xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision 82e6fdd6)
1 /*
2  *  Copyright (C) 1995-1996  Gary Thomas (gdt@linuxppc.org)
3  *  Copyright 2007-2010 Freescale Semiconductor, Inc.
4  *
5  *  This program is free software; you can redistribute it and/or
6  *  modify it under the terms of the GNU General Public License
7  *  as published by the Free Software Foundation; either version
8  *  2 of the License, or (at your option) any later version.
9  *
10  *  Modified by Cort Dougan (cort@cs.nmt.edu)
11  *  and Paul Mackerras (paulus@samba.org)
12  */
13 
14 /*
15  * This file handles the architecture-dependent parts of hardware exceptions
16  */
17 
18 #include <linux/errno.h>
19 #include <linux/sched.h>
20 #include <linux/sched/debug.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/pkeys.h>
24 #include <linux/stddef.h>
25 #include <linux/unistd.h>
26 #include <linux/ptrace.h>
27 #include <linux/user.h>
28 #include <linux/interrupt.h>
29 #include <linux/init.h>
30 #include <linux/extable.h>
31 #include <linux/module.h>	/* print_modules */
32 #include <linux/prctl.h>
33 #include <linux/delay.h>
34 #include <linux/kprobes.h>
35 #include <linux/kexec.h>
36 #include <linux/backlight.h>
37 #include <linux/bug.h>
38 #include <linux/kdebug.h>
39 #include <linux/ratelimit.h>
40 #include <linux/context_tracking.h>
41 #include <linux/smp.h>
42 #include <linux/console.h>
43 #include <linux/kmsg_dump.h>
44 
45 #include <asm/emulated_ops.h>
46 #include <asm/pgtable.h>
47 #include <linux/uaccess.h>
48 #include <asm/debugfs.h>
49 #include <asm/io.h>
50 #include <asm/machdep.h>
51 #include <asm/rtas.h>
52 #include <asm/pmc.h>
53 #include <asm/reg.h>
54 #ifdef CONFIG_PMAC_BACKLIGHT
55 #include <asm/backlight.h>
56 #endif
57 #ifdef CONFIG_PPC64
58 #include <asm/firmware.h>
59 #include <asm/processor.h>
60 #include <asm/tm.h>
61 #endif
62 #include <asm/kexec.h>
63 #include <asm/ppc-opcode.h>
64 #include <asm/rio.h>
65 #include <asm/fadump.h>
66 #include <asm/switch_to.h>
67 #include <asm/tm.h>
68 #include <asm/debug.h>
69 #include <asm/asm-prototypes.h>
70 #include <asm/hmi.h>
71 #include <sysdev/fsl_pci.h>
72 #include <asm/kprobes.h>
73 
74 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
75 int (*__debugger)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
79 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
80 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
81 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
82 
83 EXPORT_SYMBOL(__debugger);
84 EXPORT_SYMBOL(__debugger_ipi);
85 EXPORT_SYMBOL(__debugger_bpt);
86 EXPORT_SYMBOL(__debugger_sstep);
87 EXPORT_SYMBOL(__debugger_iabr_match);
88 EXPORT_SYMBOL(__debugger_break_match);
89 EXPORT_SYMBOL(__debugger_fault_handler);
90 #endif
91 
92 /* Transactional Memory trap debug */
93 #ifdef TM_DEBUG_SW
94 #define TM_DEBUG(x...) printk(KERN_INFO x)
95 #else
96 #define TM_DEBUG(x...) do { } while(0)
97 #endif
98 
99 /*
100  * Trap & Exception support
101  */
102 
103 #ifdef CONFIG_PMAC_BACKLIGHT
104 static void pmac_backlight_unblank(void)
105 {
106 	mutex_lock(&pmac_backlight_mutex);
107 	if (pmac_backlight) {
108 		struct backlight_properties *props;
109 
110 		props = &pmac_backlight->props;
111 		props->brightness = props->max_brightness;
112 		props->power = FB_BLANK_UNBLANK;
113 		backlight_update_status(pmac_backlight);
114 	}
115 	mutex_unlock(&pmac_backlight_mutex);
116 }
117 #else
118 static inline void pmac_backlight_unblank(void) { }
119 #endif
120 
121 /*
122  * If oops/die is expected to crash the machine, return true here.
123  *
124  * This should not be expected to be 100% accurate, there may be
125  * notifiers registered or other unexpected conditions that may bring
126  * down the kernel. Or if the current process in the kernel is holding
127  * locks or has other critical state, the kernel may become effectively
128  * unusable anyway.
129  */
130 bool die_will_crash(void)
131 {
132 	if (should_fadump_crash())
133 		return true;
134 	if (kexec_should_crash(current))
135 		return true;
136 	if (in_interrupt() || panic_on_oops ||
137 			!current->pid || is_global_init(current))
138 		return true;
139 
140 	return false;
141 }
142 
143 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
144 static int die_owner = -1;
145 static unsigned int die_nest_count;
146 static int die_counter;
147 
148 extern void panic_flush_kmsg_start(void)
149 {
150 	/*
151 	 * These are mostly taken from kernel/panic.c, but tries to do
152 	 * relatively minimal work. Don't use delay functions (TB may
153 	 * be broken), don't crash dump (need to set a firmware log),
154 	 * don't run notifiers. We do want to get some information to
155 	 * Linux console.
156 	 */
157 	console_verbose();
158 	bust_spinlocks(1);
159 }
160 
161 extern void panic_flush_kmsg_end(void)
162 {
163 	printk_safe_flush_on_panic();
164 	kmsg_dump(KMSG_DUMP_PANIC);
165 	bust_spinlocks(0);
166 	debug_locks_off();
167 	console_flush_on_panic();
168 }
169 
170 static unsigned long oops_begin(struct pt_regs *regs)
171 {
172 	int cpu;
173 	unsigned long flags;
174 
175 	oops_enter();
176 
177 	/* racy, but better than risking deadlock. */
178 	raw_local_irq_save(flags);
179 	cpu = smp_processor_id();
180 	if (!arch_spin_trylock(&die_lock)) {
181 		if (cpu == die_owner)
182 			/* nested oops. should stop eventually */;
183 		else
184 			arch_spin_lock(&die_lock);
185 	}
186 	die_nest_count++;
187 	die_owner = cpu;
188 	console_verbose();
189 	bust_spinlocks(1);
190 	if (machine_is(powermac))
191 		pmac_backlight_unblank();
192 	return flags;
193 }
194 NOKPROBE_SYMBOL(oops_begin);
195 
196 static void oops_end(unsigned long flags, struct pt_regs *regs,
197 			       int signr)
198 {
199 	bust_spinlocks(0);
200 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
201 	die_nest_count--;
202 	oops_exit();
203 	printk("\n");
204 	if (!die_nest_count) {
205 		/* Nest count reaches zero, release the lock. */
206 		die_owner = -1;
207 		arch_spin_unlock(&die_lock);
208 	}
209 	raw_local_irq_restore(flags);
210 
211 	crash_fadump(regs, "die oops");
212 
213 	if (kexec_should_crash(current))
214 		crash_kexec(regs);
215 
216 	if (!signr)
217 		return;
218 
219 	/*
220 	 * While our oops output is serialised by a spinlock, output
221 	 * from panic() called below can race and corrupt it. If we
222 	 * know we are going to panic, delay for 1 second so we have a
223 	 * chance to get clean backtraces from all CPUs that are oopsing.
224 	 */
225 	if (in_interrupt() || panic_on_oops || !current->pid ||
226 	    is_global_init(current)) {
227 		mdelay(MSEC_PER_SEC);
228 	}
229 
230 	if (in_interrupt())
231 		panic("Fatal exception in interrupt");
232 	if (panic_on_oops)
233 		panic("Fatal exception");
234 	do_exit(signr);
235 }
236 NOKPROBE_SYMBOL(oops_end);
237 
238 static int __die(const char *str, struct pt_regs *regs, long err)
239 {
240 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
241 
242 	if (IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN))
243 		printk("LE ");
244 	else
245 		printk("BE ");
246 
247 	if (IS_ENABLED(CONFIG_PREEMPT))
248 		pr_cont("PREEMPT ");
249 
250 	if (IS_ENABLED(CONFIG_SMP))
251 		pr_cont("SMP NR_CPUS=%d ", NR_CPUS);
252 
253 	if (debug_pagealloc_enabled())
254 		pr_cont("DEBUG_PAGEALLOC ");
255 
256 	if (IS_ENABLED(CONFIG_NUMA))
257 		pr_cont("NUMA ");
258 
259 	pr_cont("%s\n", ppc_md.name ? ppc_md.name : "");
260 
261 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
262 		return 1;
263 
264 	print_modules();
265 	show_regs(regs);
266 
267 	return 0;
268 }
269 NOKPROBE_SYMBOL(__die);
270 
271 void die(const char *str, struct pt_regs *regs, long err)
272 {
273 	unsigned long flags;
274 
275 	if (debugger(regs))
276 		return;
277 
278 	flags = oops_begin(regs);
279 	if (__die(str, regs, err))
280 		err = 0;
281 	oops_end(flags, regs, err);
282 }
283 NOKPROBE_SYMBOL(die);
284 
285 void user_single_step_siginfo(struct task_struct *tsk,
286 				struct pt_regs *regs, siginfo_t *info)
287 {
288 	memset(info, 0, sizeof(*info));
289 	info->si_signo = SIGTRAP;
290 	info->si_code = TRAP_TRACE;
291 	info->si_addr = (void __user *)regs->nip;
292 }
293 
294 
295 void _exception_pkey(int signr, struct pt_regs *regs, int code,
296 		unsigned long addr, int key)
297 {
298 	siginfo_t info;
299 	const char fmt32[] = KERN_INFO "%s[%d]: unhandled signal %d " \
300 			"at %08lx nip %08lx lr %08lx code %x\n";
301 	const char fmt64[] = KERN_INFO "%s[%d]: unhandled signal %d " \
302 			"at %016lx nip %016lx lr %016lx code %x\n";
303 
304 	if (!user_mode(regs)) {
305 		die("Exception in kernel mode", regs, signr);
306 		return;
307 	}
308 
309 	if (show_unhandled_signals && unhandled_signal(current, signr)) {
310 		printk_ratelimited(regs->msr & MSR_64BIT ? fmt64 : fmt32,
311 				   current->comm, current->pid, signr,
312 				   addr, regs->nip, regs->link, code);
313 	}
314 
315 	if (arch_irqs_disabled() && !arch_irq_disabled_regs(regs))
316 		local_irq_enable();
317 
318 	current->thread.trap_nr = code;
319 
320 	/*
321 	 * Save all the pkey registers AMR/IAMR/UAMOR. Eg: Core dumps need
322 	 * to capture the content, if the task gets killed.
323 	 */
324 	thread_pkey_regs_save(&current->thread);
325 
326 	memset(&info, 0, sizeof(info));
327 	info.si_signo = signr;
328 	info.si_code = code;
329 	info.si_addr = (void __user *) addr;
330 	info.si_pkey = key;
331 
332 	force_sig_info(signr, &info, current);
333 }
334 
335 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
336 {
337 	_exception_pkey(signr, regs, code, addr, 0);
338 }
339 
340 void system_reset_exception(struct pt_regs *regs)
341 {
342 	/*
343 	 * Avoid crashes in case of nested NMI exceptions. Recoverability
344 	 * is determined by RI and in_nmi
345 	 */
346 	bool nested = in_nmi();
347 	if (!nested)
348 		nmi_enter();
349 
350 	__this_cpu_inc(irq_stat.sreset_irqs);
351 
352 	/* See if any machine dependent calls */
353 	if (ppc_md.system_reset_exception) {
354 		if (ppc_md.system_reset_exception(regs))
355 			goto out;
356 	}
357 
358 	if (debugger(regs))
359 		goto out;
360 
361 	/*
362 	 * A system reset is a request to dump, so we always send
363 	 * it through the crashdump code (if fadump or kdump are
364 	 * registered).
365 	 */
366 	crash_fadump(regs, "System Reset");
367 
368 	crash_kexec(regs);
369 
370 	/*
371 	 * We aren't the primary crash CPU. We need to send it
372 	 * to a holding pattern to avoid it ending up in the panic
373 	 * code.
374 	 */
375 	crash_kexec_secondary(regs);
376 
377 	/*
378 	 * No debugger or crash dump registered, print logs then
379 	 * panic.
380 	 */
381 	die("System Reset", regs, SIGABRT);
382 
383 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
384 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
385 	nmi_panic(regs, "System Reset");
386 
387 out:
388 #ifdef CONFIG_PPC_BOOK3S_64
389 	BUG_ON(get_paca()->in_nmi == 0);
390 	if (get_paca()->in_nmi > 1)
391 		nmi_panic(regs, "Unrecoverable nested System Reset");
392 #endif
393 	/* Must die if the interrupt is not recoverable */
394 	if (!(regs->msr & MSR_RI))
395 		nmi_panic(regs, "Unrecoverable System Reset");
396 
397 	if (!nested)
398 		nmi_exit();
399 
400 	/* What should we do here? We could issue a shutdown or hard reset. */
401 }
402 
403 /*
404  * I/O accesses can cause machine checks on powermacs.
405  * Check if the NIP corresponds to the address of a sync
406  * instruction for which there is an entry in the exception
407  * table.
408  * Note that the 601 only takes a machine check on TEA
409  * (transfer error ack) signal assertion, and does not
410  * set any of the top 16 bits of SRR1.
411  *  -- paulus.
412  */
413 static inline int check_io_access(struct pt_regs *regs)
414 {
415 #ifdef CONFIG_PPC32
416 	unsigned long msr = regs->msr;
417 	const struct exception_table_entry *entry;
418 	unsigned int *nip = (unsigned int *)regs->nip;
419 
420 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
421 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
422 		/*
423 		 * Check that it's a sync instruction, or somewhere
424 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
425 		 * As the address is in the exception table
426 		 * we should be able to read the instr there.
427 		 * For the debug message, we look at the preceding
428 		 * load or store.
429 		 */
430 		if (*nip == PPC_INST_NOP)
431 			nip -= 2;
432 		else if (*nip == PPC_INST_ISYNC)
433 			--nip;
434 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
435 			unsigned int rb;
436 
437 			--nip;
438 			rb = (*nip >> 11) & 0x1f;
439 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
440 			       (*nip & 0x100)? "OUT to": "IN from",
441 			       regs->gpr[rb] - _IO_BASE, nip);
442 			regs->msr |= MSR_RI;
443 			regs->nip = extable_fixup(entry);
444 			return 1;
445 		}
446 	}
447 #endif /* CONFIG_PPC32 */
448 	return 0;
449 }
450 
451 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
452 /* On 4xx, the reason for the machine check or program exception
453    is in the ESR. */
454 #define get_reason(regs)	((regs)->dsisr)
455 #define REASON_FP		ESR_FP
456 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
457 #define REASON_PRIVILEGED	ESR_PPR
458 #define REASON_TRAP		ESR_PTR
459 
460 /* single-step stuff */
461 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
462 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
463 
464 #else
465 /* On non-4xx, the reason for the machine check or program
466    exception is in the MSR. */
467 #define get_reason(regs)	((regs)->msr)
468 #define REASON_TM		SRR1_PROGTM
469 #define REASON_FP		SRR1_PROGFPE
470 #define REASON_ILLEGAL		SRR1_PROGILL
471 #define REASON_PRIVILEGED	SRR1_PROGPRIV
472 #define REASON_TRAP		SRR1_PROGTRAP
473 
474 #define single_stepping(regs)	((regs)->msr & MSR_SE)
475 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
476 #endif
477 
478 #if defined(CONFIG_E500)
479 int machine_check_e500mc(struct pt_regs *regs)
480 {
481 	unsigned long mcsr = mfspr(SPRN_MCSR);
482 	unsigned long pvr = mfspr(SPRN_PVR);
483 	unsigned long reason = mcsr;
484 	int recoverable = 1;
485 
486 	if (reason & MCSR_LD) {
487 		recoverable = fsl_rio_mcheck_exception(regs);
488 		if (recoverable == 1)
489 			goto silent_out;
490 	}
491 
492 	printk("Machine check in kernel mode.\n");
493 	printk("Caused by (from MCSR=%lx): ", reason);
494 
495 	if (reason & MCSR_MCP)
496 		printk("Machine Check Signal\n");
497 
498 	if (reason & MCSR_ICPERR) {
499 		printk("Instruction Cache Parity Error\n");
500 
501 		/*
502 		 * This is recoverable by invalidating the i-cache.
503 		 */
504 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
505 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
506 			;
507 
508 		/*
509 		 * This will generally be accompanied by an instruction
510 		 * fetch error report -- only treat MCSR_IF as fatal
511 		 * if it wasn't due to an L1 parity error.
512 		 */
513 		reason &= ~MCSR_IF;
514 	}
515 
516 	if (reason & MCSR_DCPERR_MC) {
517 		printk("Data Cache Parity Error\n");
518 
519 		/*
520 		 * In write shadow mode we auto-recover from the error, but it
521 		 * may still get logged and cause a machine check.  We should
522 		 * only treat the non-write shadow case as non-recoverable.
523 		 */
524 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
525 		 * is not implemented but L1 data cache always runs in write
526 		 * shadow mode. Hence on data cache parity errors HW will
527 		 * automatically invalidate the L1 Data Cache.
528 		 */
529 		if (PVR_VER(pvr) != PVR_VER_E6500) {
530 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
531 				recoverable = 0;
532 		}
533 	}
534 
535 	if (reason & MCSR_L2MMU_MHIT) {
536 		printk("Hit on multiple TLB entries\n");
537 		recoverable = 0;
538 	}
539 
540 	if (reason & MCSR_NMI)
541 		printk("Non-maskable interrupt\n");
542 
543 	if (reason & MCSR_IF) {
544 		printk("Instruction Fetch Error Report\n");
545 		recoverable = 0;
546 	}
547 
548 	if (reason & MCSR_LD) {
549 		printk("Load Error Report\n");
550 		recoverable = 0;
551 	}
552 
553 	if (reason & MCSR_ST) {
554 		printk("Store Error Report\n");
555 		recoverable = 0;
556 	}
557 
558 	if (reason & MCSR_LDG) {
559 		printk("Guarded Load Error Report\n");
560 		recoverable = 0;
561 	}
562 
563 	if (reason & MCSR_TLBSYNC)
564 		printk("Simultaneous tlbsync operations\n");
565 
566 	if (reason & MCSR_BSL2_ERR) {
567 		printk("Level 2 Cache Error\n");
568 		recoverable = 0;
569 	}
570 
571 	if (reason & MCSR_MAV) {
572 		u64 addr;
573 
574 		addr = mfspr(SPRN_MCAR);
575 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
576 
577 		printk("Machine Check %s Address: %#llx\n",
578 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
579 	}
580 
581 silent_out:
582 	mtspr(SPRN_MCSR, mcsr);
583 	return mfspr(SPRN_MCSR) == 0 && recoverable;
584 }
585 
586 int machine_check_e500(struct pt_regs *regs)
587 {
588 	unsigned long reason = mfspr(SPRN_MCSR);
589 
590 	if (reason & MCSR_BUS_RBERR) {
591 		if (fsl_rio_mcheck_exception(regs))
592 			return 1;
593 		if (fsl_pci_mcheck_exception(regs))
594 			return 1;
595 	}
596 
597 	printk("Machine check in kernel mode.\n");
598 	printk("Caused by (from MCSR=%lx): ", reason);
599 
600 	if (reason & MCSR_MCP)
601 		printk("Machine Check Signal\n");
602 	if (reason & MCSR_ICPERR)
603 		printk("Instruction Cache Parity Error\n");
604 	if (reason & MCSR_DCP_PERR)
605 		printk("Data Cache Push Parity Error\n");
606 	if (reason & MCSR_DCPERR)
607 		printk("Data Cache Parity Error\n");
608 	if (reason & MCSR_BUS_IAERR)
609 		printk("Bus - Instruction Address Error\n");
610 	if (reason & MCSR_BUS_RAERR)
611 		printk("Bus - Read Address Error\n");
612 	if (reason & MCSR_BUS_WAERR)
613 		printk("Bus - Write Address Error\n");
614 	if (reason & MCSR_BUS_IBERR)
615 		printk("Bus - Instruction Data Error\n");
616 	if (reason & MCSR_BUS_RBERR)
617 		printk("Bus - Read Data Bus Error\n");
618 	if (reason & MCSR_BUS_WBERR)
619 		printk("Bus - Write Data Bus Error\n");
620 	if (reason & MCSR_BUS_IPERR)
621 		printk("Bus - Instruction Parity Error\n");
622 	if (reason & MCSR_BUS_RPERR)
623 		printk("Bus - Read Parity Error\n");
624 
625 	return 0;
626 }
627 
628 int machine_check_generic(struct pt_regs *regs)
629 {
630 	return 0;
631 }
632 #elif defined(CONFIG_E200)
633 int machine_check_e200(struct pt_regs *regs)
634 {
635 	unsigned long reason = mfspr(SPRN_MCSR);
636 
637 	printk("Machine check in kernel mode.\n");
638 	printk("Caused by (from MCSR=%lx): ", reason);
639 
640 	if (reason & MCSR_MCP)
641 		printk("Machine Check Signal\n");
642 	if (reason & MCSR_CP_PERR)
643 		printk("Cache Push Parity Error\n");
644 	if (reason & MCSR_CPERR)
645 		printk("Cache Parity Error\n");
646 	if (reason & MCSR_EXCP_ERR)
647 		printk("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
648 	if (reason & MCSR_BUS_IRERR)
649 		printk("Bus - Read Bus Error on instruction fetch\n");
650 	if (reason & MCSR_BUS_DRERR)
651 		printk("Bus - Read Bus Error on data load\n");
652 	if (reason & MCSR_BUS_WRERR)
653 		printk("Bus - Write Bus Error on buffered store or cache line push\n");
654 
655 	return 0;
656 }
657 #elif defined(CONFIG_PPC32)
658 int machine_check_generic(struct pt_regs *regs)
659 {
660 	unsigned long reason = regs->msr;
661 
662 	printk("Machine check in kernel mode.\n");
663 	printk("Caused by (from SRR1=%lx): ", reason);
664 	switch (reason & 0x601F0000) {
665 	case 0x80000:
666 		printk("Machine check signal\n");
667 		break;
668 	case 0:		/* for 601 */
669 	case 0x40000:
670 	case 0x140000:	/* 7450 MSS error and TEA */
671 		printk("Transfer error ack signal\n");
672 		break;
673 	case 0x20000:
674 		printk("Data parity error signal\n");
675 		break;
676 	case 0x10000:
677 		printk("Address parity error signal\n");
678 		break;
679 	case 0x20000000:
680 		printk("L1 Data Cache error\n");
681 		break;
682 	case 0x40000000:
683 		printk("L1 Instruction Cache error\n");
684 		break;
685 	case 0x00100000:
686 		printk("L2 data cache parity error\n");
687 		break;
688 	default:
689 		printk("Unknown values in msr\n");
690 	}
691 	return 0;
692 }
693 #endif /* everything else */
694 
695 void machine_check_exception(struct pt_regs *regs)
696 {
697 	int recover = 0;
698 	bool nested = in_nmi();
699 	if (!nested)
700 		nmi_enter();
701 
702 	/* 64s accounts the mce in machine_check_early when in HVMODE */
703 	if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64) || !cpu_has_feature(CPU_FTR_HVMODE))
704 		__this_cpu_inc(irq_stat.mce_exceptions);
705 
706 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
707 
708 	/* See if any machine dependent calls. In theory, we would want
709 	 * to call the CPU first, and call the ppc_md. one if the CPU
710 	 * one returns a positive number. However there is existing code
711 	 * that assumes the board gets a first chance, so let's keep it
712 	 * that way for now and fix things later. --BenH.
713 	 */
714 	if (ppc_md.machine_check_exception)
715 		recover = ppc_md.machine_check_exception(regs);
716 	else if (cur_cpu_spec->machine_check)
717 		recover = cur_cpu_spec->machine_check(regs);
718 
719 	if (recover > 0)
720 		goto bail;
721 
722 	if (debugger_fault_handler(regs))
723 		goto bail;
724 
725 	if (check_io_access(regs))
726 		goto bail;
727 
728 	die("Machine check", regs, SIGBUS);
729 
730 	/* Must die if the interrupt is not recoverable */
731 	if (!(regs->msr & MSR_RI))
732 		nmi_panic(regs, "Unrecoverable Machine check");
733 
734 bail:
735 	if (!nested)
736 		nmi_exit();
737 }
738 
739 void SMIException(struct pt_regs *regs)
740 {
741 	die("System Management Interrupt", regs, SIGABRT);
742 }
743 
744 #ifdef CONFIG_VSX
745 static void p9_hmi_special_emu(struct pt_regs *regs)
746 {
747 	unsigned int ra, rb, t, i, sel, instr, rc;
748 	const void __user *addr;
749 	u8 vbuf[16], *vdst;
750 	unsigned long ea, msr, msr_mask;
751 	bool swap;
752 
753 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
754 		return;
755 
756 	/*
757 	 * lxvb16x	opcode: 0x7c0006d8
758 	 * lxvd2x	opcode: 0x7c000698
759 	 * lxvh8x	opcode: 0x7c000658
760 	 * lxvw4x	opcode: 0x7c000618
761 	 */
762 	if ((instr & 0xfc00073e) != 0x7c000618) {
763 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
764 			 " instr=%08x\n",
765 			 smp_processor_id(), current->comm, current->pid,
766 			 regs->nip, instr);
767 		return;
768 	}
769 
770 	/* Grab vector registers into the task struct */
771 	msr = regs->msr; /* Grab msr before we flush the bits */
772 	flush_vsx_to_thread(current);
773 	enable_kernel_altivec();
774 
775 	/*
776 	 * Is userspace running with a different endian (this is rare but
777 	 * not impossible)
778 	 */
779 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
780 
781 	/* Decode the instruction */
782 	ra = (instr >> 16) & 0x1f;
783 	rb = (instr >> 11) & 0x1f;
784 	t = (instr >> 21) & 0x1f;
785 	if (instr & 1)
786 		vdst = (u8 *)&current->thread.vr_state.vr[t];
787 	else
788 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
789 
790 	/* Grab the vector address */
791 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
792 	if (is_32bit_task())
793 		ea &= 0xfffffffful;
794 	addr = (__force const void __user *)ea;
795 
796 	/* Check it */
797 	if (!access_ok(VERIFY_READ, addr, 16)) {
798 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
799 			 " instr=%08x addr=%016lx\n",
800 			 smp_processor_id(), current->comm, current->pid,
801 			 regs->nip, instr, (unsigned long)addr);
802 		return;
803 	}
804 
805 	/* Read the vector */
806 	rc = 0;
807 	if ((unsigned long)addr & 0xfUL)
808 		/* unaligned case */
809 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
810 	else
811 		__get_user_atomic_128_aligned(vbuf, addr, rc);
812 	if (rc) {
813 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
814 			 " instr=%08x addr=%016lx\n",
815 			 smp_processor_id(), current->comm, current->pid,
816 			 regs->nip, instr, (unsigned long)addr);
817 		return;
818 	}
819 
820 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
821 		 " instr=%08x addr=%016lx\n",
822 		 smp_processor_id(), current->comm, current->pid, regs->nip,
823 		 instr, (unsigned long) addr);
824 
825 	/* Grab instruction "selector" */
826 	sel = (instr >> 6) & 3;
827 
828 	/*
829 	 * Check to make sure the facility is actually enabled. This
830 	 * could happen if we get a false positive hit.
831 	 *
832 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
833 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
834 	 */
835 	msr_mask = MSR_VSX;
836 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
837 		msr_mask = MSR_VEC;
838 	if (!(msr & msr_mask)) {
839 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
840 			 " instr=%08x msr:%016lx\n",
841 			 smp_processor_id(), current->comm, current->pid,
842 			 regs->nip, instr, msr);
843 		return;
844 	}
845 
846 	/* Do logging here before we modify sel based on endian */
847 	switch (sel) {
848 	case 0:	/* lxvw4x */
849 		PPC_WARN_EMULATED(lxvw4x, regs);
850 		break;
851 	case 1: /* lxvh8x */
852 		PPC_WARN_EMULATED(lxvh8x, regs);
853 		break;
854 	case 2: /* lxvd2x */
855 		PPC_WARN_EMULATED(lxvd2x, regs);
856 		break;
857 	case 3: /* lxvb16x */
858 		PPC_WARN_EMULATED(lxvb16x, regs);
859 		break;
860 	}
861 
862 #ifdef __LITTLE_ENDIAN__
863 	/*
864 	 * An LE kernel stores the vector in the task struct as an LE
865 	 * byte array (effectively swapping both the components and
866 	 * the content of the components). Those instructions expect
867 	 * the components to remain in ascending address order, so we
868 	 * swap them back.
869 	 *
870 	 * If we are running a BE user space, the expectation is that
871 	 * of a simple memcpy, so forcing the emulation to look like
872 	 * a lxvb16x should do the trick.
873 	 */
874 	if (swap)
875 		sel = 3;
876 
877 	switch (sel) {
878 	case 0:	/* lxvw4x */
879 		for (i = 0; i < 4; i++)
880 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
881 		break;
882 	case 1: /* lxvh8x */
883 		for (i = 0; i < 8; i++)
884 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
885 		break;
886 	case 2: /* lxvd2x */
887 		for (i = 0; i < 2; i++)
888 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
889 		break;
890 	case 3: /* lxvb16x */
891 		for (i = 0; i < 16; i++)
892 			vdst[i] = vbuf[15-i];
893 		break;
894 	}
895 #else /* __LITTLE_ENDIAN__ */
896 	/* On a big endian kernel, a BE userspace only needs a memcpy */
897 	if (!swap)
898 		sel = 3;
899 
900 	/* Otherwise, we need to swap the content of the components */
901 	switch (sel) {
902 	case 0:	/* lxvw4x */
903 		for (i = 0; i < 4; i++)
904 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
905 		break;
906 	case 1: /* lxvh8x */
907 		for (i = 0; i < 8; i++)
908 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
909 		break;
910 	case 2: /* lxvd2x */
911 		for (i = 0; i < 2; i++)
912 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
913 		break;
914 	case 3: /* lxvb16x */
915 		memcpy(vdst, vbuf, 16);
916 		break;
917 	}
918 #endif /* !__LITTLE_ENDIAN__ */
919 
920 	/* Go to next instruction */
921 	regs->nip += 4;
922 }
923 #endif /* CONFIG_VSX */
924 
925 void handle_hmi_exception(struct pt_regs *regs)
926 {
927 	struct pt_regs *old_regs;
928 
929 	old_regs = set_irq_regs(regs);
930 	irq_enter();
931 
932 #ifdef CONFIG_VSX
933 	/* Real mode flagged P9 special emu is needed */
934 	if (local_paca->hmi_p9_special_emu) {
935 		local_paca->hmi_p9_special_emu = 0;
936 
937 		/*
938 		 * We don't want to take page faults while doing the
939 		 * emulation, we just replay the instruction if necessary.
940 		 */
941 		pagefault_disable();
942 		p9_hmi_special_emu(regs);
943 		pagefault_enable();
944 	}
945 #endif /* CONFIG_VSX */
946 
947 	if (ppc_md.handle_hmi_exception)
948 		ppc_md.handle_hmi_exception(regs);
949 
950 	irq_exit();
951 	set_irq_regs(old_regs);
952 }
953 
954 void unknown_exception(struct pt_regs *regs)
955 {
956 	enum ctx_state prev_state = exception_enter();
957 
958 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
959 	       regs->nip, regs->msr, regs->trap);
960 
961 	_exception(SIGTRAP, regs, TRAP_FIXME, 0);
962 
963 	exception_exit(prev_state);
964 }
965 
966 void instruction_breakpoint_exception(struct pt_regs *regs)
967 {
968 	enum ctx_state prev_state = exception_enter();
969 
970 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
971 					5, SIGTRAP) == NOTIFY_STOP)
972 		goto bail;
973 	if (debugger_iabr_match(regs))
974 		goto bail;
975 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
976 
977 bail:
978 	exception_exit(prev_state);
979 }
980 
981 void RunModeException(struct pt_regs *regs)
982 {
983 	_exception(SIGTRAP, regs, TRAP_FIXME, 0);
984 }
985 
986 void single_step_exception(struct pt_regs *regs)
987 {
988 	enum ctx_state prev_state = exception_enter();
989 
990 	clear_single_step(regs);
991 
992 	if (kprobe_post_handler(regs))
993 		return;
994 
995 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
996 					5, SIGTRAP) == NOTIFY_STOP)
997 		goto bail;
998 	if (debugger_sstep(regs))
999 		goto bail;
1000 
1001 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1002 
1003 bail:
1004 	exception_exit(prev_state);
1005 }
1006 NOKPROBE_SYMBOL(single_step_exception);
1007 
1008 /*
1009  * After we have successfully emulated an instruction, we have to
1010  * check if the instruction was being single-stepped, and if so,
1011  * pretend we got a single-step exception.  This was pointed out
1012  * by Kumar Gala.  -- paulus
1013  */
1014 static void emulate_single_step(struct pt_regs *regs)
1015 {
1016 	if (single_stepping(regs))
1017 		single_step_exception(regs);
1018 }
1019 
1020 static inline int __parse_fpscr(unsigned long fpscr)
1021 {
1022 	int ret = FPE_FIXME;
1023 
1024 	/* Invalid operation */
1025 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1026 		ret = FPE_FLTINV;
1027 
1028 	/* Overflow */
1029 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1030 		ret = FPE_FLTOVF;
1031 
1032 	/* Underflow */
1033 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1034 		ret = FPE_FLTUND;
1035 
1036 	/* Divide by zero */
1037 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1038 		ret = FPE_FLTDIV;
1039 
1040 	/* Inexact result */
1041 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1042 		ret = FPE_FLTRES;
1043 
1044 	return ret;
1045 }
1046 
1047 static void parse_fpe(struct pt_regs *regs)
1048 {
1049 	int code = 0;
1050 
1051 	flush_fp_to_thread(current);
1052 
1053 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1054 
1055 	_exception(SIGFPE, regs, code, regs->nip);
1056 }
1057 
1058 /*
1059  * Illegal instruction emulation support.  Originally written to
1060  * provide the PVR to user applications using the mfspr rd, PVR.
1061  * Return non-zero if we can't emulate, or -EFAULT if the associated
1062  * memory access caused an access fault.  Return zero on success.
1063  *
1064  * There are a couple of ways to do this, either "decode" the instruction
1065  * or directly match lots of bits.  In this case, matching lots of
1066  * bits is faster and easier.
1067  *
1068  */
1069 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1070 {
1071 	u8 rT = (instword >> 21) & 0x1f;
1072 	u8 rA = (instword >> 16) & 0x1f;
1073 	u8 NB_RB = (instword >> 11) & 0x1f;
1074 	u32 num_bytes;
1075 	unsigned long EA;
1076 	int pos = 0;
1077 
1078 	/* Early out if we are an invalid form of lswx */
1079 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1080 		if ((rT == rA) || (rT == NB_RB))
1081 			return -EINVAL;
1082 
1083 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1084 
1085 	switch (instword & PPC_INST_STRING_MASK) {
1086 		case PPC_INST_LSWX:
1087 		case PPC_INST_STSWX:
1088 			EA += NB_RB;
1089 			num_bytes = regs->xer & 0x7f;
1090 			break;
1091 		case PPC_INST_LSWI:
1092 		case PPC_INST_STSWI:
1093 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1094 			break;
1095 		default:
1096 			return -EINVAL;
1097 	}
1098 
1099 	while (num_bytes != 0)
1100 	{
1101 		u8 val;
1102 		u32 shift = 8 * (3 - (pos & 0x3));
1103 
1104 		/* if process is 32-bit, clear upper 32 bits of EA */
1105 		if ((regs->msr & MSR_64BIT) == 0)
1106 			EA &= 0xFFFFFFFF;
1107 
1108 		switch ((instword & PPC_INST_STRING_MASK)) {
1109 			case PPC_INST_LSWX:
1110 			case PPC_INST_LSWI:
1111 				if (get_user(val, (u8 __user *)EA))
1112 					return -EFAULT;
1113 				/* first time updating this reg,
1114 				 * zero it out */
1115 				if (pos == 0)
1116 					regs->gpr[rT] = 0;
1117 				regs->gpr[rT] |= val << shift;
1118 				break;
1119 			case PPC_INST_STSWI:
1120 			case PPC_INST_STSWX:
1121 				val = regs->gpr[rT] >> shift;
1122 				if (put_user(val, (u8 __user *)EA))
1123 					return -EFAULT;
1124 				break;
1125 		}
1126 		/* move EA to next address */
1127 		EA += 1;
1128 		num_bytes--;
1129 
1130 		/* manage our position within the register */
1131 		if (++pos == 4) {
1132 			pos = 0;
1133 			if (++rT == 32)
1134 				rT = 0;
1135 		}
1136 	}
1137 
1138 	return 0;
1139 }
1140 
1141 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1142 {
1143 	u32 ra,rs;
1144 	unsigned long tmp;
1145 
1146 	ra = (instword >> 16) & 0x1f;
1147 	rs = (instword >> 21) & 0x1f;
1148 
1149 	tmp = regs->gpr[rs];
1150 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1151 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1152 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1153 	regs->gpr[ra] = tmp;
1154 
1155 	return 0;
1156 }
1157 
1158 static int emulate_isel(struct pt_regs *regs, u32 instword)
1159 {
1160 	u8 rT = (instword >> 21) & 0x1f;
1161 	u8 rA = (instword >> 16) & 0x1f;
1162 	u8 rB = (instword >> 11) & 0x1f;
1163 	u8 BC = (instword >> 6) & 0x1f;
1164 	u8 bit;
1165 	unsigned long tmp;
1166 
1167 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1168 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1169 
1170 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1171 
1172 	return 0;
1173 }
1174 
1175 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1176 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1177 {
1178         /* If we're emulating a load/store in an active transaction, we cannot
1179          * emulate it as the kernel operates in transaction suspended context.
1180          * We need to abort the transaction.  This creates a persistent TM
1181          * abort so tell the user what caused it with a new code.
1182 	 */
1183 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1184 		tm_enable();
1185 		tm_abort(cause);
1186 		return true;
1187 	}
1188 	return false;
1189 }
1190 #else
1191 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1192 {
1193 	return false;
1194 }
1195 #endif
1196 
1197 static int emulate_instruction(struct pt_regs *regs)
1198 {
1199 	u32 instword;
1200 	u32 rd;
1201 
1202 	if (!user_mode(regs))
1203 		return -EINVAL;
1204 	CHECK_FULL_REGS(regs);
1205 
1206 	if (get_user(instword, (u32 __user *)(regs->nip)))
1207 		return -EFAULT;
1208 
1209 	/* Emulate the mfspr rD, PVR. */
1210 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1211 		PPC_WARN_EMULATED(mfpvr, regs);
1212 		rd = (instword >> 21) & 0x1f;
1213 		regs->gpr[rd] = mfspr(SPRN_PVR);
1214 		return 0;
1215 	}
1216 
1217 	/* Emulating the dcba insn is just a no-op.  */
1218 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1219 		PPC_WARN_EMULATED(dcba, regs);
1220 		return 0;
1221 	}
1222 
1223 	/* Emulate the mcrxr insn.  */
1224 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1225 		int shift = (instword >> 21) & 0x1c;
1226 		unsigned long msk = 0xf0000000UL >> shift;
1227 
1228 		PPC_WARN_EMULATED(mcrxr, regs);
1229 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1230 		regs->xer &= ~0xf0000000UL;
1231 		return 0;
1232 	}
1233 
1234 	/* Emulate load/store string insn. */
1235 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1236 		if (tm_abort_check(regs,
1237 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1238 			return -EINVAL;
1239 		PPC_WARN_EMULATED(string, regs);
1240 		return emulate_string_inst(regs, instword);
1241 	}
1242 
1243 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1244 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1245 		PPC_WARN_EMULATED(popcntb, regs);
1246 		return emulate_popcntb_inst(regs, instword);
1247 	}
1248 
1249 	/* Emulate isel (Integer Select) instruction */
1250 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1251 		PPC_WARN_EMULATED(isel, regs);
1252 		return emulate_isel(regs, instword);
1253 	}
1254 
1255 	/* Emulate sync instruction variants */
1256 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1257 		PPC_WARN_EMULATED(sync, regs);
1258 		asm volatile("sync");
1259 		return 0;
1260 	}
1261 
1262 #ifdef CONFIG_PPC64
1263 	/* Emulate the mfspr rD, DSCR. */
1264 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1265 		PPC_INST_MFSPR_DSCR_USER) ||
1266 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1267 		PPC_INST_MFSPR_DSCR)) &&
1268 			cpu_has_feature(CPU_FTR_DSCR)) {
1269 		PPC_WARN_EMULATED(mfdscr, regs);
1270 		rd = (instword >> 21) & 0x1f;
1271 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1272 		return 0;
1273 	}
1274 	/* Emulate the mtspr DSCR, rD. */
1275 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1276 		PPC_INST_MTSPR_DSCR_USER) ||
1277 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1278 		PPC_INST_MTSPR_DSCR)) &&
1279 			cpu_has_feature(CPU_FTR_DSCR)) {
1280 		PPC_WARN_EMULATED(mtdscr, regs);
1281 		rd = (instword >> 21) & 0x1f;
1282 		current->thread.dscr = regs->gpr[rd];
1283 		current->thread.dscr_inherit = 1;
1284 		mtspr(SPRN_DSCR, current->thread.dscr);
1285 		return 0;
1286 	}
1287 #endif
1288 
1289 	return -EINVAL;
1290 }
1291 
1292 int is_valid_bugaddr(unsigned long addr)
1293 {
1294 	return is_kernel_addr(addr);
1295 }
1296 
1297 #ifdef CONFIG_MATH_EMULATION
1298 static int emulate_math(struct pt_regs *regs)
1299 {
1300 	int ret;
1301 	extern int do_mathemu(struct pt_regs *regs);
1302 
1303 	ret = do_mathemu(regs);
1304 	if (ret >= 0)
1305 		PPC_WARN_EMULATED(math, regs);
1306 
1307 	switch (ret) {
1308 	case 0:
1309 		emulate_single_step(regs);
1310 		return 0;
1311 	case 1: {
1312 			int code = 0;
1313 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1314 			_exception(SIGFPE, regs, code, regs->nip);
1315 			return 0;
1316 		}
1317 	case -EFAULT:
1318 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1319 		return 0;
1320 	}
1321 
1322 	return -1;
1323 }
1324 #else
1325 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1326 #endif
1327 
1328 void program_check_exception(struct pt_regs *regs)
1329 {
1330 	enum ctx_state prev_state = exception_enter();
1331 	unsigned int reason = get_reason(regs);
1332 
1333 	/* We can now get here via a FP Unavailable exception if the core
1334 	 * has no FPU, in that case the reason flags will be 0 */
1335 
1336 	if (reason & REASON_FP) {
1337 		/* IEEE FP exception */
1338 		parse_fpe(regs);
1339 		goto bail;
1340 	}
1341 	if (reason & REASON_TRAP) {
1342 		unsigned long bugaddr;
1343 		/* Debugger is first in line to stop recursive faults in
1344 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1345 		if (debugger_bpt(regs))
1346 			goto bail;
1347 
1348 		if (kprobe_handler(regs))
1349 			goto bail;
1350 
1351 		/* trap exception */
1352 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1353 				== NOTIFY_STOP)
1354 			goto bail;
1355 
1356 		bugaddr = regs->nip;
1357 		/*
1358 		 * Fixup bugaddr for BUG_ON() in real mode
1359 		 */
1360 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1361 			bugaddr += PAGE_OFFSET;
1362 
1363 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1364 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1365 			regs->nip += 4;
1366 			goto bail;
1367 		}
1368 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1369 		goto bail;
1370 	}
1371 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1372 	if (reason & REASON_TM) {
1373 		/* This is a TM "Bad Thing Exception" program check.
1374 		 * This occurs when:
1375 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1376 		 *    transition in TM states.
1377 		 * -  A trechkpt is attempted when transactional.
1378 		 * -  A treclaim is attempted when non transactional.
1379 		 * -  A tend is illegally attempted.
1380 		 * -  writing a TM SPR when transactional.
1381 		 *
1382 		 * If usermode caused this, it's done something illegal and
1383 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1384 		 * operand to distinguish from the instruction just being bad
1385 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1386 		 * illegal /placement/ of a valid instruction.
1387 		 */
1388 		if (user_mode(regs)) {
1389 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1390 			goto bail;
1391 		} else {
1392 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1393 			       "at %lx (msr 0x%x)\n", regs->nip, reason);
1394 			die("Unrecoverable exception", regs, SIGABRT);
1395 		}
1396 	}
1397 #endif
1398 
1399 	/*
1400 	 * If we took the program check in the kernel skip down to sending a
1401 	 * SIGILL. The subsequent cases all relate to emulating instructions
1402 	 * which we should only do for userspace. We also do not want to enable
1403 	 * interrupts for kernel faults because that might lead to further
1404 	 * faults, and loose the context of the original exception.
1405 	 */
1406 	if (!user_mode(regs))
1407 		goto sigill;
1408 
1409 	/* We restore the interrupt state now */
1410 	if (!arch_irq_disabled_regs(regs))
1411 		local_irq_enable();
1412 
1413 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1414 	 * but there seems to be a hardware bug on the 405GP (RevD)
1415 	 * that means ESR is sometimes set incorrectly - either to
1416 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1417 	 * hardware people - not sure if it can happen on any illegal
1418 	 * instruction or only on FP instructions, whether there is a
1419 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1420 	 */
1421 	if (!emulate_math(regs))
1422 		goto bail;
1423 
1424 	/* Try to emulate it if we should. */
1425 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1426 		switch (emulate_instruction(regs)) {
1427 		case 0:
1428 			regs->nip += 4;
1429 			emulate_single_step(regs);
1430 			goto bail;
1431 		case -EFAULT:
1432 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1433 			goto bail;
1434 		}
1435 	}
1436 
1437 sigill:
1438 	if (reason & REASON_PRIVILEGED)
1439 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1440 	else
1441 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1442 
1443 bail:
1444 	exception_exit(prev_state);
1445 }
1446 NOKPROBE_SYMBOL(program_check_exception);
1447 
1448 /*
1449  * This occurs when running in hypervisor mode on POWER6 or later
1450  * and an illegal instruction is encountered.
1451  */
1452 void emulation_assist_interrupt(struct pt_regs *regs)
1453 {
1454 	regs->msr |= REASON_ILLEGAL;
1455 	program_check_exception(regs);
1456 }
1457 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1458 
1459 void alignment_exception(struct pt_regs *regs)
1460 {
1461 	enum ctx_state prev_state = exception_enter();
1462 	int sig, code, fixed = 0;
1463 
1464 	/* We restore the interrupt state now */
1465 	if (!arch_irq_disabled_regs(regs))
1466 		local_irq_enable();
1467 
1468 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1469 		goto bail;
1470 
1471 	/* we don't implement logging of alignment exceptions */
1472 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1473 		fixed = fix_alignment(regs);
1474 
1475 	if (fixed == 1) {
1476 		regs->nip += 4;	/* skip over emulated instruction */
1477 		emulate_single_step(regs);
1478 		goto bail;
1479 	}
1480 
1481 	/* Operand address was bad */
1482 	if (fixed == -EFAULT) {
1483 		sig = SIGSEGV;
1484 		code = SEGV_ACCERR;
1485 	} else {
1486 		sig = SIGBUS;
1487 		code = BUS_ADRALN;
1488 	}
1489 	if (user_mode(regs))
1490 		_exception(sig, regs, code, regs->dar);
1491 	else
1492 		bad_page_fault(regs, regs->dar, sig);
1493 
1494 bail:
1495 	exception_exit(prev_state);
1496 }
1497 
1498 void slb_miss_bad_addr(struct pt_regs *regs)
1499 {
1500 	enum ctx_state prev_state = exception_enter();
1501 
1502 	if (user_mode(regs))
1503 		_exception(SIGSEGV, regs, SEGV_BNDERR, regs->dar);
1504 	else
1505 		bad_page_fault(regs, regs->dar, SIGSEGV);
1506 
1507 	exception_exit(prev_state);
1508 }
1509 
1510 void StackOverflow(struct pt_regs *regs)
1511 {
1512 	printk(KERN_CRIT "Kernel stack overflow in process %p, r1=%lx\n",
1513 	       current, regs->gpr[1]);
1514 	debugger(regs);
1515 	show_regs(regs);
1516 	panic("kernel stack overflow");
1517 }
1518 
1519 void nonrecoverable_exception(struct pt_regs *regs)
1520 {
1521 	printk(KERN_ERR "Non-recoverable exception at PC=%lx MSR=%lx\n",
1522 	       regs->nip, regs->msr);
1523 	debugger(regs);
1524 	die("nonrecoverable exception", regs, SIGKILL);
1525 }
1526 
1527 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1528 {
1529 	enum ctx_state prev_state = exception_enter();
1530 
1531 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1532 			  "%lx at %lx\n", regs->trap, regs->nip);
1533 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1534 
1535 	exception_exit(prev_state);
1536 }
1537 
1538 void altivec_unavailable_exception(struct pt_regs *regs)
1539 {
1540 	enum ctx_state prev_state = exception_enter();
1541 
1542 	if (user_mode(regs)) {
1543 		/* A user program has executed an altivec instruction,
1544 		   but this kernel doesn't support altivec. */
1545 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1546 		goto bail;
1547 	}
1548 
1549 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1550 			"%lx at %lx\n", regs->trap, regs->nip);
1551 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1552 
1553 bail:
1554 	exception_exit(prev_state);
1555 }
1556 
1557 void vsx_unavailable_exception(struct pt_regs *regs)
1558 {
1559 	if (user_mode(regs)) {
1560 		/* A user program has executed an vsx instruction,
1561 		   but this kernel doesn't support vsx. */
1562 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1563 		return;
1564 	}
1565 
1566 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1567 			"%lx at %lx\n", regs->trap, regs->nip);
1568 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1569 }
1570 
1571 #ifdef CONFIG_PPC64
1572 static void tm_unavailable(struct pt_regs *regs)
1573 {
1574 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1575 	if (user_mode(regs)) {
1576 		current->thread.load_tm++;
1577 		regs->msr |= MSR_TM;
1578 		tm_enable();
1579 		tm_restore_sprs(&current->thread);
1580 		return;
1581 	}
1582 #endif
1583 	pr_emerg("Unrecoverable TM Unavailable Exception "
1584 			"%lx at %lx\n", regs->trap, regs->nip);
1585 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1586 }
1587 
1588 void facility_unavailable_exception(struct pt_regs *regs)
1589 {
1590 	static char *facility_strings[] = {
1591 		[FSCR_FP_LG] = "FPU",
1592 		[FSCR_VECVSX_LG] = "VMX/VSX",
1593 		[FSCR_DSCR_LG] = "DSCR",
1594 		[FSCR_PM_LG] = "PMU SPRs",
1595 		[FSCR_BHRB_LG] = "BHRB",
1596 		[FSCR_TM_LG] = "TM",
1597 		[FSCR_EBB_LG] = "EBB",
1598 		[FSCR_TAR_LG] = "TAR",
1599 		[FSCR_MSGP_LG] = "MSGP",
1600 		[FSCR_SCV_LG] = "SCV",
1601 	};
1602 	char *facility = "unknown";
1603 	u64 value;
1604 	u32 instword, rd;
1605 	u8 status;
1606 	bool hv;
1607 
1608 	hv = (TRAP(regs) == 0xf80);
1609 	if (hv)
1610 		value = mfspr(SPRN_HFSCR);
1611 	else
1612 		value = mfspr(SPRN_FSCR);
1613 
1614 	status = value >> 56;
1615 	if (status == FSCR_DSCR_LG) {
1616 		/*
1617 		 * User is accessing the DSCR register using the problem
1618 		 * state only SPR number (0x03) either through a mfspr or
1619 		 * a mtspr instruction. If it is a write attempt through
1620 		 * a mtspr, then we set the inherit bit. This also allows
1621 		 * the user to write or read the register directly in the
1622 		 * future by setting via the FSCR DSCR bit. But in case it
1623 		 * is a read DSCR attempt through a mfspr instruction, we
1624 		 * just emulate the instruction instead. This code path will
1625 		 * always emulate all the mfspr instructions till the user
1626 		 * has attempted at least one mtspr instruction. This way it
1627 		 * preserves the same behaviour when the user is accessing
1628 		 * the DSCR through privilege level only SPR number (0x11)
1629 		 * which is emulated through illegal instruction exception.
1630 		 * We always leave HFSCR DSCR set.
1631 		 */
1632 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1633 			pr_err("Failed to fetch the user instruction\n");
1634 			return;
1635 		}
1636 
1637 		/* Write into DSCR (mtspr 0x03, RS) */
1638 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1639 				== PPC_INST_MTSPR_DSCR_USER) {
1640 			rd = (instword >> 21) & 0x1f;
1641 			current->thread.dscr = regs->gpr[rd];
1642 			current->thread.dscr_inherit = 1;
1643 			current->thread.fscr |= FSCR_DSCR;
1644 			mtspr(SPRN_FSCR, current->thread.fscr);
1645 		}
1646 
1647 		/* Read from DSCR (mfspr RT, 0x03) */
1648 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1649 				== PPC_INST_MFSPR_DSCR_USER) {
1650 			if (emulate_instruction(regs)) {
1651 				pr_err("DSCR based mfspr emulation failed\n");
1652 				return;
1653 			}
1654 			regs->nip += 4;
1655 			emulate_single_step(regs);
1656 		}
1657 		return;
1658 	}
1659 
1660 	if (status == FSCR_TM_LG) {
1661 		/*
1662 		 * If we're here then the hardware is TM aware because it
1663 		 * generated an exception with FSRM_TM set.
1664 		 *
1665 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1666 		 * told us not to do TM, or the kernel is not built with TM
1667 		 * support.
1668 		 *
1669 		 * If both of those things are true, then userspace can spam the
1670 		 * console by triggering the printk() below just by continually
1671 		 * doing tbegin (or any TM instruction). So in that case just
1672 		 * send the process a SIGILL immediately.
1673 		 */
1674 		if (!cpu_has_feature(CPU_FTR_TM))
1675 			goto out;
1676 
1677 		tm_unavailable(regs);
1678 		return;
1679 	}
1680 
1681 	if ((hv || status >= 2) &&
1682 	    (status < ARRAY_SIZE(facility_strings)) &&
1683 	    facility_strings[status])
1684 		facility = facility_strings[status];
1685 
1686 	/* We restore the interrupt state now */
1687 	if (!arch_irq_disabled_regs(regs))
1688 		local_irq_enable();
1689 
1690 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1691 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1692 
1693 out:
1694 	if (user_mode(regs)) {
1695 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1696 		return;
1697 	}
1698 
1699 	die("Unexpected facility unavailable exception", regs, SIGABRT);
1700 }
1701 #endif
1702 
1703 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1704 
1705 void fp_unavailable_tm(struct pt_regs *regs)
1706 {
1707 	/* Note:  This does not handle any kind of FP laziness. */
1708 
1709 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1710 		 regs->nip, regs->msr);
1711 
1712         /* We can only have got here if the task started using FP after
1713          * beginning the transaction.  So, the transactional regs are just a
1714          * copy of the checkpointed ones.  But, we still need to recheckpoint
1715          * as we're enabling FP for the process; it will return, abort the
1716          * transaction, and probably retry but now with FP enabled.  So the
1717          * checkpointed FP registers need to be loaded.
1718 	 */
1719 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1720 	/* Reclaim didn't save out any FPRs to transact_fprs. */
1721 
1722 	/* Enable FP for the task: */
1723 	current->thread.load_fp = 1;
1724 
1725 	/* This loads and recheckpoints the FP registers from
1726 	 * thread.fpr[].  They will remain in registers after the
1727 	 * checkpoint so we don't need to reload them after.
1728 	 * If VMX is in use, the VRs now hold checkpointed values,
1729 	 * so we don't want to load the VRs from the thread_struct.
1730 	 */
1731 	tm_recheckpoint(&current->thread);
1732 }
1733 
1734 void altivec_unavailable_tm(struct pt_regs *regs)
1735 {
1736 	/* See the comments in fp_unavailable_tm().  This function operates
1737 	 * the same way.
1738 	 */
1739 
1740 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1741 		 "MSR=%lx\n",
1742 		 regs->nip, regs->msr);
1743 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1744 	current->thread.load_vec = 1;
1745 	tm_recheckpoint(&current->thread);
1746 	current->thread.used_vr = 1;
1747 }
1748 
1749 void vsx_unavailable_tm(struct pt_regs *regs)
1750 {
1751 	/* See the comments in fp_unavailable_tm().  This works similarly,
1752 	 * though we're loading both FP and VEC registers in here.
1753 	 *
1754 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1755 	 * regs.  Either way, set MSR_VSX.
1756 	 */
1757 
1758 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1759 		 "MSR=%lx\n",
1760 		 regs->nip, regs->msr);
1761 
1762 	current->thread.used_vsr = 1;
1763 
1764 	/* This reclaims FP and/or VR regs if they're already enabled */
1765 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1766 
1767 	current->thread.load_vec = 1;
1768 	current->thread.load_fp = 1;
1769 
1770 	tm_recheckpoint(&current->thread);
1771 }
1772 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1773 
1774 void performance_monitor_exception(struct pt_regs *regs)
1775 {
1776 	__this_cpu_inc(irq_stat.pmu_irqs);
1777 
1778 	perf_irq(regs);
1779 }
1780 
1781 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1782 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1783 {
1784 	int changed = 0;
1785 	/*
1786 	 * Determine the cause of the debug event, clear the
1787 	 * event flags and send a trap to the handler. Torez
1788 	 */
1789 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1790 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1791 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1792 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1793 #endif
1794 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1795 			     5);
1796 		changed |= 0x01;
1797 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1798 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1799 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1800 			     6);
1801 		changed |= 0x01;
1802 	}  else if (debug_status & DBSR_IAC1) {
1803 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1804 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1805 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1806 			     1);
1807 		changed |= 0x01;
1808 	}  else if (debug_status & DBSR_IAC2) {
1809 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1810 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1811 			     2);
1812 		changed |= 0x01;
1813 	}  else if (debug_status & DBSR_IAC3) {
1814 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1815 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1816 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1817 			     3);
1818 		changed |= 0x01;
1819 	}  else if (debug_status & DBSR_IAC4) {
1820 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1821 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1822 			     4);
1823 		changed |= 0x01;
1824 	}
1825 	/*
1826 	 * At the point this routine was called, the MSR(DE) was turned off.
1827 	 * Check all other debug flags and see if that bit needs to be turned
1828 	 * back on or not.
1829 	 */
1830 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1831 			       current->thread.debug.dbcr1))
1832 		regs->msr |= MSR_DE;
1833 	else
1834 		/* Make sure the IDM flag is off */
1835 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1836 
1837 	if (changed & 0x01)
1838 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1839 }
1840 
1841 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1842 {
1843 	current->thread.debug.dbsr = debug_status;
1844 
1845 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1846 	 * on server, it stops on the target of the branch. In order to simulate
1847 	 * the server behaviour, we thus restart right away with a single step
1848 	 * instead of stopping here when hitting a BT
1849 	 */
1850 	if (debug_status & DBSR_BT) {
1851 		regs->msr &= ~MSR_DE;
1852 
1853 		/* Disable BT */
1854 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1855 		/* Clear the BT event */
1856 		mtspr(SPRN_DBSR, DBSR_BT);
1857 
1858 		/* Do the single step trick only when coming from userspace */
1859 		if (user_mode(regs)) {
1860 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1861 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1862 			regs->msr |= MSR_DE;
1863 			return;
1864 		}
1865 
1866 		if (kprobe_post_handler(regs))
1867 			return;
1868 
1869 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1870 			       5, SIGTRAP) == NOTIFY_STOP) {
1871 			return;
1872 		}
1873 		if (debugger_sstep(regs))
1874 			return;
1875 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1876 		regs->msr &= ~MSR_DE;
1877 
1878 		/* Disable instruction completion */
1879 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
1880 		/* Clear the instruction completion event */
1881 		mtspr(SPRN_DBSR, DBSR_IC);
1882 
1883 		if (kprobe_post_handler(regs))
1884 			return;
1885 
1886 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1887 			       5, SIGTRAP) == NOTIFY_STOP) {
1888 			return;
1889 		}
1890 
1891 		if (debugger_sstep(regs))
1892 			return;
1893 
1894 		if (user_mode(regs)) {
1895 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
1896 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1897 					       current->thread.debug.dbcr1))
1898 				regs->msr |= MSR_DE;
1899 			else
1900 				/* Make sure the IDM bit is off */
1901 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1902 		}
1903 
1904 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1905 	} else
1906 		handle_debug(regs, debug_status);
1907 }
1908 NOKPROBE_SYMBOL(DebugException);
1909 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
1910 
1911 #if !defined(CONFIG_TAU_INT)
1912 void TAUException(struct pt_regs *regs)
1913 {
1914 	printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
1915 	       regs->nip, regs->msr, regs->trap, print_tainted());
1916 }
1917 #endif /* CONFIG_INT_TAU */
1918 
1919 #ifdef CONFIG_ALTIVEC
1920 void altivec_assist_exception(struct pt_regs *regs)
1921 {
1922 	int err;
1923 
1924 	if (!user_mode(regs)) {
1925 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
1926 		       " at %lx\n", regs->nip);
1927 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
1928 	}
1929 
1930 	flush_altivec_to_thread(current);
1931 
1932 	PPC_WARN_EMULATED(altivec, regs);
1933 	err = emulate_altivec(regs);
1934 	if (err == 0) {
1935 		regs->nip += 4;		/* skip emulated instruction */
1936 		emulate_single_step(regs);
1937 		return;
1938 	}
1939 
1940 	if (err == -EFAULT) {
1941 		/* got an error reading the instruction */
1942 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
1943 	} else {
1944 		/* didn't recognize the instruction */
1945 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
1946 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
1947 				   "in %s at %lx\n", current->comm, regs->nip);
1948 		current->thread.vr_state.vscr.u[3] |= 0x10000;
1949 	}
1950 }
1951 #endif /* CONFIG_ALTIVEC */
1952 
1953 #ifdef CONFIG_FSL_BOOKE
1954 void CacheLockingException(struct pt_regs *regs, unsigned long address,
1955 			   unsigned long error_code)
1956 {
1957 	/* We treat cache locking instructions from the user
1958 	 * as priv ops, in the future we could try to do
1959 	 * something smarter
1960 	 */
1961 	if (error_code & (ESR_DLK|ESR_ILK))
1962 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1963 	return;
1964 }
1965 #endif /* CONFIG_FSL_BOOKE */
1966 
1967 #ifdef CONFIG_SPE
1968 void SPEFloatingPointException(struct pt_regs *regs)
1969 {
1970 	extern int do_spe_mathemu(struct pt_regs *regs);
1971 	unsigned long spefscr;
1972 	int fpexc_mode;
1973 	int code = FPE_FIXME;
1974 	int err;
1975 
1976 	flush_spe_to_thread(current);
1977 
1978 	spefscr = current->thread.spefscr;
1979 	fpexc_mode = current->thread.fpexc_mode;
1980 
1981 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
1982 		code = FPE_FLTOVF;
1983 	}
1984 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
1985 		code = FPE_FLTUND;
1986 	}
1987 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
1988 		code = FPE_FLTDIV;
1989 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
1990 		code = FPE_FLTINV;
1991 	}
1992 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
1993 		code = FPE_FLTRES;
1994 
1995 	err = do_spe_mathemu(regs);
1996 	if (err == 0) {
1997 		regs->nip += 4;		/* skip emulated instruction */
1998 		emulate_single_step(regs);
1999 		return;
2000 	}
2001 
2002 	if (err == -EFAULT) {
2003 		/* got an error reading the instruction */
2004 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2005 	} else if (err == -EINVAL) {
2006 		/* didn't recognize the instruction */
2007 		printk(KERN_ERR "unrecognized spe instruction "
2008 		       "in %s at %lx\n", current->comm, regs->nip);
2009 	} else {
2010 		_exception(SIGFPE, regs, code, regs->nip);
2011 	}
2012 
2013 	return;
2014 }
2015 
2016 void SPEFloatingPointRoundException(struct pt_regs *regs)
2017 {
2018 	extern int speround_handler(struct pt_regs *regs);
2019 	int err;
2020 
2021 	preempt_disable();
2022 	if (regs->msr & MSR_SPE)
2023 		giveup_spe(current);
2024 	preempt_enable();
2025 
2026 	regs->nip -= 4;
2027 	err = speround_handler(regs);
2028 	if (err == 0) {
2029 		regs->nip += 4;		/* skip emulated instruction */
2030 		emulate_single_step(regs);
2031 		return;
2032 	}
2033 
2034 	if (err == -EFAULT) {
2035 		/* got an error reading the instruction */
2036 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2037 	} else if (err == -EINVAL) {
2038 		/* didn't recognize the instruction */
2039 		printk(KERN_ERR "unrecognized spe instruction "
2040 		       "in %s at %lx\n", current->comm, regs->nip);
2041 	} else {
2042 		_exception(SIGFPE, regs, FPE_FIXME, regs->nip);
2043 		return;
2044 	}
2045 }
2046 #endif
2047 
2048 /*
2049  * We enter here if we get an unrecoverable exception, that is, one
2050  * that happened at a point where the RI (recoverable interrupt) bit
2051  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2052  * we therefore lost state by taking this exception.
2053  */
2054 void unrecoverable_exception(struct pt_regs *regs)
2055 {
2056 	printk(KERN_EMERG "Unrecoverable exception %lx at %lx\n",
2057 	       regs->trap, regs->nip);
2058 	die("Unrecoverable exception", regs, SIGABRT);
2059 }
2060 NOKPROBE_SYMBOL(unrecoverable_exception);
2061 
2062 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2063 /*
2064  * Default handler for a Watchdog exception,
2065  * spins until a reboot occurs
2066  */
2067 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2068 {
2069 	/* Generic WatchdogHandler, implement your own */
2070 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2071 	return;
2072 }
2073 
2074 void WatchdogException(struct pt_regs *regs)
2075 {
2076 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2077 	WatchdogHandler(regs);
2078 }
2079 #endif
2080 
2081 /*
2082  * We enter here if we discover during exception entry that we are
2083  * running in supervisor mode with a userspace value in the stack pointer.
2084  */
2085 void kernel_bad_stack(struct pt_regs *regs)
2086 {
2087 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2088 	       regs->gpr[1], regs->nip);
2089 	die("Bad kernel stack pointer", regs, SIGABRT);
2090 }
2091 NOKPROBE_SYMBOL(kernel_bad_stack);
2092 
2093 void __init trap_init(void)
2094 {
2095 }
2096 
2097 
2098 #ifdef CONFIG_PPC_EMULATED_STATS
2099 
2100 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2101 
2102 struct ppc_emulated ppc_emulated = {
2103 #ifdef CONFIG_ALTIVEC
2104 	WARN_EMULATED_SETUP(altivec),
2105 #endif
2106 	WARN_EMULATED_SETUP(dcba),
2107 	WARN_EMULATED_SETUP(dcbz),
2108 	WARN_EMULATED_SETUP(fp_pair),
2109 	WARN_EMULATED_SETUP(isel),
2110 	WARN_EMULATED_SETUP(mcrxr),
2111 	WARN_EMULATED_SETUP(mfpvr),
2112 	WARN_EMULATED_SETUP(multiple),
2113 	WARN_EMULATED_SETUP(popcntb),
2114 	WARN_EMULATED_SETUP(spe),
2115 	WARN_EMULATED_SETUP(string),
2116 	WARN_EMULATED_SETUP(sync),
2117 	WARN_EMULATED_SETUP(unaligned),
2118 #ifdef CONFIG_MATH_EMULATION
2119 	WARN_EMULATED_SETUP(math),
2120 #endif
2121 #ifdef CONFIG_VSX
2122 	WARN_EMULATED_SETUP(vsx),
2123 #endif
2124 #ifdef CONFIG_PPC64
2125 	WARN_EMULATED_SETUP(mfdscr),
2126 	WARN_EMULATED_SETUP(mtdscr),
2127 	WARN_EMULATED_SETUP(lq_stq),
2128 	WARN_EMULATED_SETUP(lxvw4x),
2129 	WARN_EMULATED_SETUP(lxvh8x),
2130 	WARN_EMULATED_SETUP(lxvd2x),
2131 	WARN_EMULATED_SETUP(lxvb16x),
2132 #endif
2133 };
2134 
2135 u32 ppc_warn_emulated;
2136 
2137 void ppc_warn_emulated_print(const char *type)
2138 {
2139 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2140 			    type);
2141 }
2142 
2143 static int __init ppc_warn_emulated_init(void)
2144 {
2145 	struct dentry *dir, *d;
2146 	unsigned int i;
2147 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2148 
2149 	if (!powerpc_debugfs_root)
2150 		return -ENODEV;
2151 
2152 	dir = debugfs_create_dir("emulated_instructions",
2153 				 powerpc_debugfs_root);
2154 	if (!dir)
2155 		return -ENOMEM;
2156 
2157 	d = debugfs_create_u32("do_warn", 0644, dir,
2158 			       &ppc_warn_emulated);
2159 	if (!d)
2160 		goto fail;
2161 
2162 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) {
2163 		d = debugfs_create_u32(entries[i].name, 0644, dir,
2164 				       (u32 *)&entries[i].val.counter);
2165 		if (!d)
2166 			goto fail;
2167 	}
2168 
2169 	return 0;
2170 
2171 fail:
2172 	debugfs_remove_recursive(dir);
2173 	return -ENOMEM;
2174 }
2175 
2176 device_initcall(ppc_warn_emulated_init);
2177 
2178 #endif /* CONFIG_PPC_EMULATED_STATS */
2179