xref: /linux/arch/powerpc/kernel/process.c (revision c060f8168bdf22aa986970955af99702d142dfbe)
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/hw_breakpoint.h>
38 #include <linux/uaccess.h>
39 #include <linux/pkeys.h>
40 #include <linux/seq_buf.h>
41 
42 #include <asm/interrupt.h>
43 #include <asm/io.h>
44 #include <asm/processor.h>
45 #include <asm/mmu.h>
46 #include <asm/machdep.h>
47 #include <asm/time.h>
48 #include <asm/runlatch.h>
49 #include <asm/syscalls.h>
50 #include <asm/switch_to.h>
51 #include <asm/tm.h>
52 #include <asm/debug.h>
53 #ifdef CONFIG_PPC64
54 #include <asm/firmware.h>
55 #include <asm/hw_irq.h>
56 #endif
57 #include <asm/code-patching.h>
58 #include <asm/exec.h>
59 #include <asm/livepatch.h>
60 #include <asm/cpu_has_feature.h>
61 #include <asm/asm-prototypes.h>
62 #include <asm/stacktrace.h>
63 #include <asm/hw_breakpoint.h>
64 
65 #include <linux/kprobes.h>
66 #include <linux/kdebug.h>
67 
68 /* Transactional Memory debug */
69 #ifdef TM_DEBUG_SW
70 #define TM_DEBUG(x...) printk(KERN_INFO x)
71 #else
72 #define TM_DEBUG(x...) do { } while(0)
73 #endif
74 
75 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
76 /*
77  * Are we running in "Suspend disabled" mode? If so we have to block any
78  * sigreturn that would get us into suspended state, and we also warn in some
79  * other paths that we should never reach with suspend disabled.
80  */
81 bool tm_suspend_disabled __ro_after_init = false;
82 
83 static void check_if_tm_restore_required(struct task_struct *tsk)
84 {
85 	/*
86 	 * If we are saving the current thread's registers, and the
87 	 * thread is in a transactional state, set the TIF_RESTORE_TM
88 	 * bit so that we know to restore the registers before
89 	 * returning to userspace.
90 	 */
91 	if (tsk == current && tsk->thread.regs &&
92 	    MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
93 	    !test_thread_flag(TIF_RESTORE_TM)) {
94 		regs_set_return_msr(&tsk->thread.ckpt_regs,
95 						tsk->thread.regs->msr);
96 		set_thread_flag(TIF_RESTORE_TM);
97 	}
98 }
99 
100 #else
101 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
102 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
103 
104 bool strict_msr_control;
105 EXPORT_SYMBOL(strict_msr_control);
106 
107 static int __init enable_strict_msr_control(char *str)
108 {
109 	strict_msr_control = true;
110 	pr_info("Enabling strict facility control\n");
111 
112 	return 0;
113 }
114 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
115 
116 /* notrace because it's called by restore_math */
117 unsigned long notrace msr_check_and_set(unsigned long bits)
118 {
119 	unsigned long oldmsr = mfmsr();
120 	unsigned long newmsr;
121 
122 	newmsr = oldmsr | bits;
123 
124 	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
125 		newmsr |= MSR_VSX;
126 
127 	if (oldmsr != newmsr)
128 		newmsr = mtmsr_isync_irqsafe(newmsr);
129 
130 	return newmsr;
131 }
132 EXPORT_SYMBOL_GPL(msr_check_and_set);
133 
134 /* notrace because it's called by restore_math */
135 void notrace __msr_check_and_clear(unsigned long bits)
136 {
137 	unsigned long oldmsr = mfmsr();
138 	unsigned long newmsr;
139 
140 	newmsr = oldmsr & ~bits;
141 
142 	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
143 		newmsr &= ~MSR_VSX;
144 
145 	if (oldmsr != newmsr)
146 		mtmsr_isync_irqsafe(newmsr);
147 }
148 EXPORT_SYMBOL(__msr_check_and_clear);
149 
150 #ifdef CONFIG_PPC_FPU
151 static void __giveup_fpu(struct task_struct *tsk)
152 {
153 	unsigned long msr;
154 
155 	save_fpu(tsk);
156 	msr = tsk->thread.regs->msr;
157 	msr &= ~(MSR_FP|MSR_FE0|MSR_FE1);
158 	if (cpu_has_feature(CPU_FTR_VSX))
159 		msr &= ~MSR_VSX;
160 	regs_set_return_msr(tsk->thread.regs, msr);
161 }
162 
163 void giveup_fpu(struct task_struct *tsk)
164 {
165 	check_if_tm_restore_required(tsk);
166 
167 	msr_check_and_set(MSR_FP);
168 	__giveup_fpu(tsk);
169 	msr_check_and_clear(MSR_FP);
170 }
171 EXPORT_SYMBOL(giveup_fpu);
172 
173 /*
174  * Make sure the floating-point register state in the
175  * the thread_struct is up to date for task tsk.
176  */
177 void flush_fp_to_thread(struct task_struct *tsk)
178 {
179 	if (tsk->thread.regs) {
180 		/*
181 		 * We need to disable preemption here because if we didn't,
182 		 * another process could get scheduled after the regs->msr
183 		 * test but before we have finished saving the FP registers
184 		 * to the thread_struct.  That process could take over the
185 		 * FPU, and then when we get scheduled again we would store
186 		 * bogus values for the remaining FP registers.
187 		 */
188 		preempt_disable();
189 		if (tsk->thread.regs->msr & MSR_FP) {
190 			/*
191 			 * This should only ever be called for current or
192 			 * for a stopped child process.  Since we save away
193 			 * the FP register state on context switch,
194 			 * there is something wrong if a stopped child appears
195 			 * to still have its FP state in the CPU registers.
196 			 */
197 			BUG_ON(tsk != current);
198 			giveup_fpu(tsk);
199 		}
200 		preempt_enable();
201 	}
202 }
203 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
204 
205 void enable_kernel_fp(void)
206 {
207 	unsigned long cpumsr;
208 
209 	WARN_ON(preemptible());
210 
211 	cpumsr = msr_check_and_set(MSR_FP);
212 
213 	if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
214 		check_if_tm_restore_required(current);
215 		/*
216 		 * If a thread has already been reclaimed then the
217 		 * checkpointed registers are on the CPU but have definitely
218 		 * been saved by the reclaim code. Don't need to and *cannot*
219 		 * giveup as this would save  to the 'live' structure not the
220 		 * checkpointed structure.
221 		 */
222 		if (!MSR_TM_ACTIVE(cpumsr) &&
223 		     MSR_TM_ACTIVE(current->thread.regs->msr))
224 			return;
225 		__giveup_fpu(current);
226 	}
227 }
228 EXPORT_SYMBOL(enable_kernel_fp);
229 #else
230 static inline void __giveup_fpu(struct task_struct *tsk) { }
231 #endif /* CONFIG_PPC_FPU */
232 
233 #ifdef CONFIG_ALTIVEC
234 static void __giveup_altivec(struct task_struct *tsk)
235 {
236 	unsigned long msr;
237 
238 	save_altivec(tsk);
239 	msr = tsk->thread.regs->msr;
240 	msr &= ~MSR_VEC;
241 	if (cpu_has_feature(CPU_FTR_VSX))
242 		msr &= ~MSR_VSX;
243 	regs_set_return_msr(tsk->thread.regs, msr);
244 }
245 
246 void giveup_altivec(struct task_struct *tsk)
247 {
248 	check_if_tm_restore_required(tsk);
249 
250 	msr_check_and_set(MSR_VEC);
251 	__giveup_altivec(tsk);
252 	msr_check_and_clear(MSR_VEC);
253 }
254 EXPORT_SYMBOL(giveup_altivec);
255 
256 void enable_kernel_altivec(void)
257 {
258 	unsigned long cpumsr;
259 
260 	WARN_ON(preemptible());
261 
262 	cpumsr = msr_check_and_set(MSR_VEC);
263 
264 	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
265 		check_if_tm_restore_required(current);
266 		/*
267 		 * If a thread has already been reclaimed then the
268 		 * checkpointed registers are on the CPU but have definitely
269 		 * been saved by the reclaim code. Don't need to and *cannot*
270 		 * giveup as this would save  to the 'live' structure not the
271 		 * checkpointed structure.
272 		 */
273 		if (!MSR_TM_ACTIVE(cpumsr) &&
274 		     MSR_TM_ACTIVE(current->thread.regs->msr))
275 			return;
276 		__giveup_altivec(current);
277 	}
278 }
279 EXPORT_SYMBOL(enable_kernel_altivec);
280 
281 /*
282  * Make sure the VMX/Altivec register state in the
283  * the thread_struct is up to date for task tsk.
284  */
285 void flush_altivec_to_thread(struct task_struct *tsk)
286 {
287 	if (tsk->thread.regs) {
288 		preempt_disable();
289 		if (tsk->thread.regs->msr & MSR_VEC) {
290 			BUG_ON(tsk != current);
291 			giveup_altivec(tsk);
292 		}
293 		preempt_enable();
294 	}
295 }
296 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
297 #endif /* CONFIG_ALTIVEC */
298 
299 #ifdef CONFIG_VSX
300 static void __giveup_vsx(struct task_struct *tsk)
301 {
302 	unsigned long msr = tsk->thread.regs->msr;
303 
304 	/*
305 	 * We should never be setting MSR_VSX without also setting
306 	 * MSR_FP and MSR_VEC
307 	 */
308 	WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
309 
310 	/* __giveup_fpu will clear MSR_VSX */
311 	if (msr & MSR_FP)
312 		__giveup_fpu(tsk);
313 	if (msr & MSR_VEC)
314 		__giveup_altivec(tsk);
315 }
316 
317 static void giveup_vsx(struct task_struct *tsk)
318 {
319 	check_if_tm_restore_required(tsk);
320 
321 	msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
322 	__giveup_vsx(tsk);
323 	msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
324 }
325 
326 void enable_kernel_vsx(void)
327 {
328 	unsigned long cpumsr;
329 
330 	WARN_ON(preemptible());
331 
332 	cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
333 
334 	if (current->thread.regs &&
335 	    (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
336 		check_if_tm_restore_required(current);
337 		/*
338 		 * If a thread has already been reclaimed then the
339 		 * checkpointed registers are on the CPU but have definitely
340 		 * been saved by the reclaim code. Don't need to and *cannot*
341 		 * giveup as this would save  to the 'live' structure not the
342 		 * checkpointed structure.
343 		 */
344 		if (!MSR_TM_ACTIVE(cpumsr) &&
345 		     MSR_TM_ACTIVE(current->thread.regs->msr))
346 			return;
347 		__giveup_vsx(current);
348 	}
349 }
350 EXPORT_SYMBOL(enable_kernel_vsx);
351 
352 void flush_vsx_to_thread(struct task_struct *tsk)
353 {
354 	if (tsk->thread.regs) {
355 		preempt_disable();
356 		if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
357 			BUG_ON(tsk != current);
358 			giveup_vsx(tsk);
359 		}
360 		preempt_enable();
361 	}
362 }
363 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
364 #endif /* CONFIG_VSX */
365 
366 #ifdef CONFIG_SPE
367 void giveup_spe(struct task_struct *tsk)
368 {
369 	check_if_tm_restore_required(tsk);
370 
371 	msr_check_and_set(MSR_SPE);
372 	__giveup_spe(tsk);
373 	msr_check_and_clear(MSR_SPE);
374 }
375 EXPORT_SYMBOL(giveup_spe);
376 
377 void enable_kernel_spe(void)
378 {
379 	WARN_ON(preemptible());
380 
381 	msr_check_and_set(MSR_SPE);
382 
383 	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
384 		check_if_tm_restore_required(current);
385 		__giveup_spe(current);
386 	}
387 }
388 EXPORT_SYMBOL(enable_kernel_spe);
389 
390 void flush_spe_to_thread(struct task_struct *tsk)
391 {
392 	if (tsk->thread.regs) {
393 		preempt_disable();
394 		if (tsk->thread.regs->msr & MSR_SPE) {
395 			BUG_ON(tsk != current);
396 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
397 			giveup_spe(tsk);
398 		}
399 		preempt_enable();
400 	}
401 }
402 #endif /* CONFIG_SPE */
403 
404 static unsigned long msr_all_available;
405 
406 static int __init init_msr_all_available(void)
407 {
408 	if (IS_ENABLED(CONFIG_PPC_FPU))
409 		msr_all_available |= MSR_FP;
410 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
411 		msr_all_available |= MSR_VEC;
412 	if (cpu_has_feature(CPU_FTR_VSX))
413 		msr_all_available |= MSR_VSX;
414 	if (cpu_has_feature(CPU_FTR_SPE))
415 		msr_all_available |= MSR_SPE;
416 
417 	return 0;
418 }
419 early_initcall(init_msr_all_available);
420 
421 void giveup_all(struct task_struct *tsk)
422 {
423 	unsigned long usermsr;
424 
425 	if (!tsk->thread.regs)
426 		return;
427 
428 	check_if_tm_restore_required(tsk);
429 
430 	usermsr = tsk->thread.regs->msr;
431 
432 	if ((usermsr & msr_all_available) == 0)
433 		return;
434 
435 	msr_check_and_set(msr_all_available);
436 
437 	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
438 
439 	if (usermsr & MSR_FP)
440 		__giveup_fpu(tsk);
441 	if (usermsr & MSR_VEC)
442 		__giveup_altivec(tsk);
443 	if (usermsr & MSR_SPE)
444 		__giveup_spe(tsk);
445 
446 	msr_check_and_clear(msr_all_available);
447 }
448 EXPORT_SYMBOL(giveup_all);
449 
450 #ifdef CONFIG_PPC_BOOK3S_64
451 #ifdef CONFIG_PPC_FPU
452 static bool should_restore_fp(void)
453 {
454 	if (current->thread.load_fp) {
455 		current->thread.load_fp++;
456 		return true;
457 	}
458 	return false;
459 }
460 
461 static void do_restore_fp(void)
462 {
463 	load_fp_state(&current->thread.fp_state);
464 }
465 #else
466 static bool should_restore_fp(void) { return false; }
467 static void do_restore_fp(void) { }
468 #endif /* CONFIG_PPC_FPU */
469 
470 #ifdef CONFIG_ALTIVEC
471 static bool should_restore_altivec(void)
472 {
473 	if (cpu_has_feature(CPU_FTR_ALTIVEC) && (current->thread.load_vec)) {
474 		current->thread.load_vec++;
475 		return true;
476 	}
477 	return false;
478 }
479 
480 static void do_restore_altivec(void)
481 {
482 	load_vr_state(&current->thread.vr_state);
483 	current->thread.used_vr = 1;
484 }
485 #else
486 static bool should_restore_altivec(void) { return false; }
487 static void do_restore_altivec(void) { }
488 #endif /* CONFIG_ALTIVEC */
489 
490 static bool should_restore_vsx(void)
491 {
492 	if (cpu_has_feature(CPU_FTR_VSX))
493 		return true;
494 	return false;
495 }
496 #ifdef CONFIG_VSX
497 static void do_restore_vsx(void)
498 {
499 	current->thread.used_vsr = 1;
500 }
501 #else
502 static void do_restore_vsx(void) { }
503 #endif /* CONFIG_VSX */
504 
505 /*
506  * The exception exit path calls restore_math() with interrupts hard disabled
507  * but the soft irq state not "reconciled". ftrace code that calls
508  * local_irq_save/restore causes warnings.
509  *
510  * Rather than complicate the exit path, just don't trace restore_math. This
511  * could be done by having ftrace entry code check for this un-reconciled
512  * condition where MSR[EE]=0 and PACA_IRQ_HARD_DIS is not set, and
513  * temporarily fix it up for the duration of the ftrace call.
514  */
515 void notrace restore_math(struct pt_regs *regs)
516 {
517 	unsigned long msr;
518 	unsigned long new_msr = 0;
519 
520 	msr = regs->msr;
521 
522 	/*
523 	 * new_msr tracks the facilities that are to be restored. Only reload
524 	 * if the bit is not set in the user MSR (if it is set, the registers
525 	 * are live for the user thread).
526 	 */
527 	if ((!(msr & MSR_FP)) && should_restore_fp())
528 		new_msr |= MSR_FP;
529 
530 	if ((!(msr & MSR_VEC)) && should_restore_altivec())
531 		new_msr |= MSR_VEC;
532 
533 	if ((!(msr & MSR_VSX)) && should_restore_vsx()) {
534 		if (((msr | new_msr) & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC))
535 			new_msr |= MSR_VSX;
536 	}
537 
538 	if (new_msr) {
539 		unsigned long fpexc_mode = 0;
540 
541 		msr_check_and_set(new_msr);
542 
543 		if (new_msr & MSR_FP) {
544 			do_restore_fp();
545 
546 			// This also covers VSX, because VSX implies FP
547 			fpexc_mode = current->thread.fpexc_mode;
548 		}
549 
550 		if (new_msr & MSR_VEC)
551 			do_restore_altivec();
552 
553 		if (new_msr & MSR_VSX)
554 			do_restore_vsx();
555 
556 		msr_check_and_clear(new_msr);
557 
558 		regs_set_return_msr(regs, regs->msr | new_msr | fpexc_mode);
559 	}
560 }
561 #endif /* CONFIG_PPC_BOOK3S_64 */
562 
563 static void save_all(struct task_struct *tsk)
564 {
565 	unsigned long usermsr;
566 
567 	if (!tsk->thread.regs)
568 		return;
569 
570 	usermsr = tsk->thread.regs->msr;
571 
572 	if ((usermsr & msr_all_available) == 0)
573 		return;
574 
575 	msr_check_and_set(msr_all_available);
576 
577 	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
578 
579 	if (usermsr & MSR_FP)
580 		save_fpu(tsk);
581 
582 	if (usermsr & MSR_VEC)
583 		save_altivec(tsk);
584 
585 	if (usermsr & MSR_SPE)
586 		__giveup_spe(tsk);
587 
588 	msr_check_and_clear(msr_all_available);
589 }
590 
591 void flush_all_to_thread(struct task_struct *tsk)
592 {
593 	if (tsk->thread.regs) {
594 		preempt_disable();
595 		BUG_ON(tsk != current);
596 #ifdef CONFIG_SPE
597 		if (tsk->thread.regs->msr & MSR_SPE)
598 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
599 #endif
600 		save_all(tsk);
601 
602 		preempt_enable();
603 	}
604 }
605 EXPORT_SYMBOL(flush_all_to_thread);
606 
607 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
608 void do_send_trap(struct pt_regs *regs, unsigned long address,
609 		  unsigned long error_code, int breakpt)
610 {
611 	current->thread.trap_nr = TRAP_HWBKPT;
612 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
613 			11, SIGSEGV) == NOTIFY_STOP)
614 		return;
615 
616 	/* Deliver the signal to userspace */
617 	force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
618 				    (void __user *)address);
619 }
620 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
621 
622 static void do_break_handler(struct pt_regs *regs)
623 {
624 	struct arch_hw_breakpoint null_brk = {0};
625 	struct arch_hw_breakpoint *info;
626 	ppc_inst_t instr = ppc_inst(0);
627 	int type = 0;
628 	int size = 0;
629 	unsigned long ea;
630 	int i;
631 
632 	/*
633 	 * If underneath hw supports only one watchpoint, we know it
634 	 * caused exception. 8xx also falls into this category.
635 	 */
636 	if (nr_wp_slots() == 1) {
637 		__set_breakpoint(0, &null_brk);
638 		current->thread.hw_brk[0] = null_brk;
639 		current->thread.hw_brk[0].flags |= HW_BRK_FLAG_DISABLED;
640 		return;
641 	}
642 
643 	/* Otherwise find out which DAWR caused exception and disable it. */
644 	wp_get_instr_detail(regs, &instr, &type, &size, &ea);
645 
646 	for (i = 0; i < nr_wp_slots(); i++) {
647 		info = &current->thread.hw_brk[i];
648 		if (!info->address)
649 			continue;
650 
651 		if (wp_check_constraints(regs, instr, ea, type, size, info)) {
652 			__set_breakpoint(i, &null_brk);
653 			current->thread.hw_brk[i] = null_brk;
654 			current->thread.hw_brk[i].flags |= HW_BRK_FLAG_DISABLED;
655 		}
656 	}
657 }
658 
659 DEFINE_INTERRUPT_HANDLER(do_break)
660 {
661 	current->thread.trap_nr = TRAP_HWBKPT;
662 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, regs->dsisr,
663 			11, SIGSEGV) == NOTIFY_STOP)
664 		return;
665 
666 	if (debugger_break_match(regs))
667 		return;
668 
669 	/*
670 	 * We reach here only when watchpoint exception is generated by ptrace
671 	 * event (or hw is buggy!). Now if CONFIG_HAVE_HW_BREAKPOINT is set,
672 	 * watchpoint is already handled by hw_breakpoint_handler() so we don't
673 	 * have to do anything. But when CONFIG_HAVE_HW_BREAKPOINT is not set,
674 	 * we need to manually handle the watchpoint here.
675 	 */
676 	if (!IS_ENABLED(CONFIG_HAVE_HW_BREAKPOINT))
677 		do_break_handler(regs);
678 
679 	/* Deliver the signal to userspace */
680 	force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)regs->dar);
681 }
682 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
683 
684 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk[HBP_NUM_MAX]);
685 
686 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
687 /*
688  * Set the debug registers back to their default "safe" values.
689  */
690 static void set_debug_reg_defaults(struct thread_struct *thread)
691 {
692 	thread->debug.iac1 = thread->debug.iac2 = 0;
693 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
694 	thread->debug.iac3 = thread->debug.iac4 = 0;
695 #endif
696 	thread->debug.dac1 = thread->debug.dac2 = 0;
697 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
698 	thread->debug.dvc1 = thread->debug.dvc2 = 0;
699 #endif
700 	thread->debug.dbcr0 = 0;
701 #ifdef CONFIG_BOOKE
702 	/*
703 	 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
704 	 */
705 	thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
706 			DBCR1_IAC3US | DBCR1_IAC4US;
707 	/*
708 	 * Force Data Address Compare User/Supervisor bits to be User-only
709 	 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
710 	 */
711 	thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
712 #else
713 	thread->debug.dbcr1 = 0;
714 #endif
715 }
716 
717 static void prime_debug_regs(struct debug_reg *debug)
718 {
719 	/*
720 	 * We could have inherited MSR_DE from userspace, since
721 	 * it doesn't get cleared on exception entry.  Make sure
722 	 * MSR_DE is clear before we enable any debug events.
723 	 */
724 	mtmsr(mfmsr() & ~MSR_DE);
725 
726 	mtspr(SPRN_IAC1, debug->iac1);
727 	mtspr(SPRN_IAC2, debug->iac2);
728 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
729 	mtspr(SPRN_IAC3, debug->iac3);
730 	mtspr(SPRN_IAC4, debug->iac4);
731 #endif
732 	mtspr(SPRN_DAC1, debug->dac1);
733 	mtspr(SPRN_DAC2, debug->dac2);
734 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
735 	mtspr(SPRN_DVC1, debug->dvc1);
736 	mtspr(SPRN_DVC2, debug->dvc2);
737 #endif
738 	mtspr(SPRN_DBCR0, debug->dbcr0);
739 	mtspr(SPRN_DBCR1, debug->dbcr1);
740 #ifdef CONFIG_BOOKE
741 	mtspr(SPRN_DBCR2, debug->dbcr2);
742 #endif
743 }
744 /*
745  * Unless neither the old or new thread are making use of the
746  * debug registers, set the debug registers from the values
747  * stored in the new thread.
748  */
749 void switch_booke_debug_regs(struct debug_reg *new_debug)
750 {
751 	if ((current->thread.debug.dbcr0 & DBCR0_IDM)
752 		|| (new_debug->dbcr0 & DBCR0_IDM))
753 			prime_debug_regs(new_debug);
754 }
755 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
756 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
757 #ifndef CONFIG_HAVE_HW_BREAKPOINT
758 static void set_breakpoint(int i, struct arch_hw_breakpoint *brk)
759 {
760 	preempt_disable();
761 	__set_breakpoint(i, brk);
762 	preempt_enable();
763 }
764 
765 static void set_debug_reg_defaults(struct thread_struct *thread)
766 {
767 	int i;
768 	struct arch_hw_breakpoint null_brk = {0};
769 
770 	for (i = 0; i < nr_wp_slots(); i++) {
771 		thread->hw_brk[i] = null_brk;
772 		if (ppc_breakpoint_available())
773 			set_breakpoint(i, &thread->hw_brk[i]);
774 	}
775 }
776 
777 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
778 				struct arch_hw_breakpoint *b)
779 {
780 	if (a->address != b->address)
781 		return false;
782 	if (a->type != b->type)
783 		return false;
784 	if (a->len != b->len)
785 		return false;
786 	/* no need to check hw_len. it's calculated from address and len */
787 	return true;
788 }
789 
790 static void switch_hw_breakpoint(struct task_struct *new)
791 {
792 	int i;
793 
794 	for (i = 0; i < nr_wp_slots(); i++) {
795 		if (likely(hw_brk_match(this_cpu_ptr(&current_brk[i]),
796 					&new->thread.hw_brk[i])))
797 			continue;
798 
799 		__set_breakpoint(i, &new->thread.hw_brk[i]);
800 	}
801 }
802 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
803 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
804 
805 static inline int set_dabr(struct arch_hw_breakpoint *brk)
806 {
807 	unsigned long dabr, dabrx;
808 
809 	dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
810 	dabrx = ((brk->type >> 3) & 0x7);
811 
812 	if (ppc_md.set_dabr)
813 		return ppc_md.set_dabr(dabr, dabrx);
814 
815 	if (IS_ENABLED(CONFIG_PPC_ADV_DEBUG_REGS)) {
816 		mtspr(SPRN_DAC1, dabr);
817 		if (IS_ENABLED(CONFIG_PPC_47x))
818 			isync();
819 		return 0;
820 	} else if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
821 		mtspr(SPRN_DABR, dabr);
822 		if (cpu_has_feature(CPU_FTR_DABRX))
823 			mtspr(SPRN_DABRX, dabrx);
824 		return 0;
825 	} else {
826 		return -EINVAL;
827 	}
828 }
829 
830 static inline int set_breakpoint_8xx(struct arch_hw_breakpoint *brk)
831 {
832 	unsigned long lctrl1 = LCTRL1_CTE_GT | LCTRL1_CTF_LT | LCTRL1_CRWE_RW |
833 			       LCTRL1_CRWF_RW;
834 	unsigned long lctrl2 = LCTRL2_LW0EN | LCTRL2_LW0LADC | LCTRL2_SLW0EN;
835 	unsigned long start_addr = ALIGN_DOWN(brk->address, HW_BREAKPOINT_SIZE);
836 	unsigned long end_addr = ALIGN(brk->address + brk->len, HW_BREAKPOINT_SIZE);
837 
838 	if (start_addr == 0)
839 		lctrl2 |= LCTRL2_LW0LA_F;
840 	else if (end_addr == 0)
841 		lctrl2 |= LCTRL2_LW0LA_E;
842 	else
843 		lctrl2 |= LCTRL2_LW0LA_EandF;
844 
845 	mtspr(SPRN_LCTRL2, 0);
846 
847 	if ((brk->type & HW_BRK_TYPE_RDWR) == 0)
848 		return 0;
849 
850 	if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
851 		lctrl1 |= LCTRL1_CRWE_RO | LCTRL1_CRWF_RO;
852 	if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
853 		lctrl1 |= LCTRL1_CRWE_WO | LCTRL1_CRWF_WO;
854 
855 	mtspr(SPRN_CMPE, start_addr - 1);
856 	mtspr(SPRN_CMPF, end_addr);
857 	mtspr(SPRN_LCTRL1, lctrl1);
858 	mtspr(SPRN_LCTRL2, lctrl2);
859 
860 	return 0;
861 }
862 
863 static void set_hw_breakpoint(int nr, struct arch_hw_breakpoint *brk)
864 {
865 	if (dawr_enabled())
866 		// Power8 or later
867 		set_dawr(nr, brk);
868 	else if (IS_ENABLED(CONFIG_PPC_8xx))
869 		set_breakpoint_8xx(brk);
870 	else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
871 		// Power7 or earlier
872 		set_dabr(brk);
873 	else
874 		// Shouldn't happen due to higher level checks
875 		WARN_ON_ONCE(1);
876 }
877 
878 void __set_breakpoint(int nr, struct arch_hw_breakpoint *brk)
879 {
880 	memcpy(this_cpu_ptr(&current_brk[nr]), brk, sizeof(*brk));
881 	set_hw_breakpoint(nr, brk);
882 }
883 
884 /* Check if we have DAWR or DABR hardware */
885 bool ppc_breakpoint_available(void)
886 {
887 	if (dawr_enabled())
888 		return true; /* POWER8 DAWR or POWER9 forced DAWR */
889 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
890 		return false; /* POWER9 with DAWR disabled */
891 	/* DABR: Everything but POWER8 and POWER9 */
892 	return true;
893 }
894 EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
895 
896 /* Disable the breakpoint in hardware without touching current_brk[] */
897 void suspend_breakpoints(void)
898 {
899 	struct arch_hw_breakpoint brk = {0};
900 	int i;
901 
902 	if (!ppc_breakpoint_available())
903 		return;
904 
905 	for (i = 0; i < nr_wp_slots(); i++)
906 		set_hw_breakpoint(i, &brk);
907 }
908 
909 /*
910  * Re-enable breakpoints suspended by suspend_breakpoints() in hardware
911  * from current_brk[]
912  */
913 void restore_breakpoints(void)
914 {
915 	int i;
916 
917 	if (!ppc_breakpoint_available())
918 		return;
919 
920 	for (i = 0; i < nr_wp_slots(); i++)
921 		set_hw_breakpoint(i, this_cpu_ptr(&current_brk[i]));
922 }
923 
924 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
925 
926 static inline bool tm_enabled(struct task_struct *tsk)
927 {
928 	return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
929 }
930 
931 static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
932 {
933 	/*
934 	 * Use the current MSR TM suspended bit to track if we have
935 	 * checkpointed state outstanding.
936 	 * On signal delivery, we'd normally reclaim the checkpointed
937 	 * state to obtain stack pointer (see:get_tm_stackpointer()).
938 	 * This will then directly return to userspace without going
939 	 * through __switch_to(). However, if the stack frame is bad,
940 	 * we need to exit this thread which calls __switch_to() which
941 	 * will again attempt to reclaim the already saved tm state.
942 	 * Hence we need to check that we've not already reclaimed
943 	 * this state.
944 	 * We do this using the current MSR, rather tracking it in
945 	 * some specific thread_struct bit, as it has the additional
946 	 * benefit of checking for a potential TM bad thing exception.
947 	 */
948 	if (!MSR_TM_SUSPENDED(mfmsr()))
949 		return;
950 
951 	giveup_all(container_of(thr, struct task_struct, thread));
952 
953 	tm_reclaim(thr, cause);
954 
955 	/*
956 	 * If we are in a transaction and FP is off then we can't have
957 	 * used FP inside that transaction. Hence the checkpointed
958 	 * state is the same as the live state. We need to copy the
959 	 * live state to the checkpointed state so that when the
960 	 * transaction is restored, the checkpointed state is correct
961 	 * and the aborted transaction sees the correct state. We use
962 	 * ckpt_regs.msr here as that's what tm_reclaim will use to
963 	 * determine if it's going to write the checkpointed state or
964 	 * not. So either this will write the checkpointed registers,
965 	 * or reclaim will. Similarly for VMX.
966 	 */
967 	if ((thr->ckpt_regs.msr & MSR_FP) == 0)
968 		memcpy(&thr->ckfp_state, &thr->fp_state,
969 		       sizeof(struct thread_fp_state));
970 	if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
971 		memcpy(&thr->ckvr_state, &thr->vr_state,
972 		       sizeof(struct thread_vr_state));
973 }
974 
975 void tm_reclaim_current(uint8_t cause)
976 {
977 	tm_enable();
978 	tm_reclaim_thread(&current->thread, cause);
979 }
980 
981 static inline void tm_reclaim_task(struct task_struct *tsk)
982 {
983 	/* We have to work out if we're switching from/to a task that's in the
984 	 * middle of a transaction.
985 	 *
986 	 * In switching we need to maintain a 2nd register state as
987 	 * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
988 	 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
989 	 * ckvr_state
990 	 *
991 	 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
992 	 */
993 	struct thread_struct *thr = &tsk->thread;
994 
995 	if (!thr->regs)
996 		return;
997 
998 	if (!MSR_TM_ACTIVE(thr->regs->msr))
999 		goto out_and_saveregs;
1000 
1001 	WARN_ON(tm_suspend_disabled);
1002 
1003 	TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
1004 		 "ccr=%lx, msr=%lx, trap=%lx)\n",
1005 		 tsk->pid, thr->regs->nip,
1006 		 thr->regs->ccr, thr->regs->msr,
1007 		 thr->regs->trap);
1008 
1009 	tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
1010 
1011 	TM_DEBUG("--- tm_reclaim on pid %d complete\n",
1012 		 tsk->pid);
1013 
1014 out_and_saveregs:
1015 	/* Always save the regs here, even if a transaction's not active.
1016 	 * This context-switches a thread's TM info SPRs.  We do it here to
1017 	 * be consistent with the restore path (in recheckpoint) which
1018 	 * cannot happen later in _switch().
1019 	 */
1020 	tm_save_sprs(thr);
1021 }
1022 
1023 extern void __tm_recheckpoint(struct thread_struct *thread);
1024 
1025 void tm_recheckpoint(struct thread_struct *thread)
1026 {
1027 	unsigned long flags;
1028 
1029 	if (!(thread->regs->msr & MSR_TM))
1030 		return;
1031 
1032 	/* We really can't be interrupted here as the TEXASR registers can't
1033 	 * change and later in the trecheckpoint code, we have a userspace R1.
1034 	 * So let's hard disable over this region.
1035 	 */
1036 	local_irq_save(flags);
1037 	hard_irq_disable();
1038 
1039 	/* The TM SPRs are restored here, so that TEXASR.FS can be set
1040 	 * before the trecheckpoint and no explosion occurs.
1041 	 */
1042 	tm_restore_sprs(thread);
1043 
1044 	__tm_recheckpoint(thread);
1045 
1046 	local_irq_restore(flags);
1047 }
1048 
1049 static inline void tm_recheckpoint_new_task(struct task_struct *new)
1050 {
1051 	if (!cpu_has_feature(CPU_FTR_TM))
1052 		return;
1053 
1054 	/* Recheckpoint the registers of the thread we're about to switch to.
1055 	 *
1056 	 * If the task was using FP, we non-lazily reload both the original and
1057 	 * the speculative FP register states.  This is because the kernel
1058 	 * doesn't see if/when a TM rollback occurs, so if we take an FP
1059 	 * unavailable later, we are unable to determine which set of FP regs
1060 	 * need to be restored.
1061 	 */
1062 	if (!tm_enabled(new))
1063 		return;
1064 
1065 	if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
1066 		tm_restore_sprs(&new->thread);
1067 		return;
1068 	}
1069 	/* Recheckpoint to restore original checkpointed register state. */
1070 	TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
1071 		 new->pid, new->thread.regs->msr);
1072 
1073 	tm_recheckpoint(&new->thread);
1074 
1075 	/*
1076 	 * The checkpointed state has been restored but the live state has
1077 	 * not, ensure all the math functionality is turned off to trigger
1078 	 * restore_math() to reload.
1079 	 */
1080 	new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1081 
1082 	TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1083 		 "(kernel msr 0x%lx)\n",
1084 		 new->pid, mfmsr());
1085 }
1086 
1087 static inline void __switch_to_tm(struct task_struct *prev,
1088 		struct task_struct *new)
1089 {
1090 	if (cpu_has_feature(CPU_FTR_TM)) {
1091 		if (tm_enabled(prev) || tm_enabled(new))
1092 			tm_enable();
1093 
1094 		if (tm_enabled(prev)) {
1095 			prev->thread.load_tm++;
1096 			tm_reclaim_task(prev);
1097 			if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1098 				prev->thread.regs->msr &= ~MSR_TM;
1099 		}
1100 
1101 		tm_recheckpoint_new_task(new);
1102 	}
1103 }
1104 
1105 /*
1106  * This is called if we are on the way out to userspace and the
1107  * TIF_RESTORE_TM flag is set.  It checks if we need to reload
1108  * FP and/or vector state and does so if necessary.
1109  * If userspace is inside a transaction (whether active or
1110  * suspended) and FP/VMX/VSX instructions have ever been enabled
1111  * inside that transaction, then we have to keep them enabled
1112  * and keep the FP/VMX/VSX state loaded while ever the transaction
1113  * continues.  The reason is that if we didn't, and subsequently
1114  * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1115  * we don't know whether it's the same transaction, and thus we
1116  * don't know which of the checkpointed state and the transactional
1117  * state to use.
1118  */
1119 void restore_tm_state(struct pt_regs *regs)
1120 {
1121 	unsigned long msr_diff;
1122 
1123 	/*
1124 	 * This is the only moment we should clear TIF_RESTORE_TM as
1125 	 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1126 	 * again, anything else could lead to an incorrect ckpt_msr being
1127 	 * saved and therefore incorrect signal contexts.
1128 	 */
1129 	clear_thread_flag(TIF_RESTORE_TM);
1130 	if (!MSR_TM_ACTIVE(regs->msr))
1131 		return;
1132 
1133 	msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1134 	msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1135 
1136 	/* Ensure that restore_math() will restore */
1137 	if (msr_diff & MSR_FP)
1138 		current->thread.load_fp = 1;
1139 #ifdef CONFIG_ALTIVEC
1140 	if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1141 		current->thread.load_vec = 1;
1142 #endif
1143 	restore_math(regs);
1144 
1145 	regs_set_return_msr(regs, regs->msr | msr_diff);
1146 }
1147 
1148 #else /* !CONFIG_PPC_TRANSACTIONAL_MEM */
1149 #define tm_recheckpoint_new_task(new)
1150 #define __switch_to_tm(prev, new)
1151 void tm_reclaim_current(uint8_t cause) {}
1152 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1153 
1154 static inline void save_sprs(struct thread_struct *t)
1155 {
1156 #ifdef CONFIG_ALTIVEC
1157 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
1158 		t->vrsave = mfspr(SPRN_VRSAVE);
1159 #endif
1160 #ifdef CONFIG_SPE
1161 	if (cpu_has_feature(CPU_FTR_SPE))
1162 		t->spefscr = mfspr(SPRN_SPEFSCR);
1163 #endif
1164 #ifdef CONFIG_PPC_BOOK3S_64
1165 	if (cpu_has_feature(CPU_FTR_DSCR))
1166 		t->dscr = mfspr(SPRN_DSCR);
1167 
1168 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1169 		t->bescr = mfspr(SPRN_BESCR);
1170 		t->ebbhr = mfspr(SPRN_EBBHR);
1171 		t->ebbrr = mfspr(SPRN_EBBRR);
1172 
1173 		t->fscr = mfspr(SPRN_FSCR);
1174 
1175 		/*
1176 		 * Note that the TAR is not available for use in the kernel.
1177 		 * (To provide this, the TAR should be backed up/restored on
1178 		 * exception entry/exit instead, and be in pt_regs.  FIXME,
1179 		 * this should be in pt_regs anyway (for debug).)
1180 		 */
1181 		t->tar = mfspr(SPRN_TAR);
1182 	}
1183 
1184 	if (cpu_has_feature(CPU_FTR_DEXCR_NPHIE))
1185 		t->hashkeyr = mfspr(SPRN_HASHKEYR);
1186 
1187 	if (cpu_has_feature(CPU_FTR_ARCH_31))
1188 		t->dexcr = mfspr(SPRN_DEXCR);
1189 #endif
1190 }
1191 
1192 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1193 void kvmppc_save_user_regs(void)
1194 {
1195 	unsigned long usermsr;
1196 
1197 	if (!current->thread.regs)
1198 		return;
1199 
1200 	usermsr = current->thread.regs->msr;
1201 
1202 	/* Caller has enabled FP/VEC/VSX/TM in MSR */
1203 	if (usermsr & MSR_FP)
1204 		__giveup_fpu(current);
1205 	if (usermsr & MSR_VEC)
1206 		__giveup_altivec(current);
1207 
1208 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1209 	if (usermsr & MSR_TM) {
1210 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
1211 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
1212 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
1213 		current->thread.regs->msr &= ~MSR_TM;
1214 	}
1215 #endif
1216 }
1217 EXPORT_SYMBOL_GPL(kvmppc_save_user_regs);
1218 
1219 void kvmppc_save_current_sprs(void)
1220 {
1221 	save_sprs(&current->thread);
1222 }
1223 EXPORT_SYMBOL_GPL(kvmppc_save_current_sprs);
1224 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1225 
1226 static inline void restore_sprs(struct thread_struct *old_thread,
1227 				struct thread_struct *new_thread)
1228 {
1229 #ifdef CONFIG_ALTIVEC
1230 	if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1231 	    old_thread->vrsave != new_thread->vrsave)
1232 		mtspr(SPRN_VRSAVE, new_thread->vrsave);
1233 #endif
1234 #ifdef CONFIG_SPE
1235 	if (cpu_has_feature(CPU_FTR_SPE) &&
1236 	    old_thread->spefscr != new_thread->spefscr)
1237 		mtspr(SPRN_SPEFSCR, new_thread->spefscr);
1238 #endif
1239 #ifdef CONFIG_PPC_BOOK3S_64
1240 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1241 		u64 dscr = get_paca()->dscr_default;
1242 		if (new_thread->dscr_inherit)
1243 			dscr = new_thread->dscr;
1244 
1245 		if (old_thread->dscr != dscr)
1246 			mtspr(SPRN_DSCR, dscr);
1247 	}
1248 
1249 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1250 		if (old_thread->bescr != new_thread->bescr)
1251 			mtspr(SPRN_BESCR, new_thread->bescr);
1252 		if (old_thread->ebbhr != new_thread->ebbhr)
1253 			mtspr(SPRN_EBBHR, new_thread->ebbhr);
1254 		if (old_thread->ebbrr != new_thread->ebbrr)
1255 			mtspr(SPRN_EBBRR, new_thread->ebbrr);
1256 
1257 		if (old_thread->fscr != new_thread->fscr)
1258 			mtspr(SPRN_FSCR, new_thread->fscr);
1259 
1260 		if (old_thread->tar != new_thread->tar)
1261 			mtspr(SPRN_TAR, new_thread->tar);
1262 	}
1263 
1264 	if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1265 	    old_thread->tidr != new_thread->tidr)
1266 		mtspr(SPRN_TIDR, new_thread->tidr);
1267 
1268 	if (cpu_has_feature(CPU_FTR_DEXCR_NPHIE) &&
1269 	    old_thread->hashkeyr != new_thread->hashkeyr)
1270 		mtspr(SPRN_HASHKEYR, new_thread->hashkeyr);
1271 
1272 	if (cpu_has_feature(CPU_FTR_ARCH_31) &&
1273 	    old_thread->dexcr != new_thread->dexcr)
1274 		mtspr(SPRN_DEXCR, new_thread->dexcr);
1275 #endif
1276 
1277 }
1278 
1279 struct task_struct *__switch_to(struct task_struct *prev,
1280 	struct task_struct *new)
1281 {
1282 	struct thread_struct *new_thread, *old_thread;
1283 	struct task_struct *last;
1284 #ifdef CONFIG_PPC_64S_HASH_MMU
1285 	struct ppc64_tlb_batch *batch;
1286 #endif
1287 
1288 	new_thread = &new->thread;
1289 	old_thread = &current->thread;
1290 
1291 	WARN_ON(!irqs_disabled());
1292 
1293 #ifdef CONFIG_PPC_64S_HASH_MMU
1294 	batch = this_cpu_ptr(&ppc64_tlb_batch);
1295 	if (batch->active) {
1296 		current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1297 		if (batch->index)
1298 			__flush_tlb_pending(batch);
1299 		batch->active = 0;
1300 	}
1301 
1302 	/*
1303 	 * On POWER9 the copy-paste buffer can only paste into
1304 	 * foreign real addresses, so unprivileged processes can not
1305 	 * see the data or use it in any way unless they have
1306 	 * foreign real mappings. If the new process has the foreign
1307 	 * real address mappings, we must issue a cp_abort to clear
1308 	 * any state and prevent snooping, corruption or a covert
1309 	 * channel. ISA v3.1 supports paste into local memory.
1310 	 */
1311 	if (new->mm && (cpu_has_feature(CPU_FTR_ARCH_31) ||
1312 			atomic_read(&new->mm->context.vas_windows)))
1313 		asm volatile(PPC_CP_ABORT);
1314 #endif /* CONFIG_PPC_BOOK3S_64 */
1315 
1316 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1317 	switch_booke_debug_regs(&new->thread.debug);
1318 #else
1319 /*
1320  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1321  * schedule DABR
1322  */
1323 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1324 	switch_hw_breakpoint(new);
1325 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1326 #endif
1327 
1328 	/*
1329 	 * We need to save SPRs before treclaim/trecheckpoint as these will
1330 	 * change a number of them.
1331 	 */
1332 	save_sprs(&prev->thread);
1333 
1334 	/* Save FPU, Altivec, VSX and SPE state */
1335 	giveup_all(prev);
1336 
1337 	__switch_to_tm(prev, new);
1338 
1339 	if (!radix_enabled()) {
1340 		/*
1341 		 * We can't take a PMU exception inside _switch() since there
1342 		 * is a window where the kernel stack SLB and the kernel stack
1343 		 * are out of sync. Hard disable here.
1344 		 */
1345 		hard_irq_disable();
1346 	}
1347 
1348 	/*
1349 	 * Call restore_sprs() and set_return_regs_changed() before calling
1350 	 * _switch(). If we move it after _switch() then we miss out on calling
1351 	 * it for new tasks. The reason for this is we manually create a stack
1352 	 * frame for new tasks that directly returns through ret_from_fork() or
1353 	 * ret_from_kernel_thread(). See copy_thread() for details.
1354 	 */
1355 	restore_sprs(old_thread, new_thread);
1356 
1357 	set_return_regs_changed(); /* _switch changes stack (and regs) */
1358 
1359 	if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64))
1360 		kuap_assert_locked();
1361 
1362 	last = _switch(old_thread, new_thread);
1363 
1364 	/*
1365 	 * Nothing after _switch will be run for newly created tasks,
1366 	 * because they switch directly to ret_from_fork/ret_from_kernel_thread
1367 	 * etc. Code added here should have a comment explaining why that is
1368 	 * okay.
1369 	 */
1370 
1371 #ifdef CONFIG_PPC_BOOK3S_64
1372 #ifdef CONFIG_PPC_64S_HASH_MMU
1373 	/*
1374 	 * This applies to a process that was context switched while inside
1375 	 * arch_enter_lazy_mmu_mode(), to re-activate the batch that was
1376 	 * deactivated above, before _switch(). This will never be the case
1377 	 * for new tasks.
1378 	 */
1379 	if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1380 		current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1381 		batch = this_cpu_ptr(&ppc64_tlb_batch);
1382 		batch->active = 1;
1383 	}
1384 #endif
1385 
1386 	/*
1387 	 * Math facilities are masked out of the child MSR in copy_thread.
1388 	 * A new task does not need to restore_math because it will
1389 	 * demand fault them.
1390 	 */
1391 	if (current->thread.regs)
1392 		restore_math(current->thread.regs);
1393 #endif /* CONFIG_PPC_BOOK3S_64 */
1394 
1395 	return last;
1396 }
1397 
1398 #define NR_INSN_TO_PRINT	16
1399 
1400 static void show_instructions(struct pt_regs *regs)
1401 {
1402 	int i;
1403 	unsigned long nip = regs->nip;
1404 	unsigned long pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1405 
1406 	printk("Code: ");
1407 
1408 	/*
1409 	 * If we were executing with the MMU off for instructions, adjust pc
1410 	 * rather than printing XXXXXXXX.
1411 	 */
1412 	if (!IS_ENABLED(CONFIG_BOOKE) && !(regs->msr & MSR_IR)) {
1413 		pc = (unsigned long)phys_to_virt(pc);
1414 		nip = (unsigned long)phys_to_virt(regs->nip);
1415 	}
1416 
1417 	for (i = 0; i < NR_INSN_TO_PRINT; i++) {
1418 		int instr;
1419 
1420 		if (get_kernel_nofault(instr, (const void *)pc)) {
1421 			pr_cont("XXXXXXXX ");
1422 		} else {
1423 			if (nip == pc)
1424 				pr_cont("<%08x> ", instr);
1425 			else
1426 				pr_cont("%08x ", instr);
1427 		}
1428 
1429 		pc += sizeof(int);
1430 	}
1431 
1432 	pr_cont("\n");
1433 }
1434 
1435 void show_user_instructions(struct pt_regs *regs)
1436 {
1437 	unsigned long pc;
1438 	int n = NR_INSN_TO_PRINT;
1439 	struct seq_buf s;
1440 	char buf[96]; /* enough for 8 times 9 + 2 chars */
1441 
1442 	pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1443 
1444 	seq_buf_init(&s, buf, sizeof(buf));
1445 
1446 	while (n) {
1447 		int i;
1448 
1449 		seq_buf_clear(&s);
1450 
1451 		for (i = 0; i < 8 && n; i++, n--, pc += sizeof(int)) {
1452 			int instr;
1453 
1454 			if (copy_from_user_nofault(&instr, (void __user *)pc,
1455 					sizeof(instr))) {
1456 				seq_buf_printf(&s, "XXXXXXXX ");
1457 				continue;
1458 			}
1459 			seq_buf_printf(&s, regs->nip == pc ? "<%08x> " : "%08x ", instr);
1460 		}
1461 
1462 		if (!seq_buf_has_overflowed(&s))
1463 			pr_info("%s[%d]: code: %s\n", current->comm,
1464 				current->pid, s.buffer);
1465 	}
1466 }
1467 
1468 struct regbit {
1469 	unsigned long bit;
1470 	const char *name;
1471 };
1472 
1473 static struct regbit msr_bits[] = {
1474 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1475 	{MSR_SF,	"SF"},
1476 	{MSR_HV,	"HV"},
1477 #endif
1478 	{MSR_VEC,	"VEC"},
1479 	{MSR_VSX,	"VSX"},
1480 #ifdef CONFIG_BOOKE
1481 	{MSR_CE,	"CE"},
1482 #endif
1483 	{MSR_EE,	"EE"},
1484 	{MSR_PR,	"PR"},
1485 	{MSR_FP,	"FP"},
1486 	{MSR_ME,	"ME"},
1487 #ifdef CONFIG_BOOKE
1488 	{MSR_DE,	"DE"},
1489 #else
1490 	{MSR_SE,	"SE"},
1491 	{MSR_BE,	"BE"},
1492 #endif
1493 	{MSR_IR,	"IR"},
1494 	{MSR_DR,	"DR"},
1495 	{MSR_PMM,	"PMM"},
1496 #ifndef CONFIG_BOOKE
1497 	{MSR_RI,	"RI"},
1498 	{MSR_LE,	"LE"},
1499 #endif
1500 	{0,		NULL}
1501 };
1502 
1503 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1504 {
1505 	const char *s = "";
1506 
1507 	for (; bits->bit; ++bits)
1508 		if (val & bits->bit) {
1509 			pr_cont("%s%s", s, bits->name);
1510 			s = sep;
1511 		}
1512 }
1513 
1514 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1515 static struct regbit msr_tm_bits[] = {
1516 	{MSR_TS_T,	"T"},
1517 	{MSR_TS_S,	"S"},
1518 	{MSR_TM,	"E"},
1519 	{0,		NULL}
1520 };
1521 
1522 static void print_tm_bits(unsigned long val)
1523 {
1524 /*
1525  * This only prints something if at least one of the TM bit is set.
1526  * Inside the TM[], the output means:
1527  *   E: Enabled		(bit 32)
1528  *   S: Suspended	(bit 33)
1529  *   T: Transactional	(bit 34)
1530  */
1531 	if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1532 		pr_cont(",TM[");
1533 		print_bits(val, msr_tm_bits, "");
1534 		pr_cont("]");
1535 	}
1536 }
1537 #else
1538 static void print_tm_bits(unsigned long val) {}
1539 #endif
1540 
1541 static void print_msr_bits(unsigned long val)
1542 {
1543 	pr_cont("<");
1544 	print_bits(val, msr_bits, ",");
1545 	print_tm_bits(val);
1546 	pr_cont(">");
1547 }
1548 
1549 #ifdef CONFIG_PPC64
1550 #define REG		"%016lx"
1551 #define REGS_PER_LINE	4
1552 #else
1553 #define REG		"%08lx"
1554 #define REGS_PER_LINE	8
1555 #endif
1556 
1557 static void __show_regs(struct pt_regs *regs)
1558 {
1559 	int i, trap;
1560 
1561 	printk("NIP:  "REG" LR: "REG" CTR: "REG"\n",
1562 	       regs->nip, regs->link, regs->ctr);
1563 	printk("REGS: %px TRAP: %04lx   %s  (%s)\n",
1564 	       regs, regs->trap, print_tainted(), init_utsname()->release);
1565 	printk("MSR:  "REG" ", regs->msr);
1566 	print_msr_bits(regs->msr);
1567 	pr_cont("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1568 	trap = TRAP(regs);
1569 	if (!trap_is_syscall(regs) && cpu_has_feature(CPU_FTR_CFAR))
1570 		pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1571 	if (trap == INTERRUPT_MACHINE_CHECK ||
1572 	    trap == INTERRUPT_DATA_STORAGE ||
1573 	    trap == INTERRUPT_ALIGNMENT) {
1574 		if (IS_ENABLED(CONFIG_BOOKE))
1575 			pr_cont("DEAR: "REG" ESR: "REG" ", regs->dear, regs->esr);
1576 		else
1577 			pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1578 	}
1579 
1580 #ifdef CONFIG_PPC64
1581 	pr_cont("IRQMASK: %lx ", regs->softe);
1582 #endif
1583 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1584 	if (MSR_TM_ACTIVE(regs->msr))
1585 		pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1586 #endif
1587 
1588 	for (i = 0;  i < 32;  i++) {
1589 		if ((i % REGS_PER_LINE) == 0)
1590 			pr_cont("\nGPR%02d: ", i);
1591 		pr_cont(REG " ", regs->gpr[i]);
1592 	}
1593 	pr_cont("\n");
1594 	/*
1595 	 * Lookup NIP late so we have the best change of getting the
1596 	 * above info out without failing
1597 	 */
1598 	if (IS_ENABLED(CONFIG_KALLSYMS)) {
1599 		printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1600 		printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1601 	}
1602 }
1603 
1604 void show_regs(struct pt_regs *regs)
1605 {
1606 	show_regs_print_info(KERN_DEFAULT);
1607 	__show_regs(regs);
1608 	show_stack(current, (unsigned long *) regs->gpr[1], KERN_DEFAULT);
1609 	if (!user_mode(regs))
1610 		show_instructions(regs);
1611 }
1612 
1613 void flush_thread(void)
1614 {
1615 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1616 	flush_ptrace_hw_breakpoint(current);
1617 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1618 	set_debug_reg_defaults(&current->thread);
1619 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1620 }
1621 
1622 void arch_setup_new_exec(void)
1623 {
1624 
1625 #ifdef CONFIG_PPC_BOOK3S_64
1626 	if (!radix_enabled())
1627 		hash__setup_new_exec();
1628 #endif
1629 	/*
1630 	 * If we exec out of a kernel thread then thread.regs will not be
1631 	 * set.  Do it now.
1632 	 */
1633 	if (!current->thread.regs) {
1634 		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1635 		current->thread.regs = regs - 1;
1636 	}
1637 
1638 #ifdef CONFIG_PPC_MEM_KEYS
1639 	current->thread.regs->amr  = default_amr;
1640 	current->thread.regs->iamr  = default_iamr;
1641 #endif
1642 
1643 #ifdef CONFIG_PPC_BOOK3S_64
1644 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
1645 		current->thread.dexcr = current->thread.dexcr_onexec;
1646 		mtspr(SPRN_DEXCR, current->thread.dexcr);
1647 	}
1648 #endif /* CONFIG_PPC_BOOK3S_64 */
1649 }
1650 
1651 #ifdef CONFIG_PPC64
1652 /*
1653  * Assign a TIDR (thread ID) for task @t and set it in the thread
1654  * structure. For now, we only support setting TIDR for 'current' task.
1655  *
1656  * Since the TID value is a truncated form of it PID, it is possible
1657  * (but unlikely) for 2 threads to have the same TID. In the unlikely event
1658  * that 2 threads share the same TID and are waiting, one of the following
1659  * cases will happen:
1660  *
1661  * 1. The correct thread is running, the wrong thread is not
1662  * In this situation, the correct thread is woken and proceeds to pass its
1663  * condition check.
1664  *
1665  * 2. Neither threads are running
1666  * In this situation, neither thread will be woken. When scheduled, the waiting
1667  * threads will execute either a wait, which will return immediately, followed
1668  * by a condition check, which will pass for the correct thread and fail
1669  * for the wrong thread, or they will execute the condition check immediately.
1670  *
1671  * 3. The wrong thread is running, the correct thread is not
1672  * The wrong thread will be woken, but will fail its condition check and
1673  * re-execute wait. The correct thread, when scheduled, will execute either
1674  * its condition check (which will pass), or wait, which returns immediately
1675  * when called the first time after the thread is scheduled, followed by its
1676  * condition check (which will pass).
1677  *
1678  * 4. Both threads are running
1679  * Both threads will be woken. The wrong thread will fail its condition check
1680  * and execute another wait, while the correct thread will pass its condition
1681  * check.
1682  *
1683  * @t: the task to set the thread ID for
1684  */
1685 int set_thread_tidr(struct task_struct *t)
1686 {
1687 	if (!cpu_has_feature(CPU_FTR_P9_TIDR))
1688 		return -EINVAL;
1689 
1690 	if (t != current)
1691 		return -EINVAL;
1692 
1693 	if (t->thread.tidr)
1694 		return 0;
1695 
1696 	t->thread.tidr = (u16)task_pid_nr(t);
1697 	mtspr(SPRN_TIDR, t->thread.tidr);
1698 
1699 	return 0;
1700 }
1701 EXPORT_SYMBOL_GPL(set_thread_tidr);
1702 
1703 #endif /* CONFIG_PPC64 */
1704 
1705 /*
1706  * this gets called so that we can store coprocessor state into memory and
1707  * copy the current task into the new thread.
1708  */
1709 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1710 {
1711 	flush_all_to_thread(src);
1712 	/*
1713 	 * Flush TM state out so we can copy it.  __switch_to_tm() does this
1714 	 * flush but it removes the checkpointed state from the current CPU and
1715 	 * transitions the CPU out of TM mode.  Hence we need to call
1716 	 * tm_recheckpoint_new_task() (on the same task) to restore the
1717 	 * checkpointed state back and the TM mode.
1718 	 *
1719 	 * Can't pass dst because it isn't ready. Doesn't matter, passing
1720 	 * dst is only important for __switch_to()
1721 	 */
1722 	__switch_to_tm(src, src);
1723 
1724 	*dst = *src;
1725 
1726 	clear_task_ebb(dst);
1727 
1728 	return 0;
1729 }
1730 
1731 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1732 {
1733 #ifdef CONFIG_PPC_64S_HASH_MMU
1734 	unsigned long sp_vsid;
1735 	unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1736 
1737 	if (radix_enabled())
1738 		return;
1739 
1740 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1741 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1742 			<< SLB_VSID_SHIFT_1T;
1743 	else
1744 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1745 			<< SLB_VSID_SHIFT;
1746 	sp_vsid |= SLB_VSID_KERNEL | llp;
1747 	p->thread.ksp_vsid = sp_vsid;
1748 #endif
1749 }
1750 
1751 /*
1752  * Copy a thread..
1753  */
1754 
1755 /*
1756  * Copy architecture-specific thread state
1757  */
1758 int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
1759 {
1760 	struct pt_regs *kregs; /* Switch frame regs */
1761 	extern void ret_from_fork(void);
1762 	extern void ret_from_fork_scv(void);
1763 	extern void ret_from_kernel_user_thread(void);
1764 	extern void start_kernel_thread(void);
1765 	void (*f)(void);
1766 	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1767 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1768 	int i;
1769 #endif
1770 
1771 	klp_init_thread_info(p);
1772 
1773 	if (unlikely(p->flags & PF_KTHREAD)) {
1774 		/* kernel thread */
1775 
1776 		/* Create initial minimum stack frame. */
1777 		sp -= STACK_FRAME_MIN_SIZE;
1778 		((unsigned long *)sp)[0] = 0;
1779 
1780 		f = start_kernel_thread;
1781 		p->thread.regs = NULL;	/* no user register state */
1782 		clear_tsk_compat_task(p);
1783 	} else {
1784 		/* user thread */
1785 		struct pt_regs *childregs;
1786 
1787 		/* Create initial user return stack frame. */
1788 		sp -= STACK_USER_INT_FRAME_SIZE;
1789 		*(unsigned long *)(sp + STACK_INT_FRAME_MARKER) = STACK_FRAME_REGS_MARKER;
1790 
1791 		childregs = (struct pt_regs *)(sp + STACK_INT_FRAME_REGS);
1792 
1793 		if (unlikely(args->fn)) {
1794 			/*
1795 			 * A user space thread, but it first runs a kernel
1796 			 * thread, and then returns as though it had called
1797 			 * execve rather than fork, so user regs will be
1798 			 * filled in (e.g., by kernel_execve()).
1799 			 */
1800 			((unsigned long *)sp)[0] = 0;
1801 			memset(childregs, 0, sizeof(struct pt_regs));
1802 #ifdef CONFIG_PPC64
1803 			childregs->softe = IRQS_ENABLED;
1804 #endif
1805 			f = ret_from_kernel_user_thread;
1806 		} else {
1807 			struct pt_regs *regs = current_pt_regs();
1808 			unsigned long clone_flags = args->flags;
1809 			unsigned long usp = args->stack;
1810 
1811 			/* Copy registers */
1812 			*childregs = *regs;
1813 			if (usp)
1814 				childregs->gpr[1] = usp;
1815 			((unsigned long *)sp)[0] = childregs->gpr[1];
1816 #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
1817 			WARN_ON_ONCE(childregs->softe != IRQS_ENABLED);
1818 #endif
1819 			if (clone_flags & CLONE_SETTLS) {
1820 				unsigned long tls = args->tls;
1821 
1822 				if (!is_32bit_task())
1823 					childregs->gpr[13] = tls;
1824 				else
1825 					childregs->gpr[2] = tls;
1826 			}
1827 
1828 			if (trap_is_scv(regs))
1829 				f = ret_from_fork_scv;
1830 			else
1831 				f = ret_from_fork;
1832 		}
1833 
1834 		childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1835 		p->thread.regs = childregs;
1836 	}
1837 
1838 	/*
1839 	 * The way this works is that at some point in the future
1840 	 * some task will call _switch to switch to the new task.
1841 	 * That will pop off the stack frame created below and start
1842 	 * the new task running at ret_from_fork.  The new task will
1843 	 * do some house keeping and then return from the fork or clone
1844 	 * system call, using the stack frame created above.
1845 	 */
1846 	((unsigned long *)sp)[STACK_FRAME_LR_SAVE] = (unsigned long)f;
1847 	sp -= STACK_SWITCH_FRAME_SIZE;
1848 	((unsigned long *)sp)[0] = sp + STACK_SWITCH_FRAME_SIZE;
1849 	kregs = (struct pt_regs *)(sp + STACK_SWITCH_FRAME_REGS);
1850 	kregs->nip = ppc_function_entry(f);
1851 	if (unlikely(args->fn)) {
1852 		/*
1853 		 * Put kthread fn, arg parameters in non-volatile GPRs in the
1854 		 * switch frame so they are loaded by _switch before it returns
1855 		 * to ret_from_kernel_thread.
1856 		 */
1857 		kregs->gpr[14] = ppc_function_entry((void *)args->fn);
1858 		kregs->gpr[15] = (unsigned long)args->fn_arg;
1859 	}
1860 	p->thread.ksp = sp;
1861 
1862 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1863 	for (i = 0; i < nr_wp_slots(); i++)
1864 		p->thread.ptrace_bps[i] = NULL;
1865 #endif
1866 
1867 #ifdef CONFIG_PPC_FPU_REGS
1868 	p->thread.fp_save_area = NULL;
1869 #endif
1870 #ifdef CONFIG_ALTIVEC
1871 	p->thread.vr_save_area = NULL;
1872 #endif
1873 #if defined(CONFIG_PPC_BOOK3S_32) && defined(CONFIG_PPC_KUAP)
1874 	p->thread.kuap = KUAP_NONE;
1875 #endif
1876 #if defined(CONFIG_BOOKE) && defined(CONFIG_PPC_KUAP)
1877 	p->thread.pid = MMU_NO_CONTEXT;
1878 #endif
1879 
1880 	setup_ksp_vsid(p, sp);
1881 
1882 #ifdef CONFIG_PPC64
1883 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1884 		p->thread.dscr_inherit = current->thread.dscr_inherit;
1885 		p->thread.dscr = mfspr(SPRN_DSCR);
1886 	}
1887 
1888 	p->thread.tidr = 0;
1889 #endif
1890 #ifdef CONFIG_PPC_BOOK3S_64
1891 	if (cpu_has_feature(CPU_FTR_DEXCR_NPHIE))
1892 		p->thread.hashkeyr = current->thread.hashkeyr;
1893 
1894 	if (cpu_has_feature(CPU_FTR_ARCH_31))
1895 		p->thread.dexcr = mfspr(SPRN_DEXCR);
1896 #endif
1897 	return 0;
1898 }
1899 
1900 void preload_new_slb_context(unsigned long start, unsigned long sp);
1901 
1902 /*
1903  * Set up a thread for executing a new program
1904  */
1905 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1906 {
1907 #ifdef CONFIG_PPC64
1908 	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
1909 
1910 	if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && !radix_enabled())
1911 		preload_new_slb_context(start, sp);
1912 #endif
1913 
1914 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1915 	/*
1916 	 * Clear any transactional state, we're exec()ing. The cause is
1917 	 * not important as there will never be a recheckpoint so it's not
1918 	 * user visible.
1919 	 */
1920 	if (MSR_TM_SUSPENDED(mfmsr()))
1921 		tm_reclaim_current(0);
1922 #endif
1923 
1924 	memset(&regs->gpr[1], 0, sizeof(regs->gpr) - sizeof(regs->gpr[0]));
1925 	regs->ctr = 0;
1926 	regs->link = 0;
1927 	regs->xer = 0;
1928 	regs->ccr = 0;
1929 	regs->gpr[1] = sp;
1930 
1931 #ifdef CONFIG_PPC32
1932 	regs->mq = 0;
1933 	regs->nip = start;
1934 	regs->msr = MSR_USER;
1935 #else
1936 	if (!is_32bit_task()) {
1937 		unsigned long entry;
1938 
1939 		if (is_elf2_task()) {
1940 			/* Look ma, no function descriptors! */
1941 			entry = start;
1942 
1943 			/*
1944 			 * Ulrich says:
1945 			 *   The latest iteration of the ABI requires that when
1946 			 *   calling a function (at its global entry point),
1947 			 *   the caller must ensure r12 holds the entry point
1948 			 *   address (so that the function can quickly
1949 			 *   establish addressability).
1950 			 */
1951 			regs->gpr[12] = start;
1952 			/* Make sure that's restored on entry to userspace. */
1953 			set_thread_flag(TIF_RESTOREALL);
1954 		} else {
1955 			unsigned long toc;
1956 
1957 			/* start is a relocated pointer to the function
1958 			 * descriptor for the elf _start routine.  The first
1959 			 * entry in the function descriptor is the entry
1960 			 * address of _start and the second entry is the TOC
1961 			 * value we need to use.
1962 			 */
1963 			__get_user(entry, (unsigned long __user *)start);
1964 			__get_user(toc, (unsigned long __user *)start+1);
1965 
1966 			/* Check whether the e_entry function descriptor entries
1967 			 * need to be relocated before we can use them.
1968 			 */
1969 			if (load_addr != 0) {
1970 				entry += load_addr;
1971 				toc   += load_addr;
1972 			}
1973 			regs->gpr[2] = toc;
1974 		}
1975 		regs_set_return_ip(regs, entry);
1976 		regs_set_return_msr(regs, MSR_USER64);
1977 	} else {
1978 		regs->gpr[2] = 0;
1979 		regs_set_return_ip(regs, start);
1980 		regs_set_return_msr(regs, MSR_USER32);
1981 	}
1982 
1983 #endif
1984 #ifdef CONFIG_VSX
1985 	current->thread.used_vsr = 0;
1986 #endif
1987 	current->thread.load_slb = 0;
1988 	current->thread.load_fp = 0;
1989 #ifdef CONFIG_PPC_FPU_REGS
1990 	memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1991 	current->thread.fp_save_area = NULL;
1992 #endif
1993 #ifdef CONFIG_ALTIVEC
1994 	memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1995 	current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1996 	current->thread.vr_save_area = NULL;
1997 	current->thread.vrsave = 0;
1998 	current->thread.used_vr = 0;
1999 	current->thread.load_vec = 0;
2000 #endif /* CONFIG_ALTIVEC */
2001 #ifdef CONFIG_SPE
2002 	memset(current->thread.evr, 0, sizeof(current->thread.evr));
2003 	current->thread.acc = 0;
2004 	current->thread.spefscr = 0;
2005 	current->thread.used_spe = 0;
2006 #endif /* CONFIG_SPE */
2007 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2008 	current->thread.tm_tfhar = 0;
2009 	current->thread.tm_texasr = 0;
2010 	current->thread.tm_tfiar = 0;
2011 	current->thread.load_tm = 0;
2012 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
2013 #ifdef CONFIG_PPC_BOOK3S_64
2014 	if (cpu_has_feature(CPU_FTR_DEXCR_NPHIE)) {
2015 		current->thread.hashkeyr = get_random_long();
2016 		mtspr(SPRN_HASHKEYR, current->thread.hashkeyr);
2017 	}
2018 #endif /* CONFIG_PPC_BOOK3S_64 */
2019 }
2020 EXPORT_SYMBOL(start_thread);
2021 
2022 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
2023 		| PR_FP_EXC_RES | PR_FP_EXC_INV)
2024 
2025 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
2026 {
2027 	struct pt_regs *regs = tsk->thread.regs;
2028 
2029 	/* This is a bit hairy.  If we are an SPE enabled  processor
2030 	 * (have embedded fp) we store the IEEE exception enable flags in
2031 	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
2032 	 * mode (asyn, precise, disabled) for 'Classic' FP. */
2033 	if (val & PR_FP_EXC_SW_ENABLE) {
2034 		if (cpu_has_feature(CPU_FTR_SPE)) {
2035 			/*
2036 			 * When the sticky exception bits are set
2037 			 * directly by userspace, it must call prctl
2038 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
2039 			 * in the existing prctl settings) or
2040 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
2041 			 * the bits being set).  <fenv.h> functions
2042 			 * saving and restoring the whole
2043 			 * floating-point environment need to do so
2044 			 * anyway to restore the prctl settings from
2045 			 * the saved environment.
2046 			 */
2047 #ifdef CONFIG_SPE
2048 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
2049 			tsk->thread.fpexc_mode = val &
2050 				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
2051 #endif
2052 			return 0;
2053 		} else {
2054 			return -EINVAL;
2055 		}
2056 	}
2057 
2058 	/* on a CONFIG_SPE this does not hurt us.  The bits that
2059 	 * __pack_fe01 use do not overlap with bits used for
2060 	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
2061 	 * on CONFIG_SPE implementations are reserved so writing to
2062 	 * them does not change anything */
2063 	if (val > PR_FP_EXC_PRECISE)
2064 		return -EINVAL;
2065 	tsk->thread.fpexc_mode = __pack_fe01(val);
2066 	if (regs != NULL && (regs->msr & MSR_FP) != 0) {
2067 		regs_set_return_msr(regs, (regs->msr & ~(MSR_FE0|MSR_FE1))
2068 						| tsk->thread.fpexc_mode);
2069 	}
2070 	return 0;
2071 }
2072 
2073 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
2074 {
2075 	unsigned int val = 0;
2076 
2077 	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE) {
2078 		if (cpu_has_feature(CPU_FTR_SPE)) {
2079 			/*
2080 			 * When the sticky exception bits are set
2081 			 * directly by userspace, it must call prctl
2082 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
2083 			 * in the existing prctl settings) or
2084 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
2085 			 * the bits being set).  <fenv.h> functions
2086 			 * saving and restoring the whole
2087 			 * floating-point environment need to do so
2088 			 * anyway to restore the prctl settings from
2089 			 * the saved environment.
2090 			 */
2091 #ifdef CONFIG_SPE
2092 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
2093 			val = tsk->thread.fpexc_mode;
2094 #endif
2095 		} else
2096 			return -EINVAL;
2097 	} else {
2098 		val = __unpack_fe01(tsk->thread.fpexc_mode);
2099 	}
2100 	return put_user(val, (unsigned int __user *) adr);
2101 }
2102 
2103 int set_endian(struct task_struct *tsk, unsigned int val)
2104 {
2105 	struct pt_regs *regs = tsk->thread.regs;
2106 
2107 	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
2108 	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
2109 		return -EINVAL;
2110 
2111 	if (regs == NULL)
2112 		return -EINVAL;
2113 
2114 	if (val == PR_ENDIAN_BIG)
2115 		regs_set_return_msr(regs, regs->msr & ~MSR_LE);
2116 	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
2117 		regs_set_return_msr(regs, regs->msr | MSR_LE);
2118 	else
2119 		return -EINVAL;
2120 
2121 	return 0;
2122 }
2123 
2124 int get_endian(struct task_struct *tsk, unsigned long adr)
2125 {
2126 	struct pt_regs *regs = tsk->thread.regs;
2127 	unsigned int val;
2128 
2129 	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
2130 	    !cpu_has_feature(CPU_FTR_REAL_LE))
2131 		return -EINVAL;
2132 
2133 	if (regs == NULL)
2134 		return -EINVAL;
2135 
2136 	if (regs->msr & MSR_LE) {
2137 		if (cpu_has_feature(CPU_FTR_REAL_LE))
2138 			val = PR_ENDIAN_LITTLE;
2139 		else
2140 			val = PR_ENDIAN_PPC_LITTLE;
2141 	} else
2142 		val = PR_ENDIAN_BIG;
2143 
2144 	return put_user(val, (unsigned int __user *)adr);
2145 }
2146 
2147 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
2148 {
2149 	tsk->thread.align_ctl = val;
2150 	return 0;
2151 }
2152 
2153 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
2154 {
2155 	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
2156 }
2157 
2158 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
2159 				  unsigned long nbytes)
2160 {
2161 	unsigned long stack_page;
2162 	unsigned long cpu = task_cpu(p);
2163 
2164 	if (!hardirq_ctx[cpu] || !softirq_ctx[cpu])
2165 		return 0;
2166 
2167 	stack_page = (unsigned long)hardirq_ctx[cpu];
2168 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2169 		return 1;
2170 
2171 	stack_page = (unsigned long)softirq_ctx[cpu];
2172 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2173 		return 1;
2174 
2175 	return 0;
2176 }
2177 
2178 #ifdef CONFIG_PPC64
2179 static inline int valid_emergency_stack(unsigned long sp, struct task_struct *p,
2180 					unsigned long nbytes)
2181 {
2182 	unsigned long stack_page;
2183 	unsigned long cpu = task_cpu(p);
2184 
2185 	if (!paca_ptrs)
2186 		return 0;
2187 
2188 	if (!paca_ptrs[cpu]->emergency_sp)
2189 		return 0;
2190 
2191 # ifdef CONFIG_PPC_BOOK3S_64
2192 	if (!paca_ptrs[cpu]->nmi_emergency_sp || !paca_ptrs[cpu]->mc_emergency_sp)
2193 		return 0;
2194 #endif
2195 
2196 	stack_page = (unsigned long)paca_ptrs[cpu]->emergency_sp - THREAD_SIZE;
2197 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2198 		return 1;
2199 
2200 # ifdef CONFIG_PPC_BOOK3S_64
2201 	stack_page = (unsigned long)paca_ptrs[cpu]->nmi_emergency_sp - THREAD_SIZE;
2202 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2203 		return 1;
2204 
2205 	stack_page = (unsigned long)paca_ptrs[cpu]->mc_emergency_sp - THREAD_SIZE;
2206 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2207 		return 1;
2208 # endif
2209 
2210 	return 0;
2211 }
2212 #else
2213 static inline int valid_emergency_stack(unsigned long sp, struct task_struct *p,
2214 					unsigned long nbytes)
2215 {
2216 	unsigned long stack_page;
2217 	unsigned long cpu = task_cpu(p);
2218 
2219 	if (!IS_ENABLED(CONFIG_VMAP_STACK))
2220 		return 0;
2221 
2222 	stack_page = (unsigned long)emergency_ctx[cpu] - THREAD_SIZE;
2223 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2224 		return 1;
2225 
2226 	return 0;
2227 }
2228 #endif
2229 
2230 /*
2231  * validate the stack frame of a particular minimum size, used for when we are
2232  * looking at a certain object in the stack beyond the minimum.
2233  */
2234 int validate_sp_size(unsigned long sp, struct task_struct *p,
2235 		     unsigned long nbytes)
2236 {
2237 	unsigned long stack_page = (unsigned long)task_stack_page(p);
2238 
2239 	if (sp < THREAD_SIZE)
2240 		return 0;
2241 
2242 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2243 		return 1;
2244 
2245 	if (valid_irq_stack(sp, p, nbytes))
2246 		return 1;
2247 
2248 	return valid_emergency_stack(sp, p, nbytes);
2249 }
2250 
2251 int validate_sp(unsigned long sp, struct task_struct *p)
2252 {
2253 	return validate_sp_size(sp, p, STACK_FRAME_MIN_SIZE);
2254 }
2255 
2256 static unsigned long ___get_wchan(struct task_struct *p)
2257 {
2258 	unsigned long ip, sp;
2259 	int count = 0;
2260 
2261 	sp = p->thread.ksp;
2262 	if (!validate_sp(sp, p))
2263 		return 0;
2264 
2265 	do {
2266 		sp = READ_ONCE_NOCHECK(*(unsigned long *)sp);
2267 		if (!validate_sp(sp, p) || task_is_running(p))
2268 			return 0;
2269 		if (count > 0) {
2270 			ip = READ_ONCE_NOCHECK(((unsigned long *)sp)[STACK_FRAME_LR_SAVE]);
2271 			if (!in_sched_functions(ip))
2272 				return ip;
2273 		}
2274 	} while (count++ < 16);
2275 	return 0;
2276 }
2277 
2278 unsigned long __get_wchan(struct task_struct *p)
2279 {
2280 	unsigned long ret;
2281 
2282 	if (!try_get_task_stack(p))
2283 		return 0;
2284 
2285 	ret = ___get_wchan(p);
2286 
2287 	put_task_stack(p);
2288 
2289 	return ret;
2290 }
2291 
2292 static bool empty_user_regs(struct pt_regs *regs, struct task_struct *tsk)
2293 {
2294 	unsigned long stack_page;
2295 
2296 	// A non-empty pt_regs should never have a zero MSR or TRAP value.
2297 	if (regs->msr || regs->trap)
2298 		return false;
2299 
2300 	// Check it sits at the very base of the stack
2301 	stack_page = (unsigned long)task_stack_page(tsk);
2302 	if ((unsigned long)(regs + 1) != stack_page + THREAD_SIZE)
2303 		return false;
2304 
2305 	return true;
2306 }
2307 
2308 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2309 
2310 void __no_sanitize_address show_stack(struct task_struct *tsk,
2311 				      unsigned long *stack,
2312 				      const char *loglvl)
2313 {
2314 	unsigned long sp, ip, lr, newsp;
2315 	int count = 0;
2316 	int firstframe = 1;
2317 	unsigned long ret_addr;
2318 	int ftrace_idx = 0;
2319 
2320 	if (tsk == NULL)
2321 		tsk = current;
2322 
2323 	if (!try_get_task_stack(tsk))
2324 		return;
2325 
2326 	sp = (unsigned long) stack;
2327 	if (sp == 0) {
2328 		if (tsk == current)
2329 			sp = current_stack_frame();
2330 		else
2331 			sp = tsk->thread.ksp;
2332 	}
2333 
2334 	lr = 0;
2335 	printk("%sCall Trace:\n", loglvl);
2336 	do {
2337 		if (!validate_sp(sp, tsk))
2338 			break;
2339 
2340 		stack = (unsigned long *) sp;
2341 		newsp = stack[0];
2342 		ip = stack[STACK_FRAME_LR_SAVE];
2343 		if (!firstframe || ip != lr) {
2344 			printk("%s["REG"] ["REG"] %pS",
2345 				loglvl, sp, ip, (void *)ip);
2346 			ret_addr = ftrace_graph_ret_addr(current,
2347 						&ftrace_idx, ip, stack);
2348 			if (ret_addr != ip)
2349 				pr_cont(" (%pS)", (void *)ret_addr);
2350 			if (firstframe)
2351 				pr_cont(" (unreliable)");
2352 			pr_cont("\n");
2353 		}
2354 		firstframe = 0;
2355 
2356 		/*
2357 		 * See if this is an exception frame.
2358 		 * We look for the "regs" marker in the current frame.
2359 		 *
2360 		 * STACK_SWITCH_FRAME_SIZE being the smallest frame that
2361 		 * could hold a pt_regs, if that does not fit then it can't
2362 		 * have regs.
2363 		 */
2364 		if (validate_sp_size(sp, tsk, STACK_SWITCH_FRAME_SIZE)
2365 		    && stack[STACK_INT_FRAME_MARKER_LONGS] == STACK_FRAME_REGS_MARKER) {
2366 			struct pt_regs *regs = (struct pt_regs *)
2367 				(sp + STACK_INT_FRAME_REGS);
2368 
2369 			lr = regs->link;
2370 			printk("%s--- interrupt: %lx at %pS\n",
2371 			       loglvl, regs->trap, (void *)regs->nip);
2372 
2373 			// Detect the case of an empty pt_regs at the very base
2374 			// of the stack and suppress showing it in full.
2375 			if (!empty_user_regs(regs, tsk)) {
2376 				__show_regs(regs);
2377 				printk("%s--- interrupt: %lx\n", loglvl, regs->trap);
2378 			}
2379 
2380 			firstframe = 1;
2381 		}
2382 
2383 		sp = newsp;
2384 	} while (count++ < kstack_depth_to_print);
2385 
2386 	put_task_stack(tsk);
2387 }
2388 
2389 #ifdef CONFIG_PPC64
2390 /* Called with hard IRQs off */
2391 void notrace __ppc64_runlatch_on(void)
2392 {
2393 	struct thread_info *ti = current_thread_info();
2394 
2395 	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2396 		/*
2397 		 * Least significant bit (RUN) is the only writable bit of
2398 		 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2399 		 * earliest ISA where this is the case, but it's convenient.
2400 		 */
2401 		mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2402 	} else {
2403 		unsigned long ctrl;
2404 
2405 		/*
2406 		 * Some architectures (e.g., Cell) have writable fields other
2407 		 * than RUN, so do the read-modify-write.
2408 		 */
2409 		ctrl = mfspr(SPRN_CTRLF);
2410 		ctrl |= CTRL_RUNLATCH;
2411 		mtspr(SPRN_CTRLT, ctrl);
2412 	}
2413 
2414 	ti->local_flags |= _TLF_RUNLATCH;
2415 }
2416 
2417 /* Called with hard IRQs off */
2418 void notrace __ppc64_runlatch_off(void)
2419 {
2420 	struct thread_info *ti = current_thread_info();
2421 
2422 	ti->local_flags &= ~_TLF_RUNLATCH;
2423 
2424 	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2425 		mtspr(SPRN_CTRLT, 0);
2426 	} else {
2427 		unsigned long ctrl;
2428 
2429 		ctrl = mfspr(SPRN_CTRLF);
2430 		ctrl &= ~CTRL_RUNLATCH;
2431 		mtspr(SPRN_CTRLT, ctrl);
2432 	}
2433 }
2434 #endif /* CONFIG_PPC64 */
2435 
2436 unsigned long arch_align_stack(unsigned long sp)
2437 {
2438 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2439 		sp -= get_random_u32_below(PAGE_SIZE);
2440 	return sp & ~0xf;
2441 }
2442