]> Git Repo - linux.git/blob - arch/powerpc/kernel/process.c
Revert "kbuild: avoid static_assert for genksyms"
[linux.git] / arch / powerpc / kernel / process.c
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 ([email protected]) and
7  *  Paul Mackerras ([email protected])
8  *
9  *  PowerPC version
10  *    Copyright (C) 1995-1996 Gary Thomas ([email protected])
11  */
12
13 #include <linux/errno.h>
14 #include <linux/sched.h>
15 #include <linux/sched/debug.h>
16 #include <linux/sched/task.h>
17 #include <linux/sched/task_stack.h>
18 #include <linux/kernel.h>
19 #include <linux/mm.h>
20 #include <linux/smp.h>
21 #include <linux/stddef.h>
22 #include <linux/unistd.h>
23 #include <linux/ptrace.h>
24 #include <linux/slab.h>
25 #include <linux/user.h>
26 #include <linux/elf.h>
27 #include <linux/prctl.h>
28 #include <linux/init_task.h>
29 #include <linux/export.h>
30 #include <linux/kallsyms.h>
31 #include <linux/mqueue.h>
32 #include <linux/hardirq.h>
33 #include <linux/utsname.h>
34 #include <linux/ftrace.h>
35 #include <linux/kernel_stat.h>
36 #include <linux/personality.h>
37 #include <linux/random.h>
38 #include <linux/hw_breakpoint.h>
39 #include <linux/uaccess.h>
40 #include <linux/elf-randomize.h>
41 #include <linux/pkeys.h>
42 #include <linux/seq_buf.h>
43
44 #include <asm/io.h>
45 #include <asm/processor.h>
46 #include <asm/mmu.h>
47 #include <asm/prom.h>
48 #include <asm/machdep.h>
49 #include <asm/time.h>
50 #include <asm/runlatch.h>
51 #include <asm/syscalls.h>
52 #include <asm/switch_to.h>
53 #include <asm/tm.h>
54 #include <asm/debug.h>
55 #ifdef CONFIG_PPC64
56 #include <asm/firmware.h>
57 #include <asm/hw_irq.h>
58 #endif
59 #include <asm/code-patching.h>
60 #include <asm/exec.h>
61 #include <asm/livepatch.h>
62 #include <asm/cpu_has_feature.h>
63 #include <asm/asm-prototypes.h>
64 #include <asm/stacktrace.h>
65 #include <asm/hw_breakpoint.h>
66
67 #include <linux/kprobes.h>
68 #include <linux/kdebug.h>
69
70 /* Transactional Memory debug */
71 #ifdef TM_DEBUG_SW
72 #define TM_DEBUG(x...) printk(KERN_INFO x)
73 #else
74 #define TM_DEBUG(x...) do { } while(0)
75 #endif
76
77 extern unsigned long _get_SP(void);
78
79 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
80 /*
81  * Are we running in "Suspend disabled" mode? If so we have to block any
82  * sigreturn that would get us into suspended state, and we also warn in some
83  * other paths that we should never reach with suspend disabled.
84  */
85 bool tm_suspend_disabled __ro_after_init = false;
86
87 static void check_if_tm_restore_required(struct task_struct *tsk)
88 {
89         /*
90          * If we are saving the current thread's registers, and the
91          * thread is in a transactional state, set the TIF_RESTORE_TM
92          * bit so that we know to restore the registers before
93          * returning to userspace.
94          */
95         if (tsk == current && tsk->thread.regs &&
96             MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
97             !test_thread_flag(TIF_RESTORE_TM)) {
98                 tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
99                 set_thread_flag(TIF_RESTORE_TM);
100         }
101 }
102
103 #else
104 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
105 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
106
107 bool strict_msr_control;
108 EXPORT_SYMBOL(strict_msr_control);
109
110 static int __init enable_strict_msr_control(char *str)
111 {
112         strict_msr_control = true;
113         pr_info("Enabling strict facility control\n");
114
115         return 0;
116 }
117 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
118
119 /* notrace because it's called by restore_math */
120 unsigned long notrace msr_check_and_set(unsigned long bits)
121 {
122         unsigned long oldmsr = mfmsr();
123         unsigned long newmsr;
124
125         newmsr = oldmsr | bits;
126
127         if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
128                 newmsr |= MSR_VSX;
129
130         if (oldmsr != newmsr)
131                 mtmsr_isync(newmsr);
132
133         return newmsr;
134 }
135 EXPORT_SYMBOL_GPL(msr_check_and_set);
136
137 /* notrace because it's called by restore_math */
138 void notrace __msr_check_and_clear(unsigned long bits)
139 {
140         unsigned long oldmsr = mfmsr();
141         unsigned long newmsr;
142
143         newmsr = oldmsr & ~bits;
144
145         if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
146                 newmsr &= ~MSR_VSX;
147
148         if (oldmsr != newmsr)
149                 mtmsr_isync(newmsr);
150 }
151 EXPORT_SYMBOL(__msr_check_and_clear);
152
153 #ifdef CONFIG_PPC_FPU
154 static void __giveup_fpu(struct task_struct *tsk)
155 {
156         unsigned long msr;
157
158         save_fpu(tsk);
159         msr = tsk->thread.regs->msr;
160         msr &= ~(MSR_FP|MSR_FE0|MSR_FE1);
161         if (cpu_has_feature(CPU_FTR_VSX))
162                 msr &= ~MSR_VSX;
163         tsk->thread.regs->msr = msr;
164 }
165
166 void giveup_fpu(struct task_struct *tsk)
167 {
168         check_if_tm_restore_required(tsk);
169
170         msr_check_and_set(MSR_FP);
171         __giveup_fpu(tsk);
172         msr_check_and_clear(MSR_FP);
173 }
174 EXPORT_SYMBOL(giveup_fpu);
175
176 /*
177  * Make sure the floating-point register state in the
178  * the thread_struct is up to date for task tsk.
179  */
180 void flush_fp_to_thread(struct task_struct *tsk)
181 {
182         if (tsk->thread.regs) {
183                 /*
184                  * We need to disable preemption here because if we didn't,
185                  * another process could get scheduled after the regs->msr
186                  * test but before we have finished saving the FP registers
187                  * to the thread_struct.  That process could take over the
188                  * FPU, and then when we get scheduled again we would store
189                  * bogus values for the remaining FP registers.
190                  */
191                 preempt_disable();
192                 if (tsk->thread.regs->msr & MSR_FP) {
193                         /*
194                          * This should only ever be called for current or
195                          * for a stopped child process.  Since we save away
196                          * the FP register state on context switch,
197                          * there is something wrong if a stopped child appears
198                          * to still have its FP state in the CPU registers.
199                          */
200                         BUG_ON(tsk != current);
201                         giveup_fpu(tsk);
202                 }
203                 preempt_enable();
204         }
205 }
206 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
207
208 void enable_kernel_fp(void)
209 {
210         unsigned long cpumsr;
211
212         WARN_ON(preemptible());
213
214         cpumsr = msr_check_and_set(MSR_FP);
215
216         if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
217                 check_if_tm_restore_required(current);
218                 /*
219                  * If a thread has already been reclaimed then the
220                  * checkpointed registers are on the CPU but have definitely
221                  * been saved by the reclaim code. Don't need to and *cannot*
222                  * giveup as this would save  to the 'live' structure not the
223                  * checkpointed structure.
224                  */
225                 if (!MSR_TM_ACTIVE(cpumsr) &&
226                      MSR_TM_ACTIVE(current->thread.regs->msr))
227                         return;
228                 __giveup_fpu(current);
229         }
230 }
231 EXPORT_SYMBOL(enable_kernel_fp);
232 #else
233 static inline void __giveup_fpu(struct task_struct *tsk) { }
234 #endif /* CONFIG_PPC_FPU */
235
236 #ifdef CONFIG_ALTIVEC
237 static void __giveup_altivec(struct task_struct *tsk)
238 {
239         unsigned long msr;
240
241         save_altivec(tsk);
242         msr = tsk->thread.regs->msr;
243         msr &= ~MSR_VEC;
244         if (cpu_has_feature(CPU_FTR_VSX))
245                 msr &= ~MSR_VSX;
246         tsk->thread.regs->msr = msr;
247 }
248
249 void giveup_altivec(struct task_struct *tsk)
250 {
251         check_if_tm_restore_required(tsk);
252
253         msr_check_and_set(MSR_VEC);
254         __giveup_altivec(tsk);
255         msr_check_and_clear(MSR_VEC);
256 }
257 EXPORT_SYMBOL(giveup_altivec);
258
259 void enable_kernel_altivec(void)
260 {
261         unsigned long cpumsr;
262
263         WARN_ON(preemptible());
264
265         cpumsr = msr_check_and_set(MSR_VEC);
266
267         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
268                 check_if_tm_restore_required(current);
269                 /*
270                  * If a thread has already been reclaimed then the
271                  * checkpointed registers are on the CPU but have definitely
272                  * been saved by the reclaim code. Don't need to and *cannot*
273                  * giveup as this would save  to the 'live' structure not the
274                  * checkpointed structure.
275                  */
276                 if (!MSR_TM_ACTIVE(cpumsr) &&
277                      MSR_TM_ACTIVE(current->thread.regs->msr))
278                         return;
279                 __giveup_altivec(current);
280         }
281 }
282 EXPORT_SYMBOL(enable_kernel_altivec);
283
284 /*
285  * Make sure the VMX/Altivec register state in the
286  * the thread_struct is up to date for task tsk.
287  */
288 void flush_altivec_to_thread(struct task_struct *tsk)
289 {
290         if (tsk->thread.regs) {
291                 preempt_disable();
292                 if (tsk->thread.regs->msr & MSR_VEC) {
293                         BUG_ON(tsk != current);
294                         giveup_altivec(tsk);
295                 }
296                 preempt_enable();
297         }
298 }
299 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
300 #endif /* CONFIG_ALTIVEC */
301
302 #ifdef CONFIG_VSX
303 static void __giveup_vsx(struct task_struct *tsk)
304 {
305         unsigned long msr = tsk->thread.regs->msr;
306
307         /*
308          * We should never be ssetting MSR_VSX without also setting
309          * MSR_FP and MSR_VEC
310          */
311         WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
312
313         /* __giveup_fpu will clear MSR_VSX */
314         if (msr & MSR_FP)
315                 __giveup_fpu(tsk);
316         if (msr & MSR_VEC)
317                 __giveup_altivec(tsk);
318 }
319
320 static void giveup_vsx(struct task_struct *tsk)
321 {
322         check_if_tm_restore_required(tsk);
323
324         msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
325         __giveup_vsx(tsk);
326         msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
327 }
328
329 void enable_kernel_vsx(void)
330 {
331         unsigned long cpumsr;
332
333         WARN_ON(preemptible());
334
335         cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
336
337         if (current->thread.regs &&
338             (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
339                 check_if_tm_restore_required(current);
340                 /*
341                  * If a thread has already been reclaimed then the
342                  * checkpointed registers are on the CPU but have definitely
343                  * been saved by the reclaim code. Don't need to and *cannot*
344                  * giveup as this would save  to the 'live' structure not the
345                  * checkpointed structure.
346                  */
347                 if (!MSR_TM_ACTIVE(cpumsr) &&
348                      MSR_TM_ACTIVE(current->thread.regs->msr))
349                         return;
350                 __giveup_vsx(current);
351         }
352 }
353 EXPORT_SYMBOL(enable_kernel_vsx);
354
355 void flush_vsx_to_thread(struct task_struct *tsk)
356 {
357         if (tsk->thread.regs) {
358                 preempt_disable();
359                 if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
360                         BUG_ON(tsk != current);
361                         giveup_vsx(tsk);
362                 }
363                 preempt_enable();
364         }
365 }
366 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
367 #endif /* CONFIG_VSX */
368
369 #ifdef CONFIG_SPE
370 void giveup_spe(struct task_struct *tsk)
371 {
372         check_if_tm_restore_required(tsk);
373
374         msr_check_and_set(MSR_SPE);
375         __giveup_spe(tsk);
376         msr_check_and_clear(MSR_SPE);
377 }
378 EXPORT_SYMBOL(giveup_spe);
379
380 void enable_kernel_spe(void)
381 {
382         WARN_ON(preemptible());
383
384         msr_check_and_set(MSR_SPE);
385
386         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
387                 check_if_tm_restore_required(current);
388                 __giveup_spe(current);
389         }
390 }
391 EXPORT_SYMBOL(enable_kernel_spe);
392
393 void flush_spe_to_thread(struct task_struct *tsk)
394 {
395         if (tsk->thread.regs) {
396                 preempt_disable();
397                 if (tsk->thread.regs->msr & MSR_SPE) {
398                         BUG_ON(tsk != current);
399                         tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
400                         giveup_spe(tsk);
401                 }
402                 preempt_enable();
403         }
404 }
405 #endif /* CONFIG_SPE */
406
407 static unsigned long msr_all_available;
408
409 static int __init init_msr_all_available(void)
410 {
411         if (IS_ENABLED(CONFIG_PPC_FPU))
412                 msr_all_available |= MSR_FP;
413         if (cpu_has_feature(CPU_FTR_ALTIVEC))
414                 msr_all_available |= MSR_VEC;
415         if (cpu_has_feature(CPU_FTR_VSX))
416                 msr_all_available |= MSR_VSX;
417         if (cpu_has_feature(CPU_FTR_SPE))
418                 msr_all_available |= MSR_SPE;
419
420         return 0;
421 }
422 early_initcall(init_msr_all_available);
423
424 void giveup_all(struct task_struct *tsk)
425 {
426         unsigned long usermsr;
427
428         if (!tsk->thread.regs)
429                 return;
430
431         check_if_tm_restore_required(tsk);
432
433         usermsr = tsk->thread.regs->msr;
434
435         if ((usermsr & msr_all_available) == 0)
436                 return;
437
438         msr_check_and_set(msr_all_available);
439
440         WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
441
442         if (usermsr & MSR_FP)
443                 __giveup_fpu(tsk);
444         if (usermsr & MSR_VEC)
445                 __giveup_altivec(tsk);
446         if (usermsr & MSR_SPE)
447                 __giveup_spe(tsk);
448
449         msr_check_and_clear(msr_all_available);
450 }
451 EXPORT_SYMBOL(giveup_all);
452
453 #ifdef CONFIG_PPC_BOOK3S_64
454 #ifdef CONFIG_PPC_FPU
455 static bool should_restore_fp(void)
456 {
457         if (current->thread.load_fp) {
458                 current->thread.load_fp++;
459                 return true;
460         }
461         return false;
462 }
463
464 static void do_restore_fp(void)
465 {
466         load_fp_state(&current->thread.fp_state);
467 }
468 #else
469 static bool should_restore_fp(void) { return false; }
470 static void do_restore_fp(void) { }
471 #endif /* CONFIG_PPC_FPU */
472
473 #ifdef CONFIG_ALTIVEC
474 static bool should_restore_altivec(void)
475 {
476         if (cpu_has_feature(CPU_FTR_ALTIVEC) && (current->thread.load_vec)) {
477                 current->thread.load_vec++;
478                 return true;
479         }
480         return false;
481 }
482
483 static void do_restore_altivec(void)
484 {
485         load_vr_state(&current->thread.vr_state);
486         current->thread.used_vr = 1;
487 }
488 #else
489 static bool should_restore_altivec(void) { return false; }
490 static void do_restore_altivec(void) { }
491 #endif /* CONFIG_ALTIVEC */
492
493 static bool should_restore_vsx(void)
494 {
495         if (cpu_has_feature(CPU_FTR_VSX))
496                 return true;
497         return false;
498 }
499 #ifdef CONFIG_VSX
500 static void do_restore_vsx(void)
501 {
502         current->thread.used_vsr = 1;
503 }
504 #else
505 static void do_restore_vsx(void) { }
506 #endif /* CONFIG_VSX */
507
508 /*
509  * The exception exit path calls restore_math() with interrupts hard disabled
510  * but the soft irq state not "reconciled". ftrace code that calls
511  * local_irq_save/restore causes warnings.
512  *
513  * Rather than complicate the exit path, just don't trace restore_math. This
514  * could be done by having ftrace entry code check for this un-reconciled
515  * condition where MSR[EE]=0 and PACA_IRQ_HARD_DIS is not set, and
516  * temporarily fix it up for the duration of the ftrace call.
517  */
518 void notrace restore_math(struct pt_regs *regs)
519 {
520         unsigned long msr;
521         unsigned long new_msr = 0;
522
523         msr = regs->msr;
524
525         /*
526          * new_msr tracks the facilities that are to be restored. Only reload
527          * if the bit is not set in the user MSR (if it is set, the registers
528          * are live for the user thread).
529          */
530         if ((!(msr & MSR_FP)) && should_restore_fp())
531                 new_msr |= MSR_FP;
532
533         if ((!(msr & MSR_VEC)) && should_restore_altivec())
534                 new_msr |= MSR_VEC;
535
536         if ((!(msr & MSR_VSX)) && should_restore_vsx()) {
537                 if (((msr | new_msr) & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC))
538                         new_msr |= MSR_VSX;
539         }
540
541         if (new_msr) {
542                 unsigned long fpexc_mode = 0;
543
544                 msr_check_and_set(new_msr);
545
546                 if (new_msr & MSR_FP) {
547                         do_restore_fp();
548
549                         // This also covers VSX, because VSX implies FP
550                         fpexc_mode = current->thread.fpexc_mode;
551                 }
552
553                 if (new_msr & MSR_VEC)
554                         do_restore_altivec();
555
556                 if (new_msr & MSR_VSX)
557                         do_restore_vsx();
558
559                 msr_check_and_clear(new_msr);
560
561                 regs->msr |= new_msr | fpexc_mode;
562         }
563 }
564 #endif /* CONFIG_PPC_BOOK3S_64 */
565
566 static void save_all(struct task_struct *tsk)
567 {
568         unsigned long usermsr;
569
570         if (!tsk->thread.regs)
571                 return;
572
573         usermsr = tsk->thread.regs->msr;
574
575         if ((usermsr & msr_all_available) == 0)
576                 return;
577
578         msr_check_and_set(msr_all_available);
579
580         WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
581
582         if (usermsr & MSR_FP)
583                 save_fpu(tsk);
584
585         if (usermsr & MSR_VEC)
586                 save_altivec(tsk);
587
588         if (usermsr & MSR_SPE)
589                 __giveup_spe(tsk);
590
591         msr_check_and_clear(msr_all_available);
592 }
593
594 void flush_all_to_thread(struct task_struct *tsk)
595 {
596         if (tsk->thread.regs) {
597                 preempt_disable();
598                 BUG_ON(tsk != current);
599 #ifdef CONFIG_SPE
600                 if (tsk->thread.regs->msr & MSR_SPE)
601                         tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
602 #endif
603                 save_all(tsk);
604
605                 preempt_enable();
606         }
607 }
608 EXPORT_SYMBOL(flush_all_to_thread);
609
610 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
611 void do_send_trap(struct pt_regs *regs, unsigned long address,
612                   unsigned long error_code, int breakpt)
613 {
614         current->thread.trap_nr = TRAP_HWBKPT;
615         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
616                         11, SIGSEGV) == NOTIFY_STOP)
617                 return;
618
619         /* Deliver the signal to userspace */
620         force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
621                                     (void __user *)address);
622 }
623 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
624
625 static void do_break_handler(struct pt_regs *regs)
626 {
627         struct arch_hw_breakpoint null_brk = {0};
628         struct arch_hw_breakpoint *info;
629         struct ppc_inst instr = ppc_inst(0);
630         int type = 0;
631         int size = 0;
632         unsigned long ea;
633         int i;
634
635         /*
636          * If underneath hw supports only one watchpoint, we know it
637          * caused exception. 8xx also falls into this category.
638          */
639         if (nr_wp_slots() == 1) {
640                 __set_breakpoint(0, &null_brk);
641                 current->thread.hw_brk[0] = null_brk;
642                 current->thread.hw_brk[0].flags |= HW_BRK_FLAG_DISABLED;
643                 return;
644         }
645
646         /* Otherwise findout which DAWR caused exception and disable it. */
647         wp_get_instr_detail(regs, &instr, &type, &size, &ea);
648
649         for (i = 0; i < nr_wp_slots(); i++) {
650                 info = &current->thread.hw_brk[i];
651                 if (!info->address)
652                         continue;
653
654                 if (wp_check_constraints(regs, instr, ea, type, size, info)) {
655                         __set_breakpoint(i, &null_brk);
656                         current->thread.hw_brk[i] = null_brk;
657                         current->thread.hw_brk[i].flags |= HW_BRK_FLAG_DISABLED;
658                 }
659         }
660 }
661
662 void do_break (struct pt_regs *regs, unsigned long address,
663                     unsigned long error_code)
664 {
665         current->thread.trap_nr = TRAP_HWBKPT;
666         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
667                         11, SIGSEGV) == NOTIFY_STOP)
668                 return;
669
670         if (debugger_break_match(regs))
671                 return;
672
673         /*
674          * We reach here only when watchpoint exception is generated by ptrace
675          * event (or hw is buggy!). Now if CONFIG_HAVE_HW_BREAKPOINT is set,
676          * watchpoint is already handled by hw_breakpoint_handler() so we don't
677          * have to do anything. But when CONFIG_HAVE_HW_BREAKPOINT is not set,
678          * we need to manually handle the watchpoint here.
679          */
680         if (!IS_ENABLED(CONFIG_HAVE_HW_BREAKPOINT))
681                 do_break_handler(regs);
682
683         /* Deliver the signal to userspace */
684         force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)address);
685 }
686 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
687
688 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk[HBP_NUM_MAX]);
689
690 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
691 /*
692  * Set the debug registers back to their default "safe" values.
693  */
694 static void set_debug_reg_defaults(struct thread_struct *thread)
695 {
696         thread->debug.iac1 = thread->debug.iac2 = 0;
697 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
698         thread->debug.iac3 = thread->debug.iac4 = 0;
699 #endif
700         thread->debug.dac1 = thread->debug.dac2 = 0;
701 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
702         thread->debug.dvc1 = thread->debug.dvc2 = 0;
703 #endif
704         thread->debug.dbcr0 = 0;
705 #ifdef CONFIG_BOOKE
706         /*
707          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
708          */
709         thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
710                         DBCR1_IAC3US | DBCR1_IAC4US;
711         /*
712          * Force Data Address Compare User/Supervisor bits to be User-only
713          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
714          */
715         thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
716 #else
717         thread->debug.dbcr1 = 0;
718 #endif
719 }
720
721 static void prime_debug_regs(struct debug_reg *debug)
722 {
723         /*
724          * We could have inherited MSR_DE from userspace, since
725          * it doesn't get cleared on exception entry.  Make sure
726          * MSR_DE is clear before we enable any debug events.
727          */
728         mtmsr(mfmsr() & ~MSR_DE);
729
730         mtspr(SPRN_IAC1, debug->iac1);
731         mtspr(SPRN_IAC2, debug->iac2);
732 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
733         mtspr(SPRN_IAC3, debug->iac3);
734         mtspr(SPRN_IAC4, debug->iac4);
735 #endif
736         mtspr(SPRN_DAC1, debug->dac1);
737         mtspr(SPRN_DAC2, debug->dac2);
738 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
739         mtspr(SPRN_DVC1, debug->dvc1);
740         mtspr(SPRN_DVC2, debug->dvc2);
741 #endif
742         mtspr(SPRN_DBCR0, debug->dbcr0);
743         mtspr(SPRN_DBCR1, debug->dbcr1);
744 #ifdef CONFIG_BOOKE
745         mtspr(SPRN_DBCR2, debug->dbcr2);
746 #endif
747 }
748 /*
749  * Unless neither the old or new thread are making use of the
750  * debug registers, set the debug registers from the values
751  * stored in the new thread.
752  */
753 void switch_booke_debug_regs(struct debug_reg *new_debug)
754 {
755         if ((current->thread.debug.dbcr0 & DBCR0_IDM)
756                 || (new_debug->dbcr0 & DBCR0_IDM))
757                         prime_debug_regs(new_debug);
758 }
759 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
760 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
761 #ifndef CONFIG_HAVE_HW_BREAKPOINT
762 static void set_breakpoint(int i, struct arch_hw_breakpoint *brk)
763 {
764         preempt_disable();
765         __set_breakpoint(i, brk);
766         preempt_enable();
767 }
768
769 static void set_debug_reg_defaults(struct thread_struct *thread)
770 {
771         int i;
772         struct arch_hw_breakpoint null_brk = {0};
773
774         for (i = 0; i < nr_wp_slots(); i++) {
775                 thread->hw_brk[i] = null_brk;
776                 if (ppc_breakpoint_available())
777                         set_breakpoint(i, &thread->hw_brk[i]);
778         }
779 }
780
781 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
782                                 struct arch_hw_breakpoint *b)
783 {
784         if (a->address != b->address)
785                 return false;
786         if (a->type != b->type)
787                 return false;
788         if (a->len != b->len)
789                 return false;
790         /* no need to check hw_len. it's calculated from address and len */
791         return true;
792 }
793
794 static void switch_hw_breakpoint(struct task_struct *new)
795 {
796         int i;
797
798         for (i = 0; i < nr_wp_slots(); i++) {
799                 if (likely(hw_brk_match(this_cpu_ptr(&current_brk[i]),
800                                         &new->thread.hw_brk[i])))
801                         continue;
802
803                 __set_breakpoint(i, &new->thread.hw_brk[i]);
804         }
805 }
806 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
807 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
808
809 static inline int set_dabr(struct arch_hw_breakpoint *brk)
810 {
811         unsigned long dabr, dabrx;
812
813         dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
814         dabrx = ((brk->type >> 3) & 0x7);
815
816         if (ppc_md.set_dabr)
817                 return ppc_md.set_dabr(dabr, dabrx);
818
819         if (IS_ENABLED(CONFIG_PPC_ADV_DEBUG_REGS)) {
820                 mtspr(SPRN_DAC1, dabr);
821                 if (IS_ENABLED(CONFIG_PPC_47x))
822                         isync();
823                 return 0;
824         } else if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
825                 mtspr(SPRN_DABR, dabr);
826                 if (cpu_has_feature(CPU_FTR_DABRX))
827                         mtspr(SPRN_DABRX, dabrx);
828                 return 0;
829         } else {
830                 return -EINVAL;
831         }
832 }
833
834 static inline int set_breakpoint_8xx(struct arch_hw_breakpoint *brk)
835 {
836         unsigned long lctrl1 = LCTRL1_CTE_GT | LCTRL1_CTF_LT | LCTRL1_CRWE_RW |
837                                LCTRL1_CRWF_RW;
838         unsigned long lctrl2 = LCTRL2_LW0EN | LCTRL2_LW0LADC | LCTRL2_SLW0EN;
839         unsigned long start_addr = ALIGN_DOWN(brk->address, HW_BREAKPOINT_SIZE);
840         unsigned long end_addr = ALIGN(brk->address + brk->len, HW_BREAKPOINT_SIZE);
841
842         if (start_addr == 0)
843                 lctrl2 |= LCTRL2_LW0LA_F;
844         else if (end_addr == 0)
845                 lctrl2 |= LCTRL2_LW0LA_E;
846         else
847                 lctrl2 |= LCTRL2_LW0LA_EandF;
848
849         mtspr(SPRN_LCTRL2, 0);
850
851         if ((brk->type & HW_BRK_TYPE_RDWR) == 0)
852                 return 0;
853
854         if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
855                 lctrl1 |= LCTRL1_CRWE_RO | LCTRL1_CRWF_RO;
856         if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
857                 lctrl1 |= LCTRL1_CRWE_WO | LCTRL1_CRWF_WO;
858
859         mtspr(SPRN_CMPE, start_addr - 1);
860         mtspr(SPRN_CMPF, end_addr);
861         mtspr(SPRN_LCTRL1, lctrl1);
862         mtspr(SPRN_LCTRL2, lctrl2);
863
864         return 0;
865 }
866
867 void __set_breakpoint(int nr, struct arch_hw_breakpoint *brk)
868 {
869         memcpy(this_cpu_ptr(&current_brk[nr]), brk, sizeof(*brk));
870
871         if (dawr_enabled())
872                 // Power8 or later
873                 set_dawr(nr, brk);
874         else if (IS_ENABLED(CONFIG_PPC_8xx))
875                 set_breakpoint_8xx(brk);
876         else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
877                 // Power7 or earlier
878                 set_dabr(brk);
879         else
880                 // Shouldn't happen due to higher level checks
881                 WARN_ON_ONCE(1);
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 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
897
898 static inline bool tm_enabled(struct task_struct *tsk)
899 {
900         return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
901 }
902
903 static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
904 {
905         /*
906          * Use the current MSR TM suspended bit to track if we have
907          * checkpointed state outstanding.
908          * On signal delivery, we'd normally reclaim the checkpointed
909          * state to obtain stack pointer (see:get_tm_stackpointer()).
910          * This will then directly return to userspace without going
911          * through __switch_to(). However, if the stack frame is bad,
912          * we need to exit this thread which calls __switch_to() which
913          * will again attempt to reclaim the already saved tm state.
914          * Hence we need to check that we've not already reclaimed
915          * this state.
916          * We do this using the current MSR, rather tracking it in
917          * some specific thread_struct bit, as it has the additional
918          * benefit of checking for a potential TM bad thing exception.
919          */
920         if (!MSR_TM_SUSPENDED(mfmsr()))
921                 return;
922
923         giveup_all(container_of(thr, struct task_struct, thread));
924
925         tm_reclaim(thr, cause);
926
927         /*
928          * If we are in a transaction and FP is off then we can't have
929          * used FP inside that transaction. Hence the checkpointed
930          * state is the same as the live state. We need to copy the
931          * live state to the checkpointed state so that when the
932          * transaction is restored, the checkpointed state is correct
933          * and the aborted transaction sees the correct state. We use
934          * ckpt_regs.msr here as that's what tm_reclaim will use to
935          * determine if it's going to write the checkpointed state or
936          * not. So either this will write the checkpointed registers,
937          * or reclaim will. Similarly for VMX.
938          */
939         if ((thr->ckpt_regs.msr & MSR_FP) == 0)
940                 memcpy(&thr->ckfp_state, &thr->fp_state,
941                        sizeof(struct thread_fp_state));
942         if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
943                 memcpy(&thr->ckvr_state, &thr->vr_state,
944                        sizeof(struct thread_vr_state));
945 }
946
947 void tm_reclaim_current(uint8_t cause)
948 {
949         tm_enable();
950         tm_reclaim_thread(&current->thread, cause);
951 }
952
953 static inline void tm_reclaim_task(struct task_struct *tsk)
954 {
955         /* We have to work out if we're switching from/to a task that's in the
956          * middle of a transaction.
957          *
958          * In switching we need to maintain a 2nd register state as
959          * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
960          * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
961          * ckvr_state
962          *
963          * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
964          */
965         struct thread_struct *thr = &tsk->thread;
966
967         if (!thr->regs)
968                 return;
969
970         if (!MSR_TM_ACTIVE(thr->regs->msr))
971                 goto out_and_saveregs;
972
973         WARN_ON(tm_suspend_disabled);
974
975         TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
976                  "ccr=%lx, msr=%lx, trap=%lx)\n",
977                  tsk->pid, thr->regs->nip,
978                  thr->regs->ccr, thr->regs->msr,
979                  thr->regs->trap);
980
981         tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
982
983         TM_DEBUG("--- tm_reclaim on pid %d complete\n",
984                  tsk->pid);
985
986 out_and_saveregs:
987         /* Always save the regs here, even if a transaction's not active.
988          * This context-switches a thread's TM info SPRs.  We do it here to
989          * be consistent with the restore path (in recheckpoint) which
990          * cannot happen later in _switch().
991          */
992         tm_save_sprs(thr);
993 }
994
995 extern void __tm_recheckpoint(struct thread_struct *thread);
996
997 void tm_recheckpoint(struct thread_struct *thread)
998 {
999         unsigned long flags;
1000
1001         if (!(thread->regs->msr & MSR_TM))
1002                 return;
1003
1004         /* We really can't be interrupted here as the TEXASR registers can't
1005          * change and later in the trecheckpoint code, we have a userspace R1.
1006          * So let's hard disable over this region.
1007          */
1008         local_irq_save(flags);
1009         hard_irq_disable();
1010
1011         /* The TM SPRs are restored here, so that TEXASR.FS can be set
1012          * before the trecheckpoint and no explosion occurs.
1013          */
1014         tm_restore_sprs(thread);
1015
1016         __tm_recheckpoint(thread);
1017
1018         local_irq_restore(flags);
1019 }
1020
1021 static inline void tm_recheckpoint_new_task(struct task_struct *new)
1022 {
1023         if (!cpu_has_feature(CPU_FTR_TM))
1024                 return;
1025
1026         /* Recheckpoint the registers of the thread we're about to switch to.
1027          *
1028          * If the task was using FP, we non-lazily reload both the original and
1029          * the speculative FP register states.  This is because the kernel
1030          * doesn't see if/when a TM rollback occurs, so if we take an FP
1031          * unavailable later, we are unable to determine which set of FP regs
1032          * need to be restored.
1033          */
1034         if (!tm_enabled(new))
1035                 return;
1036
1037         if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
1038                 tm_restore_sprs(&new->thread);
1039                 return;
1040         }
1041         /* Recheckpoint to restore original checkpointed register state. */
1042         TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
1043                  new->pid, new->thread.regs->msr);
1044
1045         tm_recheckpoint(&new->thread);
1046
1047         /*
1048          * The checkpointed state has been restored but the live state has
1049          * not, ensure all the math functionality is turned off to trigger
1050          * restore_math() to reload.
1051          */
1052         new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1053
1054         TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1055                  "(kernel msr 0x%lx)\n",
1056                  new->pid, mfmsr());
1057 }
1058
1059 static inline void __switch_to_tm(struct task_struct *prev,
1060                 struct task_struct *new)
1061 {
1062         if (cpu_has_feature(CPU_FTR_TM)) {
1063                 if (tm_enabled(prev) || tm_enabled(new))
1064                         tm_enable();
1065
1066                 if (tm_enabled(prev)) {
1067                         prev->thread.load_tm++;
1068                         tm_reclaim_task(prev);
1069                         if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1070                                 prev->thread.regs->msr &= ~MSR_TM;
1071                 }
1072
1073                 tm_recheckpoint_new_task(new);
1074         }
1075 }
1076
1077 /*
1078  * This is called if we are on the way out to userspace and the
1079  * TIF_RESTORE_TM flag is set.  It checks if we need to reload
1080  * FP and/or vector state and does so if necessary.
1081  * If userspace is inside a transaction (whether active or
1082  * suspended) and FP/VMX/VSX instructions have ever been enabled
1083  * inside that transaction, then we have to keep them enabled
1084  * and keep the FP/VMX/VSX state loaded while ever the transaction
1085  * continues.  The reason is that if we didn't, and subsequently
1086  * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1087  * we don't know whether it's the same transaction, and thus we
1088  * don't know which of the checkpointed state and the transactional
1089  * state to use.
1090  */
1091 void restore_tm_state(struct pt_regs *regs)
1092 {
1093         unsigned long msr_diff;
1094
1095         /*
1096          * This is the only moment we should clear TIF_RESTORE_TM as
1097          * it is here that ckpt_regs.msr and pt_regs.msr become the same
1098          * again, anything else could lead to an incorrect ckpt_msr being
1099          * saved and therefore incorrect signal contexts.
1100          */
1101         clear_thread_flag(TIF_RESTORE_TM);
1102         if (!MSR_TM_ACTIVE(regs->msr))
1103                 return;
1104
1105         msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1106         msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1107
1108         /* Ensure that restore_math() will restore */
1109         if (msr_diff & MSR_FP)
1110                 current->thread.load_fp = 1;
1111 #ifdef CONFIG_ALTIVEC
1112         if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1113                 current->thread.load_vec = 1;
1114 #endif
1115         restore_math(regs);
1116
1117         regs->msr |= msr_diff;
1118 }
1119
1120 #else
1121 #define tm_recheckpoint_new_task(new)
1122 #define __switch_to_tm(prev, new)
1123 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1124
1125 static inline void save_sprs(struct thread_struct *t)
1126 {
1127 #ifdef CONFIG_ALTIVEC
1128         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1129                 t->vrsave = mfspr(SPRN_VRSAVE);
1130 #endif
1131 #ifdef CONFIG_PPC_BOOK3S_64
1132         if (cpu_has_feature(CPU_FTR_DSCR))
1133                 t->dscr = mfspr(SPRN_DSCR);
1134
1135         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1136                 t->bescr = mfspr(SPRN_BESCR);
1137                 t->ebbhr = mfspr(SPRN_EBBHR);
1138                 t->ebbrr = mfspr(SPRN_EBBRR);
1139
1140                 t->fscr = mfspr(SPRN_FSCR);
1141
1142                 /*
1143                  * Note that the TAR is not available for use in the kernel.
1144                  * (To provide this, the TAR should be backed up/restored on
1145                  * exception entry/exit instead, and be in pt_regs.  FIXME,
1146                  * this should be in pt_regs anyway (for debug).)
1147                  */
1148                 t->tar = mfspr(SPRN_TAR);
1149         }
1150 #endif
1151 }
1152
1153 static inline void restore_sprs(struct thread_struct *old_thread,
1154                                 struct thread_struct *new_thread)
1155 {
1156 #ifdef CONFIG_ALTIVEC
1157         if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1158             old_thread->vrsave != new_thread->vrsave)
1159                 mtspr(SPRN_VRSAVE, new_thread->vrsave);
1160 #endif
1161 #ifdef CONFIG_PPC_BOOK3S_64
1162         if (cpu_has_feature(CPU_FTR_DSCR)) {
1163                 u64 dscr = get_paca()->dscr_default;
1164                 if (new_thread->dscr_inherit)
1165                         dscr = new_thread->dscr;
1166
1167                 if (old_thread->dscr != dscr)
1168                         mtspr(SPRN_DSCR, dscr);
1169         }
1170
1171         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1172                 if (old_thread->bescr != new_thread->bescr)
1173                         mtspr(SPRN_BESCR, new_thread->bescr);
1174                 if (old_thread->ebbhr != new_thread->ebbhr)
1175                         mtspr(SPRN_EBBHR, new_thread->ebbhr);
1176                 if (old_thread->ebbrr != new_thread->ebbrr)
1177                         mtspr(SPRN_EBBRR, new_thread->ebbrr);
1178
1179                 if (old_thread->fscr != new_thread->fscr)
1180                         mtspr(SPRN_FSCR, new_thread->fscr);
1181
1182                 if (old_thread->tar != new_thread->tar)
1183                         mtspr(SPRN_TAR, new_thread->tar);
1184         }
1185
1186         if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1187             old_thread->tidr != new_thread->tidr)
1188                 mtspr(SPRN_TIDR, new_thread->tidr);
1189 #endif
1190
1191 }
1192
1193 struct task_struct *__switch_to(struct task_struct *prev,
1194         struct task_struct *new)
1195 {
1196         struct thread_struct *new_thread, *old_thread;
1197         struct task_struct *last;
1198 #ifdef CONFIG_PPC_BOOK3S_64
1199         struct ppc64_tlb_batch *batch;
1200 #endif
1201
1202         new_thread = &new->thread;
1203         old_thread = &current->thread;
1204
1205         WARN_ON(!irqs_disabled());
1206
1207 #ifdef CONFIG_PPC_BOOK3S_64
1208         batch = this_cpu_ptr(&ppc64_tlb_batch);
1209         if (batch->active) {
1210                 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1211                 if (batch->index)
1212                         __flush_tlb_pending(batch);
1213                 batch->active = 0;
1214         }
1215 #endif /* CONFIG_PPC_BOOK3S_64 */
1216
1217 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1218         switch_booke_debug_regs(&new->thread.debug);
1219 #else
1220 /*
1221  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1222  * schedule DABR
1223  */
1224 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1225         switch_hw_breakpoint(new);
1226 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1227 #endif
1228
1229         /*
1230          * We need to save SPRs before treclaim/trecheckpoint as these will
1231          * change a number of them.
1232          */
1233         save_sprs(&prev->thread);
1234
1235         /* Save FPU, Altivec, VSX and SPE state */
1236         giveup_all(prev);
1237
1238         __switch_to_tm(prev, new);
1239
1240         if (!radix_enabled()) {
1241                 /*
1242                  * We can't take a PMU exception inside _switch() since there
1243                  * is a window where the kernel stack SLB and the kernel stack
1244                  * are out of sync. Hard disable here.
1245                  */
1246                 hard_irq_disable();
1247         }
1248
1249         /*
1250          * Call restore_sprs() before calling _switch(). If we move it after
1251          * _switch() then we miss out on calling it for new tasks. The reason
1252          * for this is we manually create a stack frame for new tasks that
1253          * directly returns through ret_from_fork() or
1254          * ret_from_kernel_thread(). See copy_thread() for details.
1255          */
1256         restore_sprs(old_thread, new_thread);
1257
1258         last = _switch(old_thread, new_thread);
1259
1260 #ifdef CONFIG_PPC_BOOK3S_64
1261         if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1262                 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1263                 batch = this_cpu_ptr(&ppc64_tlb_batch);
1264                 batch->active = 1;
1265         }
1266
1267         if (current->thread.regs) {
1268                 restore_math(current->thread.regs);
1269
1270                 /*
1271                  * On POWER9 the copy-paste buffer can only paste into
1272                  * foreign real addresses, so unprivileged processes can not
1273                  * see the data or use it in any way unless they have
1274                  * foreign real mappings. If the new process has the foreign
1275                  * real address mappings, we must issue a cp_abort to clear
1276                  * any state and prevent snooping, corruption or a covert
1277                  * channel. ISA v3.1 supports paste into local memory.
1278                  */
1279                 if (current->mm &&
1280                         (cpu_has_feature(CPU_FTR_ARCH_31) ||
1281                         atomic_read(&current->mm->context.vas_windows)))
1282                         asm volatile(PPC_CP_ABORT);
1283         }
1284 #endif /* CONFIG_PPC_BOOK3S_64 */
1285
1286         return last;
1287 }
1288
1289 #define NR_INSN_TO_PRINT        16
1290
1291 static void show_instructions(struct pt_regs *regs)
1292 {
1293         int i;
1294         unsigned long nip = regs->nip;
1295         unsigned long pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1296
1297         printk("Instruction dump:");
1298
1299         /*
1300          * If we were executing with the MMU off for instructions, adjust pc
1301          * rather than printing XXXXXXXX.
1302          */
1303         if (!IS_ENABLED(CONFIG_BOOKE) && !(regs->msr & MSR_IR)) {
1304                 pc = (unsigned long)phys_to_virt(pc);
1305                 nip = (unsigned long)phys_to_virt(regs->nip);
1306         }
1307
1308         for (i = 0; i < NR_INSN_TO_PRINT; i++) {
1309                 int instr;
1310
1311                 if (!(i % 8))
1312                         pr_cont("\n");
1313
1314                 if (!__kernel_text_address(pc) ||
1315                     get_kernel_nofault(instr, (const void *)pc)) {
1316                         pr_cont("XXXXXXXX ");
1317                 } else {
1318                         if (nip == pc)
1319                                 pr_cont("<%08x> ", instr);
1320                         else
1321                                 pr_cont("%08x ", instr);
1322                 }
1323
1324                 pc += sizeof(int);
1325         }
1326
1327         pr_cont("\n");
1328 }
1329
1330 void show_user_instructions(struct pt_regs *regs)
1331 {
1332         unsigned long pc;
1333         int n = NR_INSN_TO_PRINT;
1334         struct seq_buf s;
1335         char buf[96]; /* enough for 8 times 9 + 2 chars */
1336
1337         pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1338
1339         seq_buf_init(&s, buf, sizeof(buf));
1340
1341         while (n) {
1342                 int i;
1343
1344                 seq_buf_clear(&s);
1345
1346                 for (i = 0; i < 8 && n; i++, n--, pc += sizeof(int)) {
1347                         int instr;
1348
1349                         if (copy_from_user_nofault(&instr, (void __user *)pc,
1350                                         sizeof(instr))) {
1351                                 seq_buf_printf(&s, "XXXXXXXX ");
1352                                 continue;
1353                         }
1354                         seq_buf_printf(&s, regs->nip == pc ? "<%08x> " : "%08x ", instr);
1355                 }
1356
1357                 if (!seq_buf_has_overflowed(&s))
1358                         pr_info("%s[%d]: code: %s\n", current->comm,
1359                                 current->pid, s.buffer);
1360         }
1361 }
1362
1363 struct regbit {
1364         unsigned long bit;
1365         const char *name;
1366 };
1367
1368 static struct regbit msr_bits[] = {
1369 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1370         {MSR_SF,        "SF"},
1371         {MSR_HV,        "HV"},
1372 #endif
1373         {MSR_VEC,       "VEC"},
1374         {MSR_VSX,       "VSX"},
1375 #ifdef CONFIG_BOOKE
1376         {MSR_CE,        "CE"},
1377 #endif
1378         {MSR_EE,        "EE"},
1379         {MSR_PR,        "PR"},
1380         {MSR_FP,        "FP"},
1381         {MSR_ME,        "ME"},
1382 #ifdef CONFIG_BOOKE
1383         {MSR_DE,        "DE"},
1384 #else
1385         {MSR_SE,        "SE"},
1386         {MSR_BE,        "BE"},
1387 #endif
1388         {MSR_IR,        "IR"},
1389         {MSR_DR,        "DR"},
1390         {MSR_PMM,       "PMM"},
1391 #ifndef CONFIG_BOOKE
1392         {MSR_RI,        "RI"},
1393         {MSR_LE,        "LE"},
1394 #endif
1395         {0,             NULL}
1396 };
1397
1398 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1399 {
1400         const char *s = "";
1401
1402         for (; bits->bit; ++bits)
1403                 if (val & bits->bit) {
1404                         pr_cont("%s%s", s, bits->name);
1405                         s = sep;
1406                 }
1407 }
1408
1409 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1410 static struct regbit msr_tm_bits[] = {
1411         {MSR_TS_T,      "T"},
1412         {MSR_TS_S,      "S"},
1413         {MSR_TM,        "E"},
1414         {0,             NULL}
1415 };
1416
1417 static void print_tm_bits(unsigned long val)
1418 {
1419 /*
1420  * This only prints something if at least one of the TM bit is set.
1421  * Inside the TM[], the output means:
1422  *   E: Enabled         (bit 32)
1423  *   S: Suspended       (bit 33)
1424  *   T: Transactional   (bit 34)
1425  */
1426         if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1427                 pr_cont(",TM[");
1428                 print_bits(val, msr_tm_bits, "");
1429                 pr_cont("]");
1430         }
1431 }
1432 #else
1433 static void print_tm_bits(unsigned long val) {}
1434 #endif
1435
1436 static void print_msr_bits(unsigned long val)
1437 {
1438         pr_cont("<");
1439         print_bits(val, msr_bits, ",");
1440         print_tm_bits(val);
1441         pr_cont(">");
1442 }
1443
1444 #ifdef CONFIG_PPC64
1445 #define REG             "%016lx"
1446 #define REGS_PER_LINE   4
1447 #define LAST_VOLATILE   13
1448 #else
1449 #define REG             "%08lx"
1450 #define REGS_PER_LINE   8
1451 #define LAST_VOLATILE   12
1452 #endif
1453
1454 static void __show_regs(struct pt_regs *regs)
1455 {
1456         int i, trap;
1457
1458         printk("NIP:  "REG" LR: "REG" CTR: "REG"\n",
1459                regs->nip, regs->link, regs->ctr);
1460         printk("REGS: %px TRAP: %04lx   %s  (%s)\n",
1461                regs, regs->trap, print_tainted(), init_utsname()->release);
1462         printk("MSR:  "REG" ", regs->msr);
1463         print_msr_bits(regs->msr);
1464         pr_cont("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1465         trap = TRAP(regs);
1466         if (!trap_is_syscall(regs) && cpu_has_feature(CPU_FTR_CFAR))
1467                 pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1468         if (trap == 0x200 || trap == 0x300 || trap == 0x600) {
1469                 if (IS_ENABLED(CONFIG_4xx) || IS_ENABLED(CONFIG_BOOKE))
1470                         pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1471                 else
1472                         pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1473         }
1474
1475 #ifdef CONFIG_PPC64
1476         pr_cont("IRQMASK: %lx ", regs->softe);
1477 #endif
1478 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1479         if (MSR_TM_ACTIVE(regs->msr))
1480                 pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1481 #endif
1482
1483         for (i = 0;  i < 32;  i++) {
1484                 if ((i % REGS_PER_LINE) == 0)
1485                         pr_cont("\nGPR%02d: ", i);
1486                 pr_cont(REG " ", regs->gpr[i]);
1487                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
1488                         break;
1489         }
1490         pr_cont("\n");
1491         /*
1492          * Lookup NIP late so we have the best change of getting the
1493          * above info out without failing
1494          */
1495         if (IS_ENABLED(CONFIG_KALLSYMS)) {
1496                 printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1497                 printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1498         }
1499 }
1500
1501 void show_regs(struct pt_regs *regs)
1502 {
1503         show_regs_print_info(KERN_DEFAULT);
1504         __show_regs(regs);
1505         show_stack(current, (unsigned long *) regs->gpr[1], KERN_DEFAULT);
1506         if (!user_mode(regs))
1507                 show_instructions(regs);
1508 }
1509
1510 void flush_thread(void)
1511 {
1512 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1513         flush_ptrace_hw_breakpoint(current);
1514 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1515         set_debug_reg_defaults(&current->thread);
1516 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1517 }
1518
1519 void arch_setup_new_exec(void)
1520 {
1521
1522 #ifdef CONFIG_PPC_BOOK3S_64
1523         if (!radix_enabled())
1524                 hash__setup_new_exec();
1525 #endif
1526         /*
1527          * If we exec out of a kernel thread then thread.regs will not be
1528          * set.  Do it now.
1529          */
1530         if (!current->thread.regs) {
1531                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1532                 current->thread.regs = regs - 1;
1533         }
1534
1535 #ifdef CONFIG_PPC_MEM_KEYS
1536         current->thread.regs->amr  = default_amr;
1537         current->thread.regs->iamr  = default_iamr;
1538 #endif
1539 }
1540
1541 #ifdef CONFIG_PPC64
1542 /**
1543  * Assign a TIDR (thread ID) for task @t and set it in the thread
1544  * structure. For now, we only support setting TIDR for 'current' task.
1545  *
1546  * Since the TID value is a truncated form of it PID, it is possible
1547  * (but unlikely) for 2 threads to have the same TID. In the unlikely event
1548  * that 2 threads share the same TID and are waiting, one of the following
1549  * cases will happen:
1550  *
1551  * 1. The correct thread is running, the wrong thread is not
1552  * In this situation, the correct thread is woken and proceeds to pass it's
1553  * condition check.
1554  *
1555  * 2. Neither threads are running
1556  * In this situation, neither thread will be woken. When scheduled, the waiting
1557  * threads will execute either a wait, which will return immediately, followed
1558  * by a condition check, which will pass for the correct thread and fail
1559  * for the wrong thread, or they will execute the condition check immediately.
1560  *
1561  * 3. The wrong thread is running, the correct thread is not
1562  * The wrong thread will be woken, but will fail it's condition check and
1563  * re-execute wait. The correct thread, when scheduled, will execute either
1564  * it's condition check (which will pass), or wait, which returns immediately
1565  * when called the first time after the thread is scheduled, followed by it's
1566  * condition check (which will pass).
1567  *
1568  * 4. Both threads are running
1569  * Both threads will be woken. The wrong thread will fail it's condition check
1570  * and execute another wait, while the correct thread will pass it's condition
1571  * check.
1572  *
1573  * @t: the task to set the thread ID for
1574  */
1575 int set_thread_tidr(struct task_struct *t)
1576 {
1577         if (!cpu_has_feature(CPU_FTR_P9_TIDR))
1578                 return -EINVAL;
1579
1580         if (t != current)
1581                 return -EINVAL;
1582
1583         if (t->thread.tidr)
1584                 return 0;
1585
1586         t->thread.tidr = (u16)task_pid_nr(t);
1587         mtspr(SPRN_TIDR, t->thread.tidr);
1588
1589         return 0;
1590 }
1591 EXPORT_SYMBOL_GPL(set_thread_tidr);
1592
1593 #endif /* CONFIG_PPC64 */
1594
1595 void
1596 release_thread(struct task_struct *t)
1597 {
1598 }
1599
1600 /*
1601  * this gets called so that we can store coprocessor state into memory and
1602  * copy the current task into the new thread.
1603  */
1604 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1605 {
1606         flush_all_to_thread(src);
1607         /*
1608          * Flush TM state out so we can copy it.  __switch_to_tm() does this
1609          * flush but it removes the checkpointed state from the current CPU and
1610          * transitions the CPU out of TM mode.  Hence we need to call
1611          * tm_recheckpoint_new_task() (on the same task) to restore the
1612          * checkpointed state back and the TM mode.
1613          *
1614          * Can't pass dst because it isn't ready. Doesn't matter, passing
1615          * dst is only important for __switch_to()
1616          */
1617         __switch_to_tm(src, src);
1618
1619         *dst = *src;
1620
1621         clear_task_ebb(dst);
1622
1623         return 0;
1624 }
1625
1626 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1627 {
1628 #ifdef CONFIG_PPC_BOOK3S_64
1629         unsigned long sp_vsid;
1630         unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1631
1632         if (radix_enabled())
1633                 return;
1634
1635         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1636                 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1637                         << SLB_VSID_SHIFT_1T;
1638         else
1639                 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1640                         << SLB_VSID_SHIFT;
1641         sp_vsid |= SLB_VSID_KERNEL | llp;
1642         p->thread.ksp_vsid = sp_vsid;
1643 #endif
1644 }
1645
1646 /*
1647  * Copy a thread..
1648  */
1649
1650 /*
1651  * Copy architecture-specific thread state
1652  */
1653 int copy_thread(unsigned long clone_flags, unsigned long usp,
1654                 unsigned long kthread_arg, struct task_struct *p,
1655                 unsigned long tls)
1656 {
1657         struct pt_regs *childregs, *kregs;
1658         extern void ret_from_fork(void);
1659         extern void ret_from_fork_scv(void);
1660         extern void ret_from_kernel_thread(void);
1661         void (*f)(void);
1662         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1663         struct thread_info *ti = task_thread_info(p);
1664 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1665         int i;
1666 #endif
1667
1668         klp_init_thread_info(p);
1669
1670         /* Copy registers */
1671         sp -= sizeof(struct pt_regs);
1672         childregs = (struct pt_regs *) sp;
1673         if (unlikely(p->flags & PF_KTHREAD)) {
1674                 /* kernel thread */
1675                 memset(childregs, 0, sizeof(struct pt_regs));
1676                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
1677                 /* function */
1678                 if (usp)
1679                         childregs->gpr[14] = ppc_function_entry((void *)usp);
1680 #ifdef CONFIG_PPC64
1681                 clear_tsk_thread_flag(p, TIF_32BIT);
1682                 childregs->softe = IRQS_ENABLED;
1683 #endif
1684                 childregs->gpr[15] = kthread_arg;
1685                 p->thread.regs = NULL;  /* no user register state */
1686                 ti->flags |= _TIF_RESTOREALL;
1687                 f = ret_from_kernel_thread;
1688         } else {
1689                 /* user thread */
1690                 struct pt_regs *regs = current_pt_regs();
1691                 CHECK_FULL_REGS(regs);
1692                 *childregs = *regs;
1693                 if (usp)
1694                         childregs->gpr[1] = usp;
1695                 p->thread.regs = childregs;
1696                 /* 64s sets this in ret_from_fork */
1697                 if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64))
1698                         childregs->gpr[3] = 0;  /* Result from fork() */
1699                 if (clone_flags & CLONE_SETTLS) {
1700                         if (!is_32bit_task())
1701                                 childregs->gpr[13] = tls;
1702                         else
1703                                 childregs->gpr[2] = tls;
1704                 }
1705
1706                 if (trap_is_scv(regs))
1707                         f = ret_from_fork_scv;
1708                 else
1709                         f = ret_from_fork;
1710         }
1711         childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1712         sp -= STACK_FRAME_OVERHEAD;
1713
1714         /*
1715          * The way this works is that at some point in the future
1716          * some task will call _switch to switch to the new task.
1717          * That will pop off the stack frame created below and start
1718          * the new task running at ret_from_fork.  The new task will
1719          * do some house keeping and then return from the fork or clone
1720          * system call, using the stack frame created above.
1721          */
1722         ((unsigned long *)sp)[0] = 0;
1723         sp -= sizeof(struct pt_regs);
1724         kregs = (struct pt_regs *) sp;
1725         sp -= STACK_FRAME_OVERHEAD;
1726         p->thread.ksp = sp;
1727 #ifdef CONFIG_PPC32
1728         p->thread.ksp_limit = (unsigned long)end_of_stack(p);
1729 #endif
1730 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1731         for (i = 0; i < nr_wp_slots(); i++)
1732                 p->thread.ptrace_bps[i] = NULL;
1733 #endif
1734
1735 #ifdef CONFIG_PPC_FPU_REGS
1736         p->thread.fp_save_area = NULL;
1737 #endif
1738 #ifdef CONFIG_ALTIVEC
1739         p->thread.vr_save_area = NULL;
1740 #endif
1741
1742         setup_ksp_vsid(p, sp);
1743
1744 #ifdef CONFIG_PPC64 
1745         if (cpu_has_feature(CPU_FTR_DSCR)) {
1746                 p->thread.dscr_inherit = current->thread.dscr_inherit;
1747                 p->thread.dscr = mfspr(SPRN_DSCR);
1748         }
1749         if (cpu_has_feature(CPU_FTR_HAS_PPR))
1750                 childregs->ppr = DEFAULT_PPR;
1751
1752         p->thread.tidr = 0;
1753 #endif
1754         /*
1755          * Run with the current AMR value of the kernel
1756          */
1757 #ifdef CONFIG_PPC_PKEY
1758         if (mmu_has_feature(MMU_FTR_BOOK3S_KUAP))
1759                 kregs->amr = AMR_KUAP_BLOCKED;
1760
1761         if (mmu_has_feature(MMU_FTR_BOOK3S_KUEP))
1762                 kregs->iamr = AMR_KUEP_BLOCKED;
1763 #endif
1764         kregs->nip = ppc_function_entry(f);
1765         return 0;
1766 }
1767
1768 void preload_new_slb_context(unsigned long start, unsigned long sp);
1769
1770 /*
1771  * Set up a thread for executing a new program
1772  */
1773 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1774 {
1775 #ifdef CONFIG_PPC64
1776         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1777
1778         if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && !radix_enabled())
1779                 preload_new_slb_context(start, sp);
1780 #endif
1781
1782 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1783         /*
1784          * Clear any transactional state, we're exec()ing. The cause is
1785          * not important as there will never be a recheckpoint so it's not
1786          * user visible.
1787          */
1788         if (MSR_TM_SUSPENDED(mfmsr()))
1789                 tm_reclaim_current(0);
1790 #endif
1791
1792         memset(regs->gpr, 0, sizeof(regs->gpr));
1793         regs->ctr = 0;
1794         regs->link = 0;
1795         regs->xer = 0;
1796         regs->ccr = 0;
1797         regs->gpr[1] = sp;
1798
1799         /*
1800          * We have just cleared all the nonvolatile GPRs, so make
1801          * FULL_REGS(regs) return true.  This is necessary to allow
1802          * ptrace to examine the thread immediately after exec.
1803          */
1804         SET_FULL_REGS(regs);
1805
1806 #ifdef CONFIG_PPC32
1807         regs->mq = 0;
1808         regs->nip = start;
1809         regs->msr = MSR_USER;
1810 #else
1811         if (!is_32bit_task()) {
1812                 unsigned long entry;
1813
1814                 if (is_elf2_task()) {
1815                         /* Look ma, no function descriptors! */
1816                         entry = start;
1817
1818                         /*
1819                          * Ulrich says:
1820                          *   The latest iteration of the ABI requires that when
1821                          *   calling a function (at its global entry point),
1822                          *   the caller must ensure r12 holds the entry point
1823                          *   address (so that the function can quickly
1824                          *   establish addressability).
1825                          */
1826                         regs->gpr[12] = start;
1827                         /* Make sure that's restored on entry to userspace. */
1828                         set_thread_flag(TIF_RESTOREALL);
1829                 } else {
1830                         unsigned long toc;
1831
1832                         /* start is a relocated pointer to the function
1833                          * descriptor for the elf _start routine.  The first
1834                          * entry in the function descriptor is the entry
1835                          * address of _start and the second entry is the TOC
1836                          * value we need to use.
1837                          */
1838                         __get_user(entry, (unsigned long __user *)start);
1839                         __get_user(toc, (unsigned long __user *)start+1);
1840
1841                         /* Check whether the e_entry function descriptor entries
1842                          * need to be relocated before we can use them.
1843                          */
1844                         if (load_addr != 0) {
1845                                 entry += load_addr;
1846                                 toc   += load_addr;
1847                         }
1848                         regs->gpr[2] = toc;
1849                 }
1850                 regs->nip = entry;
1851                 regs->msr = MSR_USER64;
1852         } else {
1853                 regs->nip = start;
1854                 regs->gpr[2] = 0;
1855                 regs->msr = MSR_USER32;
1856         }
1857 #endif
1858 #ifdef CONFIG_VSX
1859         current->thread.used_vsr = 0;
1860 #endif
1861         current->thread.load_slb = 0;
1862         current->thread.load_fp = 0;
1863 #ifdef CONFIG_PPC_FPU_REGS
1864         memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1865         current->thread.fp_save_area = NULL;
1866 #endif
1867 #ifdef CONFIG_ALTIVEC
1868         memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1869         current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1870         current->thread.vr_save_area = NULL;
1871         current->thread.vrsave = 0;
1872         current->thread.used_vr = 0;
1873         current->thread.load_vec = 0;
1874 #endif /* CONFIG_ALTIVEC */
1875 #ifdef CONFIG_SPE
1876         memset(current->thread.evr, 0, sizeof(current->thread.evr));
1877         current->thread.acc = 0;
1878         current->thread.spefscr = 0;
1879         current->thread.used_spe = 0;
1880 #endif /* CONFIG_SPE */
1881 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1882         current->thread.tm_tfhar = 0;
1883         current->thread.tm_texasr = 0;
1884         current->thread.tm_tfiar = 0;
1885         current->thread.load_tm = 0;
1886 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1887
1888 }
1889 EXPORT_SYMBOL(start_thread);
1890
1891 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1892                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
1893
1894 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1895 {
1896         struct pt_regs *regs = tsk->thread.regs;
1897
1898         /* This is a bit hairy.  If we are an SPE enabled  processor
1899          * (have embedded fp) we store the IEEE exception enable flags in
1900          * fpexc_mode.  fpexc_mode is also used for setting FP exception
1901          * mode (asyn, precise, disabled) for 'Classic' FP. */
1902         if (val & PR_FP_EXC_SW_ENABLE) {
1903                 if (cpu_has_feature(CPU_FTR_SPE)) {
1904                         /*
1905                          * When the sticky exception bits are set
1906                          * directly by userspace, it must call prctl
1907                          * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1908                          * in the existing prctl settings) or
1909                          * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1910                          * the bits being set).  <fenv.h> functions
1911                          * saving and restoring the whole
1912                          * floating-point environment need to do so
1913                          * anyway to restore the prctl settings from
1914                          * the saved environment.
1915                          */
1916 #ifdef CONFIG_SPE
1917                         tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1918                         tsk->thread.fpexc_mode = val &
1919                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1920 #endif
1921                         return 0;
1922                 } else {
1923                         return -EINVAL;
1924                 }
1925         }
1926
1927         /* on a CONFIG_SPE this does not hurt us.  The bits that
1928          * __pack_fe01 use do not overlap with bits used for
1929          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1930          * on CONFIG_SPE implementations are reserved so writing to
1931          * them does not change anything */
1932         if (val > PR_FP_EXC_PRECISE)
1933                 return -EINVAL;
1934         tsk->thread.fpexc_mode = __pack_fe01(val);
1935         if (regs != NULL && (regs->msr & MSR_FP) != 0)
1936                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1937                         | tsk->thread.fpexc_mode;
1938         return 0;
1939 }
1940
1941 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1942 {
1943         unsigned int val = 0;
1944
1945         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE) {
1946                 if (cpu_has_feature(CPU_FTR_SPE)) {
1947                         /*
1948                          * When the sticky exception bits are set
1949                          * directly by userspace, it must call prctl
1950                          * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1951                          * in the existing prctl settings) or
1952                          * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1953                          * the bits being set).  <fenv.h> functions
1954                          * saving and restoring the whole
1955                          * floating-point environment need to do so
1956                          * anyway to restore the prctl settings from
1957                          * the saved environment.
1958                          */
1959 #ifdef CONFIG_SPE
1960                         tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1961                         val = tsk->thread.fpexc_mode;
1962 #endif
1963                 } else
1964                         return -EINVAL;
1965         } else {
1966                 val = __unpack_fe01(tsk->thread.fpexc_mode);
1967         }
1968         return put_user(val, (unsigned int __user *) adr);
1969 }
1970
1971 int set_endian(struct task_struct *tsk, unsigned int val)
1972 {
1973         struct pt_regs *regs = tsk->thread.regs;
1974
1975         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1976             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1977                 return -EINVAL;
1978
1979         if (regs == NULL)
1980                 return -EINVAL;
1981
1982         if (val == PR_ENDIAN_BIG)
1983                 regs->msr &= ~MSR_LE;
1984         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1985                 regs->msr |= MSR_LE;
1986         else
1987                 return -EINVAL;
1988
1989         return 0;
1990 }
1991
1992 int get_endian(struct task_struct *tsk, unsigned long adr)
1993 {
1994         struct pt_regs *regs = tsk->thread.regs;
1995         unsigned int val;
1996
1997         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1998             !cpu_has_feature(CPU_FTR_REAL_LE))
1999                 return -EINVAL;
2000
2001         if (regs == NULL)
2002                 return -EINVAL;
2003
2004         if (regs->msr & MSR_LE) {
2005                 if (cpu_has_feature(CPU_FTR_REAL_LE))
2006                         val = PR_ENDIAN_LITTLE;
2007                 else
2008                         val = PR_ENDIAN_PPC_LITTLE;
2009         } else
2010                 val = PR_ENDIAN_BIG;
2011
2012         return put_user(val, (unsigned int __user *)adr);
2013 }
2014
2015 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
2016 {
2017         tsk->thread.align_ctl = val;
2018         return 0;
2019 }
2020
2021 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
2022 {
2023         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
2024 }
2025
2026 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
2027                                   unsigned long nbytes)
2028 {
2029         unsigned long stack_page;
2030         unsigned long cpu = task_cpu(p);
2031
2032         stack_page = (unsigned long)hardirq_ctx[cpu];
2033         if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2034                 return 1;
2035
2036         stack_page = (unsigned long)softirq_ctx[cpu];
2037         if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2038                 return 1;
2039
2040         return 0;
2041 }
2042
2043 static inline int valid_emergency_stack(unsigned long sp, struct task_struct *p,
2044                                         unsigned long nbytes)
2045 {
2046 #ifdef CONFIG_PPC64
2047         unsigned long stack_page;
2048         unsigned long cpu = task_cpu(p);
2049
2050         stack_page = (unsigned long)paca_ptrs[cpu]->emergency_sp - THREAD_SIZE;
2051         if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2052                 return 1;
2053
2054 # ifdef CONFIG_PPC_BOOK3S_64
2055         stack_page = (unsigned long)paca_ptrs[cpu]->nmi_emergency_sp - THREAD_SIZE;
2056         if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2057                 return 1;
2058
2059         stack_page = (unsigned long)paca_ptrs[cpu]->mc_emergency_sp - THREAD_SIZE;
2060         if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2061                 return 1;
2062 # endif
2063 #endif
2064
2065         return 0;
2066 }
2067
2068
2069 int validate_sp(unsigned long sp, struct task_struct *p,
2070                        unsigned long nbytes)
2071 {
2072         unsigned long stack_page = (unsigned long)task_stack_page(p);
2073
2074         if (sp < THREAD_SIZE)
2075                 return 0;
2076
2077         if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2078                 return 1;
2079
2080         if (valid_irq_stack(sp, p, nbytes))
2081                 return 1;
2082
2083         return valid_emergency_stack(sp, p, nbytes);
2084 }
2085
2086 EXPORT_SYMBOL(validate_sp);
2087
2088 static unsigned long __get_wchan(struct task_struct *p)
2089 {
2090         unsigned long ip, sp;
2091         int count = 0;
2092
2093         if (!p || p == current || p->state == TASK_RUNNING)
2094                 return 0;
2095
2096         sp = p->thread.ksp;
2097         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2098                 return 0;
2099
2100         do {
2101                 sp = *(unsigned long *)sp;
2102                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2103                     p->state == TASK_RUNNING)
2104                         return 0;
2105                 if (count > 0) {
2106                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2107                         if (!in_sched_functions(ip))
2108                                 return ip;
2109                 }
2110         } while (count++ < 16);
2111         return 0;
2112 }
2113
2114 unsigned long get_wchan(struct task_struct *p)
2115 {
2116         unsigned long ret;
2117
2118         if (!try_get_task_stack(p))
2119                 return 0;
2120
2121         ret = __get_wchan(p);
2122
2123         put_task_stack(p);
2124
2125         return ret;
2126 }
2127
2128 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2129
2130 void show_stack(struct task_struct *tsk, unsigned long *stack,
2131                 const char *loglvl)
2132 {
2133         unsigned long sp, ip, lr, newsp;
2134         int count = 0;
2135         int firstframe = 1;
2136         unsigned long ret_addr;
2137         int ftrace_idx = 0;
2138
2139         if (tsk == NULL)
2140                 tsk = current;
2141
2142         if (!try_get_task_stack(tsk))
2143                 return;
2144
2145         sp = (unsigned long) stack;
2146         if (sp == 0) {
2147                 if (tsk == current)
2148                         sp = current_stack_frame();
2149                 else
2150                         sp = tsk->thread.ksp;
2151         }
2152
2153         lr = 0;
2154         printk("%sCall Trace:\n", loglvl);
2155         do {
2156                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2157                         break;
2158
2159                 stack = (unsigned long *) sp;
2160                 newsp = stack[0];
2161                 ip = stack[STACK_FRAME_LR_SAVE];
2162                 if (!firstframe || ip != lr) {
2163                         printk("%s["REG"] ["REG"] %pS",
2164                                 loglvl, sp, ip, (void *)ip);
2165                         ret_addr = ftrace_graph_ret_addr(current,
2166                                                 &ftrace_idx, ip, stack);
2167                         if (ret_addr != ip)
2168                                 pr_cont(" (%pS)", (void *)ret_addr);
2169                         if (firstframe)
2170                                 pr_cont(" (unreliable)");
2171                         pr_cont("\n");
2172                 }
2173                 firstframe = 0;
2174
2175                 /*
2176                  * See if this is an exception frame.
2177                  * We look for the "regshere" marker in the current frame.
2178                  */
2179                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
2180                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2181                         struct pt_regs *regs = (struct pt_regs *)
2182                                 (sp + STACK_FRAME_OVERHEAD);
2183
2184                         lr = regs->link;
2185                         printk("%s--- interrupt: %lx at %pS\n",
2186                                loglvl, regs->trap, (void *)regs->nip);
2187                         __show_regs(regs);
2188                         printk("%s--- interrupt: %lx\n",
2189                                loglvl, regs->trap);
2190
2191                         firstframe = 1;
2192                 }
2193
2194                 sp = newsp;
2195         } while (count++ < kstack_depth_to_print);
2196
2197         put_task_stack(tsk);
2198 }
2199
2200 #ifdef CONFIG_PPC64
2201 /* Called with hard IRQs off */
2202 void notrace __ppc64_runlatch_on(void)
2203 {
2204         struct thread_info *ti = current_thread_info();
2205
2206         if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2207                 /*
2208                  * Least significant bit (RUN) is the only writable bit of
2209                  * the CTRL register, so we can avoid mfspr. 2.06 is not the
2210                  * earliest ISA where this is the case, but it's convenient.
2211                  */
2212                 mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2213         } else {
2214                 unsigned long ctrl;
2215
2216                 /*
2217                  * Some architectures (e.g., Cell) have writable fields other
2218                  * than RUN, so do the read-modify-write.
2219                  */
2220                 ctrl = mfspr(SPRN_CTRLF);
2221                 ctrl |= CTRL_RUNLATCH;
2222                 mtspr(SPRN_CTRLT, ctrl);
2223         }
2224
2225         ti->local_flags |= _TLF_RUNLATCH;
2226 }
2227
2228 /* Called with hard IRQs off */
2229 void notrace __ppc64_runlatch_off(void)
2230 {
2231         struct thread_info *ti = current_thread_info();
2232
2233         ti->local_flags &= ~_TLF_RUNLATCH;
2234
2235         if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2236                 mtspr(SPRN_CTRLT, 0);
2237         } else {
2238                 unsigned long ctrl;
2239
2240                 ctrl = mfspr(SPRN_CTRLF);
2241                 ctrl &= ~CTRL_RUNLATCH;
2242                 mtspr(SPRN_CTRLT, ctrl);
2243         }
2244 }
2245 #endif /* CONFIG_PPC64 */
2246
2247 unsigned long arch_align_stack(unsigned long sp)
2248 {
2249         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2250                 sp -= get_random_int() & ~PAGE_MASK;
2251         return sp & ~0xf;
2252 }
2253
2254 static inline unsigned long brk_rnd(void)
2255 {
2256         unsigned long rnd = 0;
2257
2258         /* 8MB for 32bit, 1GB for 64bit */
2259         if (is_32bit_task())
2260                 rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2261         else
2262                 rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2263
2264         return rnd << PAGE_SHIFT;
2265 }
2266
2267 unsigned long arch_randomize_brk(struct mm_struct *mm)
2268 {
2269         unsigned long base = mm->brk;
2270         unsigned long ret;
2271
2272 #ifdef CONFIG_PPC_BOOK3S_64
2273         /*
2274          * If we are using 1TB segments and we are allowed to randomise
2275          * the heap, we can put it above 1TB so it is backed by a 1TB
2276          * segment. Otherwise the heap will be in the bottom 1TB
2277          * which always uses 256MB segments and this may result in a
2278          * performance penalty. We don't need to worry about radix. For
2279          * radix, mmu_highuser_ssize remains unchanged from 256MB.
2280          */
2281         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2282                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2283 #endif
2284
2285         ret = PAGE_ALIGN(base + brk_rnd());
2286
2287         if (ret < mm->brk)
2288                 return mm->brk;
2289
2290         return ret;
2291 }
2292
This page took 0.166559 seconds and 4 git commands to generate.