xref: /linux/arch/powerpc/kernel/traps.c (revision 89aca4753eb451a48f65a12b02640365dba3d4ea)
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/stddef.h>
24 #include <linux/unistd.h>
25 #include <linux/ptrace.h>
26 #include <linux/user.h>
27 #include <linux/interrupt.h>
28 #include <linux/init.h>
29 #include <linux/extable.h>
30 #include <linux/module.h>	/* print_modules */
31 #include <linux/prctl.h>
32 #include <linux/delay.h>
33 #include <linux/kprobes.h>
34 #include <linux/kexec.h>
35 #include <linux/backlight.h>
36 #include <linux/bug.h>
37 #include <linux/kdebug.h>
38 #include <linux/ratelimit.h>
39 #include <linux/context_tracking.h>
40 #include <linux/smp.h>
41 
42 #include <asm/emulated_ops.h>
43 #include <asm/pgtable.h>
44 #include <linux/uaccess.h>
45 #include <asm/debugfs.h>
46 #include <asm/io.h>
47 #include <asm/machdep.h>
48 #include <asm/rtas.h>
49 #include <asm/pmc.h>
50 #include <asm/reg.h>
51 #ifdef CONFIG_PMAC_BACKLIGHT
52 #include <asm/backlight.h>
53 #endif
54 #ifdef CONFIG_PPC64
55 #include <asm/firmware.h>
56 #include <asm/processor.h>
57 #include <asm/tm.h>
58 #endif
59 #include <asm/kexec.h>
60 #include <asm/ppc-opcode.h>
61 #include <asm/rio.h>
62 #include <asm/fadump.h>
63 #include <asm/switch_to.h>
64 #include <asm/tm.h>
65 #include <asm/debug.h>
66 #include <asm/asm-prototypes.h>
67 #include <asm/hmi.h>
68 #include <sysdev/fsl_pci.h>
69 #include <asm/kprobes.h>
70 
71 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
72 int (*__debugger)(struct pt_regs *regs) __read_mostly;
73 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
74 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
75 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
79 
80 EXPORT_SYMBOL(__debugger);
81 EXPORT_SYMBOL(__debugger_ipi);
82 EXPORT_SYMBOL(__debugger_bpt);
83 EXPORT_SYMBOL(__debugger_sstep);
84 EXPORT_SYMBOL(__debugger_iabr_match);
85 EXPORT_SYMBOL(__debugger_break_match);
86 EXPORT_SYMBOL(__debugger_fault_handler);
87 #endif
88 
89 /* Transactional Memory trap debug */
90 #ifdef TM_DEBUG_SW
91 #define TM_DEBUG(x...) printk(KERN_INFO x)
92 #else
93 #define TM_DEBUG(x...) do { } while(0)
94 #endif
95 
96 /*
97  * Trap & Exception support
98  */
99 
100 #ifdef CONFIG_PMAC_BACKLIGHT
101 static void pmac_backlight_unblank(void)
102 {
103 	mutex_lock(&pmac_backlight_mutex);
104 	if (pmac_backlight) {
105 		struct backlight_properties *props;
106 
107 		props = &pmac_backlight->props;
108 		props->brightness = props->max_brightness;
109 		props->power = FB_BLANK_UNBLANK;
110 		backlight_update_status(pmac_backlight);
111 	}
112 	mutex_unlock(&pmac_backlight_mutex);
113 }
114 #else
115 static inline void pmac_backlight_unblank(void) { }
116 #endif
117 
118 /*
119  * If oops/die is expected to crash the machine, return true here.
120  *
121  * This should not be expected to be 100% accurate, there may be
122  * notifiers registered or other unexpected conditions that may bring
123  * down the kernel. Or if the current process in the kernel is holding
124  * locks or has other critical state, the kernel may become effectively
125  * unusable anyway.
126  */
127 bool die_will_crash(void)
128 {
129 	if (should_fadump_crash())
130 		return true;
131 	if (kexec_should_crash(current))
132 		return true;
133 	if (in_interrupt() || panic_on_oops ||
134 			!current->pid || is_global_init(current))
135 		return true;
136 
137 	return false;
138 }
139 
140 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
141 static int die_owner = -1;
142 static unsigned int die_nest_count;
143 static int die_counter;
144 
145 static unsigned long oops_begin(struct pt_regs *regs)
146 {
147 	int cpu;
148 	unsigned long flags;
149 
150 	oops_enter();
151 
152 	/* racy, but better than risking deadlock. */
153 	raw_local_irq_save(flags);
154 	cpu = smp_processor_id();
155 	if (!arch_spin_trylock(&die_lock)) {
156 		if (cpu == die_owner)
157 			/* nested oops. should stop eventually */;
158 		else
159 			arch_spin_lock(&die_lock);
160 	}
161 	die_nest_count++;
162 	die_owner = cpu;
163 	console_verbose();
164 	bust_spinlocks(1);
165 	if (machine_is(powermac))
166 		pmac_backlight_unblank();
167 	return flags;
168 }
169 NOKPROBE_SYMBOL(oops_begin);
170 
171 static void oops_end(unsigned long flags, struct pt_regs *regs,
172 			       int signr)
173 {
174 	bust_spinlocks(0);
175 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
176 	die_nest_count--;
177 	oops_exit();
178 	printk("\n");
179 	if (!die_nest_count) {
180 		/* Nest count reaches zero, release the lock. */
181 		die_owner = -1;
182 		arch_spin_unlock(&die_lock);
183 	}
184 	raw_local_irq_restore(flags);
185 
186 	crash_fadump(regs, "die oops");
187 
188 	if (kexec_should_crash(current))
189 		crash_kexec(regs);
190 
191 	if (!signr)
192 		return;
193 
194 	/*
195 	 * While our oops output is serialised by a spinlock, output
196 	 * from panic() called below can race and corrupt it. If we
197 	 * know we are going to panic, delay for 1 second so we have a
198 	 * chance to get clean backtraces from all CPUs that are oopsing.
199 	 */
200 	if (in_interrupt() || panic_on_oops || !current->pid ||
201 	    is_global_init(current)) {
202 		mdelay(MSEC_PER_SEC);
203 	}
204 
205 	if (in_interrupt())
206 		panic("Fatal exception in interrupt");
207 	if (panic_on_oops)
208 		panic("Fatal exception");
209 	do_exit(signr);
210 }
211 NOKPROBE_SYMBOL(oops_end);
212 
213 static int __die(const char *str, struct pt_regs *regs, long err)
214 {
215 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
216 
217 	if (IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN))
218 		printk("LE ");
219 	else
220 		printk("BE ");
221 
222 	if (IS_ENABLED(CONFIG_PREEMPT))
223 		pr_cont("PREEMPT ");
224 
225 	if (IS_ENABLED(CONFIG_SMP))
226 		pr_cont("SMP NR_CPUS=%d ", NR_CPUS);
227 
228 	if (debug_pagealloc_enabled())
229 		pr_cont("DEBUG_PAGEALLOC ");
230 
231 	if (IS_ENABLED(CONFIG_NUMA))
232 		pr_cont("NUMA ");
233 
234 	pr_cont("%s\n", ppc_md.name ? ppc_md.name : "");
235 
236 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
237 		return 1;
238 
239 	print_modules();
240 	show_regs(regs);
241 
242 	return 0;
243 }
244 NOKPROBE_SYMBOL(__die);
245 
246 void die(const char *str, struct pt_regs *regs, long err)
247 {
248 	unsigned long flags;
249 
250 	if (debugger(regs))
251 		return;
252 
253 	flags = oops_begin(regs);
254 	if (__die(str, regs, err))
255 		err = 0;
256 	oops_end(flags, regs, err);
257 }
258 NOKPROBE_SYMBOL(die);
259 
260 void user_single_step_siginfo(struct task_struct *tsk,
261 				struct pt_regs *regs, siginfo_t *info)
262 {
263 	memset(info, 0, sizeof(*info));
264 	info->si_signo = SIGTRAP;
265 	info->si_code = TRAP_TRACE;
266 	info->si_addr = (void __user *)regs->nip;
267 }
268 
269 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
270 {
271 	siginfo_t info;
272 	const char fmt32[] = KERN_INFO "%s[%d]: unhandled signal %d " \
273 			"at %08lx nip %08lx lr %08lx code %x\n";
274 	const char fmt64[] = KERN_INFO "%s[%d]: unhandled signal %d " \
275 			"at %016lx nip %016lx lr %016lx code %x\n";
276 
277 	if (!user_mode(regs)) {
278 		die("Exception in kernel mode", regs, signr);
279 		return;
280 	}
281 
282 	if (show_unhandled_signals && unhandled_signal(current, signr)) {
283 		printk_ratelimited(regs->msr & MSR_64BIT ? fmt64 : fmt32,
284 				   current->comm, current->pid, signr,
285 				   addr, regs->nip, regs->link, code);
286 	}
287 
288 	if (arch_irqs_disabled() && !arch_irq_disabled_regs(regs))
289 		local_irq_enable();
290 
291 	current->thread.trap_nr = code;
292 	memset(&info, 0, sizeof(info));
293 	info.si_signo = signr;
294 	info.si_code = code;
295 	info.si_addr = (void __user *) addr;
296 	force_sig_info(signr, &info, current);
297 }
298 
299 void system_reset_exception(struct pt_regs *regs)
300 {
301 	/*
302 	 * Avoid crashes in case of nested NMI exceptions. Recoverability
303 	 * is determined by RI and in_nmi
304 	 */
305 	bool nested = in_nmi();
306 	if (!nested)
307 		nmi_enter();
308 
309 	__this_cpu_inc(irq_stat.sreset_irqs);
310 
311 	/* See if any machine dependent calls */
312 	if (ppc_md.system_reset_exception) {
313 		if (ppc_md.system_reset_exception(regs))
314 			goto out;
315 	}
316 
317 	if (debugger(regs))
318 		goto out;
319 
320 	/*
321 	 * A system reset is a request to dump, so we always send
322 	 * it through the crashdump code (if fadump or kdump are
323 	 * registered).
324 	 */
325 	crash_fadump(regs, "System Reset");
326 
327 	crash_kexec(regs);
328 
329 	/*
330 	 * We aren't the primary crash CPU. We need to send it
331 	 * to a holding pattern to avoid it ending up in the panic
332 	 * code.
333 	 */
334 	crash_kexec_secondary(regs);
335 
336 	/*
337 	 * No debugger or crash dump registered, print logs then
338 	 * panic.
339 	 */
340 	__die("System Reset", regs, SIGABRT);
341 
342 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
343 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
344 	nmi_panic(regs, "System Reset");
345 
346 out:
347 #ifdef CONFIG_PPC_BOOK3S_64
348 	BUG_ON(get_paca()->in_nmi == 0);
349 	if (get_paca()->in_nmi > 1)
350 		nmi_panic(regs, "Unrecoverable nested System Reset");
351 #endif
352 	/* Must die if the interrupt is not recoverable */
353 	if (!(regs->msr & MSR_RI))
354 		nmi_panic(regs, "Unrecoverable System Reset");
355 
356 	if (!nested)
357 		nmi_exit();
358 
359 	/* What should we do here? We could issue a shutdown or hard reset. */
360 }
361 
362 /*
363  * I/O accesses can cause machine checks on powermacs.
364  * Check if the NIP corresponds to the address of a sync
365  * instruction for which there is an entry in the exception
366  * table.
367  * Note that the 601 only takes a machine check on TEA
368  * (transfer error ack) signal assertion, and does not
369  * set any of the top 16 bits of SRR1.
370  *  -- paulus.
371  */
372 static inline int check_io_access(struct pt_regs *regs)
373 {
374 #ifdef CONFIG_PPC32
375 	unsigned long msr = regs->msr;
376 	const struct exception_table_entry *entry;
377 	unsigned int *nip = (unsigned int *)regs->nip;
378 
379 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
380 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
381 		/*
382 		 * Check that it's a sync instruction, or somewhere
383 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
384 		 * As the address is in the exception table
385 		 * we should be able to read the instr there.
386 		 * For the debug message, we look at the preceding
387 		 * load or store.
388 		 */
389 		if (*nip == PPC_INST_NOP)
390 			nip -= 2;
391 		else if (*nip == PPC_INST_ISYNC)
392 			--nip;
393 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
394 			unsigned int rb;
395 
396 			--nip;
397 			rb = (*nip >> 11) & 0x1f;
398 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
399 			       (*nip & 0x100)? "OUT to": "IN from",
400 			       regs->gpr[rb] - _IO_BASE, nip);
401 			regs->msr |= MSR_RI;
402 			regs->nip = extable_fixup(entry);
403 			return 1;
404 		}
405 	}
406 #endif /* CONFIG_PPC32 */
407 	return 0;
408 }
409 
410 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
411 /* On 4xx, the reason for the machine check or program exception
412    is in the ESR. */
413 #define get_reason(regs)	((regs)->dsisr)
414 #define REASON_FP		ESR_FP
415 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
416 #define REASON_PRIVILEGED	ESR_PPR
417 #define REASON_TRAP		ESR_PTR
418 
419 /* single-step stuff */
420 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
421 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
422 
423 #else
424 /* On non-4xx, the reason for the machine check or program
425    exception is in the MSR. */
426 #define get_reason(regs)	((regs)->msr)
427 #define REASON_TM		SRR1_PROGTM
428 #define REASON_FP		SRR1_PROGFPE
429 #define REASON_ILLEGAL		SRR1_PROGILL
430 #define REASON_PRIVILEGED	SRR1_PROGPRIV
431 #define REASON_TRAP		SRR1_PROGTRAP
432 
433 #define single_stepping(regs)	((regs)->msr & MSR_SE)
434 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
435 #endif
436 
437 #if defined(CONFIG_E500)
438 int machine_check_e500mc(struct pt_regs *regs)
439 {
440 	unsigned long mcsr = mfspr(SPRN_MCSR);
441 	unsigned long pvr = mfspr(SPRN_PVR);
442 	unsigned long reason = mcsr;
443 	int recoverable = 1;
444 
445 	if (reason & MCSR_LD) {
446 		recoverable = fsl_rio_mcheck_exception(regs);
447 		if (recoverable == 1)
448 			goto silent_out;
449 	}
450 
451 	printk("Machine check in kernel mode.\n");
452 	printk("Caused by (from MCSR=%lx): ", reason);
453 
454 	if (reason & MCSR_MCP)
455 		printk("Machine Check Signal\n");
456 
457 	if (reason & MCSR_ICPERR) {
458 		printk("Instruction Cache Parity Error\n");
459 
460 		/*
461 		 * This is recoverable by invalidating the i-cache.
462 		 */
463 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
464 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
465 			;
466 
467 		/*
468 		 * This will generally be accompanied by an instruction
469 		 * fetch error report -- only treat MCSR_IF as fatal
470 		 * if it wasn't due to an L1 parity error.
471 		 */
472 		reason &= ~MCSR_IF;
473 	}
474 
475 	if (reason & MCSR_DCPERR_MC) {
476 		printk("Data Cache Parity Error\n");
477 
478 		/*
479 		 * In write shadow mode we auto-recover from the error, but it
480 		 * may still get logged and cause a machine check.  We should
481 		 * only treat the non-write shadow case as non-recoverable.
482 		 */
483 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
484 		 * is not implemented but L1 data cache always runs in write
485 		 * shadow mode. Hence on data cache parity errors HW will
486 		 * automatically invalidate the L1 Data Cache.
487 		 */
488 		if (PVR_VER(pvr) != PVR_VER_E6500) {
489 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
490 				recoverable = 0;
491 		}
492 	}
493 
494 	if (reason & MCSR_L2MMU_MHIT) {
495 		printk("Hit on multiple TLB entries\n");
496 		recoverable = 0;
497 	}
498 
499 	if (reason & MCSR_NMI)
500 		printk("Non-maskable interrupt\n");
501 
502 	if (reason & MCSR_IF) {
503 		printk("Instruction Fetch Error Report\n");
504 		recoverable = 0;
505 	}
506 
507 	if (reason & MCSR_LD) {
508 		printk("Load Error Report\n");
509 		recoverable = 0;
510 	}
511 
512 	if (reason & MCSR_ST) {
513 		printk("Store Error Report\n");
514 		recoverable = 0;
515 	}
516 
517 	if (reason & MCSR_LDG) {
518 		printk("Guarded Load Error Report\n");
519 		recoverable = 0;
520 	}
521 
522 	if (reason & MCSR_TLBSYNC)
523 		printk("Simultaneous tlbsync operations\n");
524 
525 	if (reason & MCSR_BSL2_ERR) {
526 		printk("Level 2 Cache Error\n");
527 		recoverable = 0;
528 	}
529 
530 	if (reason & MCSR_MAV) {
531 		u64 addr;
532 
533 		addr = mfspr(SPRN_MCAR);
534 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
535 
536 		printk("Machine Check %s Address: %#llx\n",
537 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
538 	}
539 
540 silent_out:
541 	mtspr(SPRN_MCSR, mcsr);
542 	return mfspr(SPRN_MCSR) == 0 && recoverable;
543 }
544 
545 int machine_check_e500(struct pt_regs *regs)
546 {
547 	unsigned long reason = mfspr(SPRN_MCSR);
548 
549 	if (reason & MCSR_BUS_RBERR) {
550 		if (fsl_rio_mcheck_exception(regs))
551 			return 1;
552 		if (fsl_pci_mcheck_exception(regs))
553 			return 1;
554 	}
555 
556 	printk("Machine check in kernel mode.\n");
557 	printk("Caused by (from MCSR=%lx): ", reason);
558 
559 	if (reason & MCSR_MCP)
560 		printk("Machine Check Signal\n");
561 	if (reason & MCSR_ICPERR)
562 		printk("Instruction Cache Parity Error\n");
563 	if (reason & MCSR_DCP_PERR)
564 		printk("Data Cache Push Parity Error\n");
565 	if (reason & MCSR_DCPERR)
566 		printk("Data Cache Parity Error\n");
567 	if (reason & MCSR_BUS_IAERR)
568 		printk("Bus - Instruction Address Error\n");
569 	if (reason & MCSR_BUS_RAERR)
570 		printk("Bus - Read Address Error\n");
571 	if (reason & MCSR_BUS_WAERR)
572 		printk("Bus - Write Address Error\n");
573 	if (reason & MCSR_BUS_IBERR)
574 		printk("Bus - Instruction Data Error\n");
575 	if (reason & MCSR_BUS_RBERR)
576 		printk("Bus - Read Data Bus Error\n");
577 	if (reason & MCSR_BUS_WBERR)
578 		printk("Bus - Write Data Bus Error\n");
579 	if (reason & MCSR_BUS_IPERR)
580 		printk("Bus - Instruction Parity Error\n");
581 	if (reason & MCSR_BUS_RPERR)
582 		printk("Bus - Read Parity Error\n");
583 
584 	return 0;
585 }
586 
587 int machine_check_generic(struct pt_regs *regs)
588 {
589 	return 0;
590 }
591 #elif defined(CONFIG_E200)
592 int machine_check_e200(struct pt_regs *regs)
593 {
594 	unsigned long reason = mfspr(SPRN_MCSR);
595 
596 	printk("Machine check in kernel mode.\n");
597 	printk("Caused by (from MCSR=%lx): ", reason);
598 
599 	if (reason & MCSR_MCP)
600 		printk("Machine Check Signal\n");
601 	if (reason & MCSR_CP_PERR)
602 		printk("Cache Push Parity Error\n");
603 	if (reason & MCSR_CPERR)
604 		printk("Cache Parity Error\n");
605 	if (reason & MCSR_EXCP_ERR)
606 		printk("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
607 	if (reason & MCSR_BUS_IRERR)
608 		printk("Bus - Read Bus Error on instruction fetch\n");
609 	if (reason & MCSR_BUS_DRERR)
610 		printk("Bus - Read Bus Error on data load\n");
611 	if (reason & MCSR_BUS_WRERR)
612 		printk("Bus - Write Bus Error on buffered store or cache line push\n");
613 
614 	return 0;
615 }
616 #elif defined(CONFIG_PPC32)
617 int machine_check_generic(struct pt_regs *regs)
618 {
619 	unsigned long reason = regs->msr;
620 
621 	printk("Machine check in kernel mode.\n");
622 	printk("Caused by (from SRR1=%lx): ", reason);
623 	switch (reason & 0x601F0000) {
624 	case 0x80000:
625 		printk("Machine check signal\n");
626 		break;
627 	case 0:		/* for 601 */
628 	case 0x40000:
629 	case 0x140000:	/* 7450 MSS error and TEA */
630 		printk("Transfer error ack signal\n");
631 		break;
632 	case 0x20000:
633 		printk("Data parity error signal\n");
634 		break;
635 	case 0x10000:
636 		printk("Address parity error signal\n");
637 		break;
638 	case 0x20000000:
639 		printk("L1 Data Cache error\n");
640 		break;
641 	case 0x40000000:
642 		printk("L1 Instruction Cache error\n");
643 		break;
644 	case 0x00100000:
645 		printk("L2 data cache parity error\n");
646 		break;
647 	default:
648 		printk("Unknown values in msr\n");
649 	}
650 	return 0;
651 }
652 #endif /* everything else */
653 
654 void machine_check_exception(struct pt_regs *regs)
655 {
656 	int recover = 0;
657 	bool nested = in_nmi();
658 	if (!nested)
659 		nmi_enter();
660 
661 	/* 64s accounts the mce in machine_check_early when in HVMODE */
662 	if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64) || !cpu_has_feature(CPU_FTR_HVMODE))
663 		__this_cpu_inc(irq_stat.mce_exceptions);
664 
665 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
666 
667 	/* See if any machine dependent calls. In theory, we would want
668 	 * to call the CPU first, and call the ppc_md. one if the CPU
669 	 * one returns a positive number. However there is existing code
670 	 * that assumes the board gets a first chance, so let's keep it
671 	 * that way for now and fix things later. --BenH.
672 	 */
673 	if (ppc_md.machine_check_exception)
674 		recover = ppc_md.machine_check_exception(regs);
675 	else if (cur_cpu_spec->machine_check)
676 		recover = cur_cpu_spec->machine_check(regs);
677 
678 	if (recover > 0)
679 		goto bail;
680 
681 	if (debugger_fault_handler(regs))
682 		goto bail;
683 
684 	if (check_io_access(regs))
685 		goto bail;
686 
687 	die("Machine check", regs, SIGBUS);
688 
689 	/* Must die if the interrupt is not recoverable */
690 	if (!(regs->msr & MSR_RI))
691 		nmi_panic(regs, "Unrecoverable Machine check");
692 
693 bail:
694 	if (!nested)
695 		nmi_exit();
696 }
697 
698 void SMIException(struct pt_regs *regs)
699 {
700 	die("System Management Interrupt", regs, SIGABRT);
701 }
702 
703 #ifdef CONFIG_VSX
704 static void p9_hmi_special_emu(struct pt_regs *regs)
705 {
706 	unsigned int ra, rb, t, i, sel, instr, rc;
707 	const void __user *addr;
708 	u8 vbuf[16], *vdst;
709 	unsigned long ea, msr, msr_mask;
710 	bool swap;
711 
712 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
713 		return;
714 
715 	/*
716 	 * lxvb16x	opcode: 0x7c0006d8
717 	 * lxvd2x	opcode: 0x7c000698
718 	 * lxvh8x	opcode: 0x7c000658
719 	 * lxvw4x	opcode: 0x7c000618
720 	 */
721 	if ((instr & 0xfc00073e) != 0x7c000618) {
722 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
723 			 " instr=%08x\n",
724 			 smp_processor_id(), current->comm, current->pid,
725 			 regs->nip, instr);
726 		return;
727 	}
728 
729 	/* Grab vector registers into the task struct */
730 	msr = regs->msr; /* Grab msr before we flush the bits */
731 	flush_vsx_to_thread(current);
732 	enable_kernel_altivec();
733 
734 	/*
735 	 * Is userspace running with a different endian (this is rare but
736 	 * not impossible)
737 	 */
738 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
739 
740 	/* Decode the instruction */
741 	ra = (instr >> 16) & 0x1f;
742 	rb = (instr >> 11) & 0x1f;
743 	t = (instr >> 21) & 0x1f;
744 	if (instr & 1)
745 		vdst = (u8 *)&current->thread.vr_state.vr[t];
746 	else
747 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
748 
749 	/* Grab the vector address */
750 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
751 	if (is_32bit_task())
752 		ea &= 0xfffffffful;
753 	addr = (__force const void __user *)ea;
754 
755 	/* Check it */
756 	if (!access_ok(VERIFY_READ, addr, 16)) {
757 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
758 			 " instr=%08x addr=%016lx\n",
759 			 smp_processor_id(), current->comm, current->pid,
760 			 regs->nip, instr, (unsigned long)addr);
761 		return;
762 	}
763 
764 	/* Read the vector */
765 	rc = 0;
766 	if ((unsigned long)addr & 0xfUL)
767 		/* unaligned case */
768 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
769 	else
770 		__get_user_atomic_128_aligned(vbuf, addr, rc);
771 	if (rc) {
772 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
773 			 " instr=%08x addr=%016lx\n",
774 			 smp_processor_id(), current->comm, current->pid,
775 			 regs->nip, instr, (unsigned long)addr);
776 		return;
777 	}
778 
779 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
780 		 " instr=%08x addr=%016lx\n",
781 		 smp_processor_id(), current->comm, current->pid, regs->nip,
782 		 instr, (unsigned long) addr);
783 
784 	/* Grab instruction "selector" */
785 	sel = (instr >> 6) & 3;
786 
787 	/*
788 	 * Check to make sure the facility is actually enabled. This
789 	 * could happen if we get a false positive hit.
790 	 *
791 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
792 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
793 	 */
794 	msr_mask = MSR_VSX;
795 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
796 		msr_mask = MSR_VEC;
797 	if (!(msr & msr_mask)) {
798 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
799 			 " instr=%08x msr:%016lx\n",
800 			 smp_processor_id(), current->comm, current->pid,
801 			 regs->nip, instr, msr);
802 		return;
803 	}
804 
805 	/* Do logging here before we modify sel based on endian */
806 	switch (sel) {
807 	case 0:	/* lxvw4x */
808 		PPC_WARN_EMULATED(lxvw4x, regs);
809 		break;
810 	case 1: /* lxvh8x */
811 		PPC_WARN_EMULATED(lxvh8x, regs);
812 		break;
813 	case 2: /* lxvd2x */
814 		PPC_WARN_EMULATED(lxvd2x, regs);
815 		break;
816 	case 3: /* lxvb16x */
817 		PPC_WARN_EMULATED(lxvb16x, regs);
818 		break;
819 	}
820 
821 #ifdef __LITTLE_ENDIAN__
822 	/*
823 	 * An LE kernel stores the vector in the task struct as an LE
824 	 * byte array (effectively swapping both the components and
825 	 * the content of the components). Those instructions expect
826 	 * the components to remain in ascending address order, so we
827 	 * swap them back.
828 	 *
829 	 * If we are running a BE user space, the expectation is that
830 	 * of a simple memcpy, so forcing the emulation to look like
831 	 * a lxvb16x should do the trick.
832 	 */
833 	if (swap)
834 		sel = 3;
835 
836 	switch (sel) {
837 	case 0:	/* lxvw4x */
838 		for (i = 0; i < 4; i++)
839 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
840 		break;
841 	case 1: /* lxvh8x */
842 		for (i = 0; i < 8; i++)
843 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
844 		break;
845 	case 2: /* lxvd2x */
846 		for (i = 0; i < 2; i++)
847 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
848 		break;
849 	case 3: /* lxvb16x */
850 		for (i = 0; i < 16; i++)
851 			vdst[i] = vbuf[15-i];
852 		break;
853 	}
854 #else /* __LITTLE_ENDIAN__ */
855 	/* On a big endian kernel, a BE userspace only needs a memcpy */
856 	if (!swap)
857 		sel = 3;
858 
859 	/* Otherwise, we need to swap the content of the components */
860 	switch (sel) {
861 	case 0:	/* lxvw4x */
862 		for (i = 0; i < 4; i++)
863 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
864 		break;
865 	case 1: /* lxvh8x */
866 		for (i = 0; i < 8; i++)
867 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
868 		break;
869 	case 2: /* lxvd2x */
870 		for (i = 0; i < 2; i++)
871 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
872 		break;
873 	case 3: /* lxvb16x */
874 		memcpy(vdst, vbuf, 16);
875 		break;
876 	}
877 #endif /* !__LITTLE_ENDIAN__ */
878 
879 	/* Go to next instruction */
880 	regs->nip += 4;
881 }
882 #endif /* CONFIG_VSX */
883 
884 void handle_hmi_exception(struct pt_regs *regs)
885 {
886 	struct pt_regs *old_regs;
887 
888 	old_regs = set_irq_regs(regs);
889 	irq_enter();
890 
891 #ifdef CONFIG_VSX
892 	/* Real mode flagged P9 special emu is needed */
893 	if (local_paca->hmi_p9_special_emu) {
894 		local_paca->hmi_p9_special_emu = 0;
895 
896 		/*
897 		 * We don't want to take page faults while doing the
898 		 * emulation, we just replay the instruction if necessary.
899 		 */
900 		pagefault_disable();
901 		p9_hmi_special_emu(regs);
902 		pagefault_enable();
903 	}
904 #endif /* CONFIG_VSX */
905 
906 	if (ppc_md.handle_hmi_exception)
907 		ppc_md.handle_hmi_exception(regs);
908 
909 	irq_exit();
910 	set_irq_regs(old_regs);
911 }
912 
913 void unknown_exception(struct pt_regs *regs)
914 {
915 	enum ctx_state prev_state = exception_enter();
916 
917 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
918 	       regs->nip, regs->msr, regs->trap);
919 
920 	_exception(SIGTRAP, regs, 0, 0);
921 
922 	exception_exit(prev_state);
923 }
924 
925 void instruction_breakpoint_exception(struct pt_regs *regs)
926 {
927 	enum ctx_state prev_state = exception_enter();
928 
929 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
930 					5, SIGTRAP) == NOTIFY_STOP)
931 		goto bail;
932 	if (debugger_iabr_match(regs))
933 		goto bail;
934 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
935 
936 bail:
937 	exception_exit(prev_state);
938 }
939 
940 void RunModeException(struct pt_regs *regs)
941 {
942 	_exception(SIGTRAP, regs, 0, 0);
943 }
944 
945 void single_step_exception(struct pt_regs *regs)
946 {
947 	enum ctx_state prev_state = exception_enter();
948 
949 	clear_single_step(regs);
950 
951 	if (kprobe_post_handler(regs))
952 		return;
953 
954 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
955 					5, SIGTRAP) == NOTIFY_STOP)
956 		goto bail;
957 	if (debugger_sstep(regs))
958 		goto bail;
959 
960 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
961 
962 bail:
963 	exception_exit(prev_state);
964 }
965 NOKPROBE_SYMBOL(single_step_exception);
966 
967 /*
968  * After we have successfully emulated an instruction, we have to
969  * check if the instruction was being single-stepped, and if so,
970  * pretend we got a single-step exception.  This was pointed out
971  * by Kumar Gala.  -- paulus
972  */
973 static void emulate_single_step(struct pt_regs *regs)
974 {
975 	if (single_stepping(regs))
976 		single_step_exception(regs);
977 }
978 
979 static inline int __parse_fpscr(unsigned long fpscr)
980 {
981 	int ret = 0;
982 
983 	/* Invalid operation */
984 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
985 		ret = FPE_FLTINV;
986 
987 	/* Overflow */
988 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
989 		ret = FPE_FLTOVF;
990 
991 	/* Underflow */
992 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
993 		ret = FPE_FLTUND;
994 
995 	/* Divide by zero */
996 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
997 		ret = FPE_FLTDIV;
998 
999 	/* Inexact result */
1000 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1001 		ret = FPE_FLTRES;
1002 
1003 	return ret;
1004 }
1005 
1006 static void parse_fpe(struct pt_regs *regs)
1007 {
1008 	int code = 0;
1009 
1010 	flush_fp_to_thread(current);
1011 
1012 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1013 
1014 	_exception(SIGFPE, regs, code, regs->nip);
1015 }
1016 
1017 /*
1018  * Illegal instruction emulation support.  Originally written to
1019  * provide the PVR to user applications using the mfspr rd, PVR.
1020  * Return non-zero if we can't emulate, or -EFAULT if the associated
1021  * memory access caused an access fault.  Return zero on success.
1022  *
1023  * There are a couple of ways to do this, either "decode" the instruction
1024  * or directly match lots of bits.  In this case, matching lots of
1025  * bits is faster and easier.
1026  *
1027  */
1028 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1029 {
1030 	u8 rT = (instword >> 21) & 0x1f;
1031 	u8 rA = (instword >> 16) & 0x1f;
1032 	u8 NB_RB = (instword >> 11) & 0x1f;
1033 	u32 num_bytes;
1034 	unsigned long EA;
1035 	int pos = 0;
1036 
1037 	/* Early out if we are an invalid form of lswx */
1038 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1039 		if ((rT == rA) || (rT == NB_RB))
1040 			return -EINVAL;
1041 
1042 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1043 
1044 	switch (instword & PPC_INST_STRING_MASK) {
1045 		case PPC_INST_LSWX:
1046 		case PPC_INST_STSWX:
1047 			EA += NB_RB;
1048 			num_bytes = regs->xer & 0x7f;
1049 			break;
1050 		case PPC_INST_LSWI:
1051 		case PPC_INST_STSWI:
1052 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1053 			break;
1054 		default:
1055 			return -EINVAL;
1056 	}
1057 
1058 	while (num_bytes != 0)
1059 	{
1060 		u8 val;
1061 		u32 shift = 8 * (3 - (pos & 0x3));
1062 
1063 		/* if process is 32-bit, clear upper 32 bits of EA */
1064 		if ((regs->msr & MSR_64BIT) == 0)
1065 			EA &= 0xFFFFFFFF;
1066 
1067 		switch ((instword & PPC_INST_STRING_MASK)) {
1068 			case PPC_INST_LSWX:
1069 			case PPC_INST_LSWI:
1070 				if (get_user(val, (u8 __user *)EA))
1071 					return -EFAULT;
1072 				/* first time updating this reg,
1073 				 * zero it out */
1074 				if (pos == 0)
1075 					regs->gpr[rT] = 0;
1076 				regs->gpr[rT] |= val << shift;
1077 				break;
1078 			case PPC_INST_STSWI:
1079 			case PPC_INST_STSWX:
1080 				val = regs->gpr[rT] >> shift;
1081 				if (put_user(val, (u8 __user *)EA))
1082 					return -EFAULT;
1083 				break;
1084 		}
1085 		/* move EA to next address */
1086 		EA += 1;
1087 		num_bytes--;
1088 
1089 		/* manage our position within the register */
1090 		if (++pos == 4) {
1091 			pos = 0;
1092 			if (++rT == 32)
1093 				rT = 0;
1094 		}
1095 	}
1096 
1097 	return 0;
1098 }
1099 
1100 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1101 {
1102 	u32 ra,rs;
1103 	unsigned long tmp;
1104 
1105 	ra = (instword >> 16) & 0x1f;
1106 	rs = (instword >> 21) & 0x1f;
1107 
1108 	tmp = regs->gpr[rs];
1109 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1110 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1111 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1112 	regs->gpr[ra] = tmp;
1113 
1114 	return 0;
1115 }
1116 
1117 static int emulate_isel(struct pt_regs *regs, u32 instword)
1118 {
1119 	u8 rT = (instword >> 21) & 0x1f;
1120 	u8 rA = (instword >> 16) & 0x1f;
1121 	u8 rB = (instword >> 11) & 0x1f;
1122 	u8 BC = (instword >> 6) & 0x1f;
1123 	u8 bit;
1124 	unsigned long tmp;
1125 
1126 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1127 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1128 
1129 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1130 
1131 	return 0;
1132 }
1133 
1134 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1135 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1136 {
1137         /* If we're emulating a load/store in an active transaction, we cannot
1138          * emulate it as the kernel operates in transaction suspended context.
1139          * We need to abort the transaction.  This creates a persistent TM
1140          * abort so tell the user what caused it with a new code.
1141 	 */
1142 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1143 		tm_enable();
1144 		tm_abort(cause);
1145 		return true;
1146 	}
1147 	return false;
1148 }
1149 #else
1150 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1151 {
1152 	return false;
1153 }
1154 #endif
1155 
1156 static int emulate_instruction(struct pt_regs *regs)
1157 {
1158 	u32 instword;
1159 	u32 rd;
1160 
1161 	if (!user_mode(regs))
1162 		return -EINVAL;
1163 	CHECK_FULL_REGS(regs);
1164 
1165 	if (get_user(instword, (u32 __user *)(regs->nip)))
1166 		return -EFAULT;
1167 
1168 	/* Emulate the mfspr rD, PVR. */
1169 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1170 		PPC_WARN_EMULATED(mfpvr, regs);
1171 		rd = (instword >> 21) & 0x1f;
1172 		regs->gpr[rd] = mfspr(SPRN_PVR);
1173 		return 0;
1174 	}
1175 
1176 	/* Emulating the dcba insn is just a no-op.  */
1177 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1178 		PPC_WARN_EMULATED(dcba, regs);
1179 		return 0;
1180 	}
1181 
1182 	/* Emulate the mcrxr insn.  */
1183 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1184 		int shift = (instword >> 21) & 0x1c;
1185 		unsigned long msk = 0xf0000000UL >> shift;
1186 
1187 		PPC_WARN_EMULATED(mcrxr, regs);
1188 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1189 		regs->xer &= ~0xf0000000UL;
1190 		return 0;
1191 	}
1192 
1193 	/* Emulate load/store string insn. */
1194 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1195 		if (tm_abort_check(regs,
1196 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1197 			return -EINVAL;
1198 		PPC_WARN_EMULATED(string, regs);
1199 		return emulate_string_inst(regs, instword);
1200 	}
1201 
1202 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1203 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1204 		PPC_WARN_EMULATED(popcntb, regs);
1205 		return emulate_popcntb_inst(regs, instword);
1206 	}
1207 
1208 	/* Emulate isel (Integer Select) instruction */
1209 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1210 		PPC_WARN_EMULATED(isel, regs);
1211 		return emulate_isel(regs, instword);
1212 	}
1213 
1214 	/* Emulate sync instruction variants */
1215 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1216 		PPC_WARN_EMULATED(sync, regs);
1217 		asm volatile("sync");
1218 		return 0;
1219 	}
1220 
1221 #ifdef CONFIG_PPC64
1222 	/* Emulate the mfspr rD, DSCR. */
1223 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1224 		PPC_INST_MFSPR_DSCR_USER) ||
1225 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1226 		PPC_INST_MFSPR_DSCR)) &&
1227 			cpu_has_feature(CPU_FTR_DSCR)) {
1228 		PPC_WARN_EMULATED(mfdscr, regs);
1229 		rd = (instword >> 21) & 0x1f;
1230 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1231 		return 0;
1232 	}
1233 	/* Emulate the mtspr DSCR, rD. */
1234 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1235 		PPC_INST_MTSPR_DSCR_USER) ||
1236 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1237 		PPC_INST_MTSPR_DSCR)) &&
1238 			cpu_has_feature(CPU_FTR_DSCR)) {
1239 		PPC_WARN_EMULATED(mtdscr, regs);
1240 		rd = (instword >> 21) & 0x1f;
1241 		current->thread.dscr = regs->gpr[rd];
1242 		current->thread.dscr_inherit = 1;
1243 		mtspr(SPRN_DSCR, current->thread.dscr);
1244 		return 0;
1245 	}
1246 #endif
1247 
1248 	return -EINVAL;
1249 }
1250 
1251 int is_valid_bugaddr(unsigned long addr)
1252 {
1253 	return is_kernel_addr(addr);
1254 }
1255 
1256 #ifdef CONFIG_MATH_EMULATION
1257 static int emulate_math(struct pt_regs *regs)
1258 {
1259 	int ret;
1260 	extern int do_mathemu(struct pt_regs *regs);
1261 
1262 	ret = do_mathemu(regs);
1263 	if (ret >= 0)
1264 		PPC_WARN_EMULATED(math, regs);
1265 
1266 	switch (ret) {
1267 	case 0:
1268 		emulate_single_step(regs);
1269 		return 0;
1270 	case 1: {
1271 			int code = 0;
1272 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1273 			_exception(SIGFPE, regs, code, regs->nip);
1274 			return 0;
1275 		}
1276 	case -EFAULT:
1277 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1278 		return 0;
1279 	}
1280 
1281 	return -1;
1282 }
1283 #else
1284 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1285 #endif
1286 
1287 void program_check_exception(struct pt_regs *regs)
1288 {
1289 	enum ctx_state prev_state = exception_enter();
1290 	unsigned int reason = get_reason(regs);
1291 
1292 	/* We can now get here via a FP Unavailable exception if the core
1293 	 * has no FPU, in that case the reason flags will be 0 */
1294 
1295 	if (reason & REASON_FP) {
1296 		/* IEEE FP exception */
1297 		parse_fpe(regs);
1298 		goto bail;
1299 	}
1300 	if (reason & REASON_TRAP) {
1301 		unsigned long bugaddr;
1302 		/* Debugger is first in line to stop recursive faults in
1303 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1304 		if (debugger_bpt(regs))
1305 			goto bail;
1306 
1307 		if (kprobe_handler(regs))
1308 			goto bail;
1309 
1310 		/* trap exception */
1311 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1312 				== NOTIFY_STOP)
1313 			goto bail;
1314 
1315 		bugaddr = regs->nip;
1316 		/*
1317 		 * Fixup bugaddr for BUG_ON() in real mode
1318 		 */
1319 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1320 			bugaddr += PAGE_OFFSET;
1321 
1322 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1323 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1324 			regs->nip += 4;
1325 			goto bail;
1326 		}
1327 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1328 		goto bail;
1329 	}
1330 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1331 	if (reason & REASON_TM) {
1332 		/* This is a TM "Bad Thing Exception" program check.
1333 		 * This occurs when:
1334 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1335 		 *    transition in TM states.
1336 		 * -  A trechkpt is attempted when transactional.
1337 		 * -  A treclaim is attempted when non transactional.
1338 		 * -  A tend is illegally attempted.
1339 		 * -  writing a TM SPR when transactional.
1340 		 */
1341 		if (!user_mode(regs) &&
1342 		    report_bug(regs->nip, regs) == BUG_TRAP_TYPE_WARN) {
1343 			regs->nip += 4;
1344 			goto bail;
1345 		}
1346 		/* If usermode caused this, it's done something illegal and
1347 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1348 		 * operand to distinguish from the instruction just being bad
1349 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1350 		 * illegal /placement/ of a valid instruction.
1351 		 */
1352 		if (user_mode(regs)) {
1353 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1354 			goto bail;
1355 		} else {
1356 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1357 			       "at %lx (msr 0x%x)\n", regs->nip, reason);
1358 			die("Unrecoverable exception", regs, SIGABRT);
1359 		}
1360 	}
1361 #endif
1362 
1363 	/*
1364 	 * If we took the program check in the kernel skip down to sending a
1365 	 * SIGILL. The subsequent cases all relate to emulating instructions
1366 	 * which we should only do for userspace. We also do not want to enable
1367 	 * interrupts for kernel faults because that might lead to further
1368 	 * faults, and loose the context of the original exception.
1369 	 */
1370 	if (!user_mode(regs))
1371 		goto sigill;
1372 
1373 	/* We restore the interrupt state now */
1374 	if (!arch_irq_disabled_regs(regs))
1375 		local_irq_enable();
1376 
1377 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1378 	 * but there seems to be a hardware bug on the 405GP (RevD)
1379 	 * that means ESR is sometimes set incorrectly - either to
1380 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1381 	 * hardware people - not sure if it can happen on any illegal
1382 	 * instruction or only on FP instructions, whether there is a
1383 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1384 	 */
1385 	if (!emulate_math(regs))
1386 		goto bail;
1387 
1388 	/* Try to emulate it if we should. */
1389 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1390 		switch (emulate_instruction(regs)) {
1391 		case 0:
1392 			regs->nip += 4;
1393 			emulate_single_step(regs);
1394 			goto bail;
1395 		case -EFAULT:
1396 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1397 			goto bail;
1398 		}
1399 	}
1400 
1401 sigill:
1402 	if (reason & REASON_PRIVILEGED)
1403 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1404 	else
1405 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1406 
1407 bail:
1408 	exception_exit(prev_state);
1409 }
1410 NOKPROBE_SYMBOL(program_check_exception);
1411 
1412 /*
1413  * This occurs when running in hypervisor mode on POWER6 or later
1414  * and an illegal instruction is encountered.
1415  */
1416 void emulation_assist_interrupt(struct pt_regs *regs)
1417 {
1418 	regs->msr |= REASON_ILLEGAL;
1419 	program_check_exception(regs);
1420 }
1421 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1422 
1423 void alignment_exception(struct pt_regs *regs)
1424 {
1425 	enum ctx_state prev_state = exception_enter();
1426 	int sig, code, fixed = 0;
1427 
1428 	/* We restore the interrupt state now */
1429 	if (!arch_irq_disabled_regs(regs))
1430 		local_irq_enable();
1431 
1432 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1433 		goto bail;
1434 
1435 	/* we don't implement logging of alignment exceptions */
1436 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1437 		fixed = fix_alignment(regs);
1438 
1439 	if (fixed == 1) {
1440 		regs->nip += 4;	/* skip over emulated instruction */
1441 		emulate_single_step(regs);
1442 		goto bail;
1443 	}
1444 
1445 	/* Operand address was bad */
1446 	if (fixed == -EFAULT) {
1447 		sig = SIGSEGV;
1448 		code = SEGV_ACCERR;
1449 	} else {
1450 		sig = SIGBUS;
1451 		code = BUS_ADRALN;
1452 	}
1453 	if (user_mode(regs))
1454 		_exception(sig, regs, code, regs->dar);
1455 	else
1456 		bad_page_fault(regs, regs->dar, sig);
1457 
1458 bail:
1459 	exception_exit(prev_state);
1460 }
1461 
1462 void slb_miss_bad_addr(struct pt_regs *regs)
1463 {
1464 	enum ctx_state prev_state = exception_enter();
1465 
1466 	if (user_mode(regs))
1467 		_exception(SIGSEGV, regs, SEGV_BNDERR, regs->dar);
1468 	else
1469 		bad_page_fault(regs, regs->dar, SIGSEGV);
1470 
1471 	exception_exit(prev_state);
1472 }
1473 
1474 void StackOverflow(struct pt_regs *regs)
1475 {
1476 	printk(KERN_CRIT "Kernel stack overflow in process %p, r1=%lx\n",
1477 	       current, regs->gpr[1]);
1478 	debugger(regs);
1479 	show_regs(regs);
1480 	panic("kernel stack overflow");
1481 }
1482 
1483 void nonrecoverable_exception(struct pt_regs *regs)
1484 {
1485 	printk(KERN_ERR "Non-recoverable exception at PC=%lx MSR=%lx\n",
1486 	       regs->nip, regs->msr);
1487 	debugger(regs);
1488 	die("nonrecoverable exception", regs, SIGKILL);
1489 }
1490 
1491 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1492 {
1493 	enum ctx_state prev_state = exception_enter();
1494 
1495 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1496 			  "%lx at %lx\n", regs->trap, regs->nip);
1497 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1498 
1499 	exception_exit(prev_state);
1500 }
1501 
1502 void altivec_unavailable_exception(struct pt_regs *regs)
1503 {
1504 	enum ctx_state prev_state = exception_enter();
1505 
1506 	if (user_mode(regs)) {
1507 		/* A user program has executed an altivec instruction,
1508 		   but this kernel doesn't support altivec. */
1509 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1510 		goto bail;
1511 	}
1512 
1513 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1514 			"%lx at %lx\n", regs->trap, regs->nip);
1515 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1516 
1517 bail:
1518 	exception_exit(prev_state);
1519 }
1520 
1521 void vsx_unavailable_exception(struct pt_regs *regs)
1522 {
1523 	if (user_mode(regs)) {
1524 		/* A user program has executed an vsx instruction,
1525 		   but this kernel doesn't support vsx. */
1526 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1527 		return;
1528 	}
1529 
1530 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1531 			"%lx at %lx\n", regs->trap, regs->nip);
1532 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1533 }
1534 
1535 #ifdef CONFIG_PPC64
1536 static void tm_unavailable(struct pt_regs *regs)
1537 {
1538 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1539 	if (user_mode(regs)) {
1540 		current->thread.load_tm++;
1541 		regs->msr |= MSR_TM;
1542 		tm_enable();
1543 		tm_restore_sprs(&current->thread);
1544 		return;
1545 	}
1546 #endif
1547 	pr_emerg("Unrecoverable TM Unavailable Exception "
1548 			"%lx at %lx\n", regs->trap, regs->nip);
1549 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1550 }
1551 
1552 void facility_unavailable_exception(struct pt_regs *regs)
1553 {
1554 	static char *facility_strings[] = {
1555 		[FSCR_FP_LG] = "FPU",
1556 		[FSCR_VECVSX_LG] = "VMX/VSX",
1557 		[FSCR_DSCR_LG] = "DSCR",
1558 		[FSCR_PM_LG] = "PMU SPRs",
1559 		[FSCR_BHRB_LG] = "BHRB",
1560 		[FSCR_TM_LG] = "TM",
1561 		[FSCR_EBB_LG] = "EBB",
1562 		[FSCR_TAR_LG] = "TAR",
1563 		[FSCR_MSGP_LG] = "MSGP",
1564 		[FSCR_SCV_LG] = "SCV",
1565 	};
1566 	char *facility = "unknown";
1567 	u64 value;
1568 	u32 instword, rd;
1569 	u8 status;
1570 	bool hv;
1571 
1572 	hv = (regs->trap == 0xf80);
1573 	if (hv)
1574 		value = mfspr(SPRN_HFSCR);
1575 	else
1576 		value = mfspr(SPRN_FSCR);
1577 
1578 	status = value >> 56;
1579 	if (status == FSCR_DSCR_LG) {
1580 		/*
1581 		 * User is accessing the DSCR register using the problem
1582 		 * state only SPR number (0x03) either through a mfspr or
1583 		 * a mtspr instruction. If it is a write attempt through
1584 		 * a mtspr, then we set the inherit bit. This also allows
1585 		 * the user to write or read the register directly in the
1586 		 * future by setting via the FSCR DSCR bit. But in case it
1587 		 * is a read DSCR attempt through a mfspr instruction, we
1588 		 * just emulate the instruction instead. This code path will
1589 		 * always emulate all the mfspr instructions till the user
1590 		 * has attempted at least one mtspr instruction. This way it
1591 		 * preserves the same behaviour when the user is accessing
1592 		 * the DSCR through privilege level only SPR number (0x11)
1593 		 * which is emulated through illegal instruction exception.
1594 		 * We always leave HFSCR DSCR set.
1595 		 */
1596 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1597 			pr_err("Failed to fetch the user instruction\n");
1598 			return;
1599 		}
1600 
1601 		/* Write into DSCR (mtspr 0x03, RS) */
1602 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1603 				== PPC_INST_MTSPR_DSCR_USER) {
1604 			rd = (instword >> 21) & 0x1f;
1605 			current->thread.dscr = regs->gpr[rd];
1606 			current->thread.dscr_inherit = 1;
1607 			current->thread.fscr |= FSCR_DSCR;
1608 			mtspr(SPRN_FSCR, current->thread.fscr);
1609 		}
1610 
1611 		/* Read from DSCR (mfspr RT, 0x03) */
1612 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1613 				== PPC_INST_MFSPR_DSCR_USER) {
1614 			if (emulate_instruction(regs)) {
1615 				pr_err("DSCR based mfspr emulation failed\n");
1616 				return;
1617 			}
1618 			regs->nip += 4;
1619 			emulate_single_step(regs);
1620 		}
1621 		return;
1622 	}
1623 
1624 	if (status == FSCR_TM_LG) {
1625 		/*
1626 		 * If we're here then the hardware is TM aware because it
1627 		 * generated an exception with FSRM_TM set.
1628 		 *
1629 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1630 		 * told us not to do TM, or the kernel is not built with TM
1631 		 * support.
1632 		 *
1633 		 * If both of those things are true, then userspace can spam the
1634 		 * console by triggering the printk() below just by continually
1635 		 * doing tbegin (or any TM instruction). So in that case just
1636 		 * send the process a SIGILL immediately.
1637 		 */
1638 		if (!cpu_has_feature(CPU_FTR_TM))
1639 			goto out;
1640 
1641 		tm_unavailable(regs);
1642 		return;
1643 	}
1644 
1645 	if ((hv || status >= 2) &&
1646 	    (status < ARRAY_SIZE(facility_strings)) &&
1647 	    facility_strings[status])
1648 		facility = facility_strings[status];
1649 
1650 	/* We restore the interrupt state now */
1651 	if (!arch_irq_disabled_regs(regs))
1652 		local_irq_enable();
1653 
1654 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1655 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1656 
1657 out:
1658 	if (user_mode(regs)) {
1659 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1660 		return;
1661 	}
1662 
1663 	die("Unexpected facility unavailable exception", regs, SIGABRT);
1664 }
1665 #endif
1666 
1667 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1668 
1669 void fp_unavailable_tm(struct pt_regs *regs)
1670 {
1671 	/* Note:  This does not handle any kind of FP laziness. */
1672 
1673 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1674 		 regs->nip, regs->msr);
1675 
1676         /* We can only have got here if the task started using FP after
1677          * beginning the transaction.  So, the transactional regs are just a
1678          * copy of the checkpointed ones.  But, we still need to recheckpoint
1679          * as we're enabling FP for the process; it will return, abort the
1680          * transaction, and probably retry but now with FP enabled.  So the
1681          * checkpointed FP registers need to be loaded.
1682 	 */
1683 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1684 	/* Reclaim didn't save out any FPRs to transact_fprs. */
1685 
1686 	/* Enable FP for the task: */
1687 	regs->msr |= (MSR_FP | current->thread.fpexc_mode);
1688 
1689 	/* This loads and recheckpoints the FP registers from
1690 	 * thread.fpr[].  They will remain in registers after the
1691 	 * checkpoint so we don't need to reload them after.
1692 	 * If VMX is in use, the VRs now hold checkpointed values,
1693 	 * so we don't want to load the VRs from the thread_struct.
1694 	 */
1695 	tm_recheckpoint(&current->thread, MSR_FP);
1696 
1697 	/* If VMX is in use, get the transactional values back */
1698 	if (regs->msr & MSR_VEC) {
1699 		msr_check_and_set(MSR_VEC);
1700 		load_vr_state(&current->thread.vr_state);
1701 		/* At this point all the VSX state is loaded, so enable it */
1702 		regs->msr |= MSR_VSX;
1703 	}
1704 }
1705 
1706 void altivec_unavailable_tm(struct pt_regs *regs)
1707 {
1708 	/* See the comments in fp_unavailable_tm().  This function operates
1709 	 * the same way.
1710 	 */
1711 
1712 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1713 		 "MSR=%lx\n",
1714 		 regs->nip, regs->msr);
1715 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1716 	regs->msr |= MSR_VEC;
1717 	tm_recheckpoint(&current->thread, MSR_VEC);
1718 	current->thread.used_vr = 1;
1719 
1720 	if (regs->msr & MSR_FP) {
1721 		msr_check_and_set(MSR_FP);
1722 		load_fp_state(&current->thread.fp_state);
1723 		regs->msr |= MSR_VSX;
1724 	}
1725 }
1726 
1727 void vsx_unavailable_tm(struct pt_regs *regs)
1728 {
1729 	unsigned long orig_msr = regs->msr;
1730 
1731 	/* See the comments in fp_unavailable_tm().  This works similarly,
1732 	 * though we're loading both FP and VEC registers in here.
1733 	 *
1734 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1735 	 * regs.  Either way, set MSR_VSX.
1736 	 */
1737 
1738 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1739 		 "MSR=%lx\n",
1740 		 regs->nip, regs->msr);
1741 
1742 	current->thread.used_vsr = 1;
1743 
1744 	/* If FP and VMX are already loaded, we have all the state we need */
1745 	if ((orig_msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC)) {
1746 		regs->msr |= MSR_VSX;
1747 		return;
1748 	}
1749 
1750 	/* This reclaims FP and/or VR regs if they're already enabled */
1751 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1752 
1753 	regs->msr |= MSR_VEC | MSR_FP | current->thread.fpexc_mode |
1754 		MSR_VSX;
1755 
1756 	/* This loads & recheckpoints FP and VRs; but we have
1757 	 * to be sure not to overwrite previously-valid state.
1758 	 */
1759 	tm_recheckpoint(&current->thread, regs->msr & ~orig_msr);
1760 
1761 	msr_check_and_set(orig_msr & (MSR_FP | MSR_VEC));
1762 
1763 	if (orig_msr & MSR_FP)
1764 		load_fp_state(&current->thread.fp_state);
1765 	if (orig_msr & MSR_VEC)
1766 		load_vr_state(&current->thread.vr_state);
1767 }
1768 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1769 
1770 void performance_monitor_exception(struct pt_regs *regs)
1771 {
1772 	__this_cpu_inc(irq_stat.pmu_irqs);
1773 
1774 	perf_irq(regs);
1775 }
1776 
1777 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1778 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1779 {
1780 	int changed = 0;
1781 	/*
1782 	 * Determine the cause of the debug event, clear the
1783 	 * event flags and send a trap to the handler. Torez
1784 	 */
1785 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1786 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1787 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1788 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1789 #endif
1790 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status, TRAP_HWBKPT,
1791 			     5);
1792 		changed |= 0x01;
1793 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1794 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1795 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status, TRAP_HWBKPT,
1796 			     6);
1797 		changed |= 0x01;
1798 	}  else if (debug_status & DBSR_IAC1) {
1799 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1800 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1801 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status, TRAP_HWBKPT,
1802 			     1);
1803 		changed |= 0x01;
1804 	}  else if (debug_status & DBSR_IAC2) {
1805 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1806 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status, TRAP_HWBKPT,
1807 			     2);
1808 		changed |= 0x01;
1809 	}  else if (debug_status & DBSR_IAC3) {
1810 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1811 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1812 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status, TRAP_HWBKPT,
1813 			     3);
1814 		changed |= 0x01;
1815 	}  else if (debug_status & DBSR_IAC4) {
1816 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1817 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status, TRAP_HWBKPT,
1818 			     4);
1819 		changed |= 0x01;
1820 	}
1821 	/*
1822 	 * At the point this routine was called, the MSR(DE) was turned off.
1823 	 * Check all other debug flags and see if that bit needs to be turned
1824 	 * back on or not.
1825 	 */
1826 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1827 			       current->thread.debug.dbcr1))
1828 		regs->msr |= MSR_DE;
1829 	else
1830 		/* Make sure the IDM flag is off */
1831 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1832 
1833 	if (changed & 0x01)
1834 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1835 }
1836 
1837 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1838 {
1839 	current->thread.debug.dbsr = debug_status;
1840 
1841 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1842 	 * on server, it stops on the target of the branch. In order to simulate
1843 	 * the server behaviour, we thus restart right away with a single step
1844 	 * instead of stopping here when hitting a BT
1845 	 */
1846 	if (debug_status & DBSR_BT) {
1847 		regs->msr &= ~MSR_DE;
1848 
1849 		/* Disable BT */
1850 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1851 		/* Clear the BT event */
1852 		mtspr(SPRN_DBSR, DBSR_BT);
1853 
1854 		/* Do the single step trick only when coming from userspace */
1855 		if (user_mode(regs)) {
1856 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1857 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1858 			regs->msr |= MSR_DE;
1859 			return;
1860 		}
1861 
1862 		if (kprobe_post_handler(regs))
1863 			return;
1864 
1865 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1866 			       5, SIGTRAP) == NOTIFY_STOP) {
1867 			return;
1868 		}
1869 		if (debugger_sstep(regs))
1870 			return;
1871 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1872 		regs->msr &= ~MSR_DE;
1873 
1874 		/* Disable instruction completion */
1875 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
1876 		/* Clear the instruction completion event */
1877 		mtspr(SPRN_DBSR, DBSR_IC);
1878 
1879 		if (kprobe_post_handler(regs))
1880 			return;
1881 
1882 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1883 			       5, SIGTRAP) == NOTIFY_STOP) {
1884 			return;
1885 		}
1886 
1887 		if (debugger_sstep(regs))
1888 			return;
1889 
1890 		if (user_mode(regs)) {
1891 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
1892 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1893 					       current->thread.debug.dbcr1))
1894 				regs->msr |= MSR_DE;
1895 			else
1896 				/* Make sure the IDM bit is off */
1897 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1898 		}
1899 
1900 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1901 	} else
1902 		handle_debug(regs, debug_status);
1903 }
1904 NOKPROBE_SYMBOL(DebugException);
1905 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
1906 
1907 #if !defined(CONFIG_TAU_INT)
1908 void TAUException(struct pt_regs *regs)
1909 {
1910 	printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
1911 	       regs->nip, regs->msr, regs->trap, print_tainted());
1912 }
1913 #endif /* CONFIG_INT_TAU */
1914 
1915 #ifdef CONFIG_ALTIVEC
1916 void altivec_assist_exception(struct pt_regs *regs)
1917 {
1918 	int err;
1919 
1920 	if (!user_mode(regs)) {
1921 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
1922 		       " at %lx\n", regs->nip);
1923 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
1924 	}
1925 
1926 	flush_altivec_to_thread(current);
1927 
1928 	PPC_WARN_EMULATED(altivec, regs);
1929 	err = emulate_altivec(regs);
1930 	if (err == 0) {
1931 		regs->nip += 4;		/* skip emulated instruction */
1932 		emulate_single_step(regs);
1933 		return;
1934 	}
1935 
1936 	if (err == -EFAULT) {
1937 		/* got an error reading the instruction */
1938 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
1939 	} else {
1940 		/* didn't recognize the instruction */
1941 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
1942 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
1943 				   "in %s at %lx\n", current->comm, regs->nip);
1944 		current->thread.vr_state.vscr.u[3] |= 0x10000;
1945 	}
1946 }
1947 #endif /* CONFIG_ALTIVEC */
1948 
1949 #ifdef CONFIG_FSL_BOOKE
1950 void CacheLockingException(struct pt_regs *regs, unsigned long address,
1951 			   unsigned long error_code)
1952 {
1953 	/* We treat cache locking instructions from the user
1954 	 * as priv ops, in the future we could try to do
1955 	 * something smarter
1956 	 */
1957 	if (error_code & (ESR_DLK|ESR_ILK))
1958 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1959 	return;
1960 }
1961 #endif /* CONFIG_FSL_BOOKE */
1962 
1963 #ifdef CONFIG_SPE
1964 void SPEFloatingPointException(struct pt_regs *regs)
1965 {
1966 	extern int do_spe_mathemu(struct pt_regs *regs);
1967 	unsigned long spefscr;
1968 	int fpexc_mode;
1969 	int code = 0;
1970 	int err;
1971 
1972 	flush_spe_to_thread(current);
1973 
1974 	spefscr = current->thread.spefscr;
1975 	fpexc_mode = current->thread.fpexc_mode;
1976 
1977 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
1978 		code = FPE_FLTOVF;
1979 	}
1980 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
1981 		code = FPE_FLTUND;
1982 	}
1983 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
1984 		code = FPE_FLTDIV;
1985 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
1986 		code = FPE_FLTINV;
1987 	}
1988 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
1989 		code = FPE_FLTRES;
1990 
1991 	err = do_spe_mathemu(regs);
1992 	if (err == 0) {
1993 		regs->nip += 4;		/* skip emulated instruction */
1994 		emulate_single_step(regs);
1995 		return;
1996 	}
1997 
1998 	if (err == -EFAULT) {
1999 		/* got an error reading the instruction */
2000 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2001 	} else if (err == -EINVAL) {
2002 		/* didn't recognize the instruction */
2003 		printk(KERN_ERR "unrecognized spe instruction "
2004 		       "in %s at %lx\n", current->comm, regs->nip);
2005 	} else {
2006 		_exception(SIGFPE, regs, code, regs->nip);
2007 	}
2008 
2009 	return;
2010 }
2011 
2012 void SPEFloatingPointRoundException(struct pt_regs *regs)
2013 {
2014 	extern int speround_handler(struct pt_regs *regs);
2015 	int err;
2016 
2017 	preempt_disable();
2018 	if (regs->msr & MSR_SPE)
2019 		giveup_spe(current);
2020 	preempt_enable();
2021 
2022 	regs->nip -= 4;
2023 	err = speround_handler(regs);
2024 	if (err == 0) {
2025 		regs->nip += 4;		/* skip emulated instruction */
2026 		emulate_single_step(regs);
2027 		return;
2028 	}
2029 
2030 	if (err == -EFAULT) {
2031 		/* got an error reading the instruction */
2032 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2033 	} else if (err == -EINVAL) {
2034 		/* didn't recognize the instruction */
2035 		printk(KERN_ERR "unrecognized spe instruction "
2036 		       "in %s at %lx\n", current->comm, regs->nip);
2037 	} else {
2038 		_exception(SIGFPE, regs, 0, regs->nip);
2039 		return;
2040 	}
2041 }
2042 #endif
2043 
2044 /*
2045  * We enter here if we get an unrecoverable exception, that is, one
2046  * that happened at a point where the RI (recoverable interrupt) bit
2047  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2048  * we therefore lost state by taking this exception.
2049  */
2050 void unrecoverable_exception(struct pt_regs *regs)
2051 {
2052 	printk(KERN_EMERG "Unrecoverable exception %lx at %lx\n",
2053 	       regs->trap, regs->nip);
2054 	die("Unrecoverable exception", regs, SIGABRT);
2055 }
2056 NOKPROBE_SYMBOL(unrecoverable_exception);
2057 
2058 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2059 /*
2060  * Default handler for a Watchdog exception,
2061  * spins until a reboot occurs
2062  */
2063 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2064 {
2065 	/* Generic WatchdogHandler, implement your own */
2066 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2067 	return;
2068 }
2069 
2070 void WatchdogException(struct pt_regs *regs)
2071 {
2072 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2073 	WatchdogHandler(regs);
2074 }
2075 #endif
2076 
2077 /*
2078  * We enter here if we discover during exception entry that we are
2079  * running in supervisor mode with a userspace value in the stack pointer.
2080  */
2081 void kernel_bad_stack(struct pt_regs *regs)
2082 {
2083 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2084 	       regs->gpr[1], regs->nip);
2085 	die("Bad kernel stack pointer", regs, SIGABRT);
2086 }
2087 NOKPROBE_SYMBOL(kernel_bad_stack);
2088 
2089 void __init trap_init(void)
2090 {
2091 }
2092 
2093 
2094 #ifdef CONFIG_PPC_EMULATED_STATS
2095 
2096 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2097 
2098 struct ppc_emulated ppc_emulated = {
2099 #ifdef CONFIG_ALTIVEC
2100 	WARN_EMULATED_SETUP(altivec),
2101 #endif
2102 	WARN_EMULATED_SETUP(dcba),
2103 	WARN_EMULATED_SETUP(dcbz),
2104 	WARN_EMULATED_SETUP(fp_pair),
2105 	WARN_EMULATED_SETUP(isel),
2106 	WARN_EMULATED_SETUP(mcrxr),
2107 	WARN_EMULATED_SETUP(mfpvr),
2108 	WARN_EMULATED_SETUP(multiple),
2109 	WARN_EMULATED_SETUP(popcntb),
2110 	WARN_EMULATED_SETUP(spe),
2111 	WARN_EMULATED_SETUP(string),
2112 	WARN_EMULATED_SETUP(sync),
2113 	WARN_EMULATED_SETUP(unaligned),
2114 #ifdef CONFIG_MATH_EMULATION
2115 	WARN_EMULATED_SETUP(math),
2116 #endif
2117 #ifdef CONFIG_VSX
2118 	WARN_EMULATED_SETUP(vsx),
2119 #endif
2120 #ifdef CONFIG_PPC64
2121 	WARN_EMULATED_SETUP(mfdscr),
2122 	WARN_EMULATED_SETUP(mtdscr),
2123 	WARN_EMULATED_SETUP(lq_stq),
2124 	WARN_EMULATED_SETUP(lxvw4x),
2125 	WARN_EMULATED_SETUP(lxvh8x),
2126 	WARN_EMULATED_SETUP(lxvd2x),
2127 	WARN_EMULATED_SETUP(lxvb16x),
2128 #endif
2129 };
2130 
2131 u32 ppc_warn_emulated;
2132 
2133 void ppc_warn_emulated_print(const char *type)
2134 {
2135 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2136 			    type);
2137 }
2138 
2139 static int __init ppc_warn_emulated_init(void)
2140 {
2141 	struct dentry *dir, *d;
2142 	unsigned int i;
2143 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2144 
2145 	if (!powerpc_debugfs_root)
2146 		return -ENODEV;
2147 
2148 	dir = debugfs_create_dir("emulated_instructions",
2149 				 powerpc_debugfs_root);
2150 	if (!dir)
2151 		return -ENOMEM;
2152 
2153 	d = debugfs_create_u32("do_warn", S_IRUGO | S_IWUSR, dir,
2154 			       &ppc_warn_emulated);
2155 	if (!d)
2156 		goto fail;
2157 
2158 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) {
2159 		d = debugfs_create_u32(entries[i].name, S_IRUGO | S_IWUSR, dir,
2160 				       (u32 *)&entries[i].val.counter);
2161 		if (!d)
2162 			goto fail;
2163 	}
2164 
2165 	return 0;
2166 
2167 fail:
2168 	debugfs_remove_recursive(dir);
2169 	return -ENOMEM;
2170 }
2171 
2172 device_initcall(ppc_warn_emulated_init);
2173 
2174 #endif /* CONFIG_PPC_EMULATED_STATS */
2175