xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision 1cac4f26)
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 #include <asm/stacktrace.h>
74 
75 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
76 int (*__debugger)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
79 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
80 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
81 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
82 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
83 
84 EXPORT_SYMBOL(__debugger);
85 EXPORT_SYMBOL(__debugger_ipi);
86 EXPORT_SYMBOL(__debugger_bpt);
87 EXPORT_SYMBOL(__debugger_sstep);
88 EXPORT_SYMBOL(__debugger_iabr_match);
89 EXPORT_SYMBOL(__debugger_break_match);
90 EXPORT_SYMBOL(__debugger_fault_handler);
91 #endif
92 
93 /* Transactional Memory trap debug */
94 #ifdef TM_DEBUG_SW
95 #define TM_DEBUG(x...) printk(KERN_INFO x)
96 #else
97 #define TM_DEBUG(x...) do { } while(0)
98 #endif
99 
100 static const char *signame(int signr)
101 {
102 	switch (signr) {
103 	case SIGBUS:	return "bus error";
104 	case SIGFPE:	return "floating point exception";
105 	case SIGILL:	return "illegal instruction";
106 	case SIGSEGV:	return "segfault";
107 	case SIGTRAP:	return "unhandled trap";
108 	}
109 
110 	return "unknown signal";
111 }
112 
113 /*
114  * Trap & Exception support
115  */
116 
117 #ifdef CONFIG_PMAC_BACKLIGHT
118 static void pmac_backlight_unblank(void)
119 {
120 	mutex_lock(&pmac_backlight_mutex);
121 	if (pmac_backlight) {
122 		struct backlight_properties *props;
123 
124 		props = &pmac_backlight->props;
125 		props->brightness = props->max_brightness;
126 		props->power = FB_BLANK_UNBLANK;
127 		backlight_update_status(pmac_backlight);
128 	}
129 	mutex_unlock(&pmac_backlight_mutex);
130 }
131 #else
132 static inline void pmac_backlight_unblank(void) { }
133 #endif
134 
135 /*
136  * If oops/die is expected to crash the machine, return true here.
137  *
138  * This should not be expected to be 100% accurate, there may be
139  * notifiers registered or other unexpected conditions that may bring
140  * down the kernel. Or if the current process in the kernel is holding
141  * locks or has other critical state, the kernel may become effectively
142  * unusable anyway.
143  */
144 bool die_will_crash(void)
145 {
146 	if (should_fadump_crash())
147 		return true;
148 	if (kexec_should_crash(current))
149 		return true;
150 	if (in_interrupt() || panic_on_oops ||
151 			!current->pid || is_global_init(current))
152 		return true;
153 
154 	return false;
155 }
156 
157 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
158 static int die_owner = -1;
159 static unsigned int die_nest_count;
160 static int die_counter;
161 
162 extern void panic_flush_kmsg_start(void)
163 {
164 	/*
165 	 * These are mostly taken from kernel/panic.c, but tries to do
166 	 * relatively minimal work. Don't use delay functions (TB may
167 	 * be broken), don't crash dump (need to set a firmware log),
168 	 * don't run notifiers. We do want to get some information to
169 	 * Linux console.
170 	 */
171 	console_verbose();
172 	bust_spinlocks(1);
173 }
174 
175 extern void panic_flush_kmsg_end(void)
176 {
177 	printk_safe_flush_on_panic();
178 	kmsg_dump(KMSG_DUMP_PANIC);
179 	bust_spinlocks(0);
180 	debug_locks_off();
181 	console_flush_on_panic();
182 }
183 
184 static unsigned long oops_begin(struct pt_regs *regs)
185 {
186 	int cpu;
187 	unsigned long flags;
188 
189 	oops_enter();
190 
191 	/* racy, but better than risking deadlock. */
192 	raw_local_irq_save(flags);
193 	cpu = smp_processor_id();
194 	if (!arch_spin_trylock(&die_lock)) {
195 		if (cpu == die_owner)
196 			/* nested oops. should stop eventually */;
197 		else
198 			arch_spin_lock(&die_lock);
199 	}
200 	die_nest_count++;
201 	die_owner = cpu;
202 	console_verbose();
203 	bust_spinlocks(1);
204 	if (machine_is(powermac))
205 		pmac_backlight_unblank();
206 	return flags;
207 }
208 NOKPROBE_SYMBOL(oops_begin);
209 
210 static void oops_end(unsigned long flags, struct pt_regs *regs,
211 			       int signr)
212 {
213 	bust_spinlocks(0);
214 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
215 	die_nest_count--;
216 	oops_exit();
217 	printk("\n");
218 	if (!die_nest_count) {
219 		/* Nest count reaches zero, release the lock. */
220 		die_owner = -1;
221 		arch_spin_unlock(&die_lock);
222 	}
223 	raw_local_irq_restore(flags);
224 
225 	/*
226 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
227 	 */
228 	if (TRAP(regs) == 0x100)
229 		return;
230 
231 	crash_fadump(regs, "die oops");
232 
233 	if (kexec_should_crash(current))
234 		crash_kexec(regs);
235 
236 	if (!signr)
237 		return;
238 
239 	/*
240 	 * While our oops output is serialised by a spinlock, output
241 	 * from panic() called below can race and corrupt it. If we
242 	 * know we are going to panic, delay for 1 second so we have a
243 	 * chance to get clean backtraces from all CPUs that are oopsing.
244 	 */
245 	if (in_interrupt() || panic_on_oops || !current->pid ||
246 	    is_global_init(current)) {
247 		mdelay(MSEC_PER_SEC);
248 	}
249 
250 	if (panic_on_oops)
251 		panic("Fatal exception");
252 	do_exit(signr);
253 }
254 NOKPROBE_SYMBOL(oops_end);
255 
256 static int __die(const char *str, struct pt_regs *regs, long err)
257 {
258 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
259 
260 	if (IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN))
261 		printk("LE ");
262 	else
263 		printk("BE ");
264 
265 	if (IS_ENABLED(CONFIG_PREEMPT))
266 		pr_cont("PREEMPT ");
267 
268 	if (IS_ENABLED(CONFIG_SMP))
269 		pr_cont("SMP NR_CPUS=%d ", NR_CPUS);
270 
271 	if (debug_pagealloc_enabled())
272 		pr_cont("DEBUG_PAGEALLOC ");
273 
274 	if (IS_ENABLED(CONFIG_NUMA))
275 		pr_cont("NUMA ");
276 
277 	pr_cont("%s\n", ppc_md.name ? ppc_md.name : "");
278 
279 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
280 		return 1;
281 
282 	print_modules();
283 	show_regs(regs);
284 
285 	return 0;
286 }
287 NOKPROBE_SYMBOL(__die);
288 
289 void die(const char *str, struct pt_regs *regs, long err)
290 {
291 	unsigned long flags;
292 
293 	/*
294 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
295 	 */
296 	if (TRAP(regs) != 0x100) {
297 		if (debugger(regs))
298 			return;
299 	}
300 
301 	flags = oops_begin(regs);
302 	if (__die(str, regs, err))
303 		err = 0;
304 	oops_end(flags, regs, err);
305 }
306 NOKPROBE_SYMBOL(die);
307 
308 void user_single_step_report(struct pt_regs *regs)
309 {
310 	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip, current);
311 }
312 
313 static void show_signal_msg(int signr, struct pt_regs *regs, int code,
314 			    unsigned long addr)
315 {
316 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
317 				      DEFAULT_RATELIMIT_BURST);
318 
319 	if (!show_unhandled_signals)
320 		return;
321 
322 	if (!unhandled_signal(current, signr))
323 		return;
324 
325 	if (!__ratelimit(&rs))
326 		return;
327 
328 	pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x",
329 		current->comm, current->pid, signame(signr), signr,
330 		addr, regs->nip, regs->link, code);
331 
332 	print_vma_addr(KERN_CONT " in ", regs->nip);
333 
334 	pr_cont("\n");
335 
336 	show_user_instructions(regs);
337 }
338 
339 static bool exception_common(int signr, struct pt_regs *regs, int code,
340 			      unsigned long addr)
341 {
342 	if (!user_mode(regs)) {
343 		die("Exception in kernel mode", regs, signr);
344 		return false;
345 	}
346 
347 	show_signal_msg(signr, regs, code, addr);
348 
349 	if (arch_irqs_disabled() && !arch_irq_disabled_regs(regs))
350 		local_irq_enable();
351 
352 	current->thread.trap_nr = code;
353 
354 	/*
355 	 * Save all the pkey registers AMR/IAMR/UAMOR. Eg: Core dumps need
356 	 * to capture the content, if the task gets killed.
357 	 */
358 	thread_pkey_regs_save(&current->thread);
359 
360 	return true;
361 }
362 
363 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key)
364 {
365 	if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr))
366 		return;
367 
368 	force_sig_pkuerr((void __user *) addr, key);
369 }
370 
371 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
372 {
373 	if (!exception_common(signr, regs, code, addr))
374 		return;
375 
376 	force_sig_fault(signr, code, (void __user *)addr, current);
377 }
378 
379 void system_reset_exception(struct pt_regs *regs)
380 {
381 	/*
382 	 * Avoid crashes in case of nested NMI exceptions. Recoverability
383 	 * is determined by RI and in_nmi
384 	 */
385 	bool nested = in_nmi();
386 	if (!nested)
387 		nmi_enter();
388 
389 	__this_cpu_inc(irq_stat.sreset_irqs);
390 
391 	/* See if any machine dependent calls */
392 	if (ppc_md.system_reset_exception) {
393 		if (ppc_md.system_reset_exception(regs))
394 			goto out;
395 	}
396 
397 	if (debugger(regs))
398 		goto out;
399 
400 	/*
401 	 * A system reset is a request to dump, so we always send
402 	 * it through the crashdump code (if fadump or kdump are
403 	 * registered).
404 	 */
405 	crash_fadump(regs, "System Reset");
406 
407 	crash_kexec(regs);
408 
409 	/*
410 	 * We aren't the primary crash CPU. We need to send it
411 	 * to a holding pattern to avoid it ending up in the panic
412 	 * code.
413 	 */
414 	crash_kexec_secondary(regs);
415 
416 	/*
417 	 * No debugger or crash dump registered, print logs then
418 	 * panic.
419 	 */
420 	die("System Reset", regs, SIGABRT);
421 
422 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
423 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
424 	nmi_panic(regs, "System Reset");
425 
426 out:
427 #ifdef CONFIG_PPC_BOOK3S_64
428 	BUG_ON(get_paca()->in_nmi == 0);
429 	if (get_paca()->in_nmi > 1)
430 		nmi_panic(regs, "Unrecoverable nested System Reset");
431 #endif
432 	/* Must die if the interrupt is not recoverable */
433 	if (!(regs->msr & MSR_RI))
434 		nmi_panic(regs, "Unrecoverable System Reset");
435 
436 	if (!nested)
437 		nmi_exit();
438 
439 	/* What should we do here? We could issue a shutdown or hard reset. */
440 }
441 
442 /*
443  * I/O accesses can cause machine checks on powermacs.
444  * Check if the NIP corresponds to the address of a sync
445  * instruction for which there is an entry in the exception
446  * table.
447  * Note that the 601 only takes a machine check on TEA
448  * (transfer error ack) signal assertion, and does not
449  * set any of the top 16 bits of SRR1.
450  *  -- paulus.
451  */
452 static inline int check_io_access(struct pt_regs *regs)
453 {
454 #ifdef CONFIG_PPC32
455 	unsigned long msr = regs->msr;
456 	const struct exception_table_entry *entry;
457 	unsigned int *nip = (unsigned int *)regs->nip;
458 
459 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
460 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
461 		/*
462 		 * Check that it's a sync instruction, or somewhere
463 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
464 		 * As the address is in the exception table
465 		 * we should be able to read the instr there.
466 		 * For the debug message, we look at the preceding
467 		 * load or store.
468 		 */
469 		if (*nip == PPC_INST_NOP)
470 			nip -= 2;
471 		else if (*nip == PPC_INST_ISYNC)
472 			--nip;
473 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
474 			unsigned int rb;
475 
476 			--nip;
477 			rb = (*nip >> 11) & 0x1f;
478 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
479 			       (*nip & 0x100)? "OUT to": "IN from",
480 			       regs->gpr[rb] - _IO_BASE, nip);
481 			regs->msr |= MSR_RI;
482 			regs->nip = extable_fixup(entry);
483 			return 1;
484 		}
485 	}
486 #endif /* CONFIG_PPC32 */
487 	return 0;
488 }
489 
490 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
491 /* On 4xx, the reason for the machine check or program exception
492    is in the ESR. */
493 #define get_reason(regs)	((regs)->dsisr)
494 #define REASON_FP		ESR_FP
495 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
496 #define REASON_PRIVILEGED	ESR_PPR
497 #define REASON_TRAP		ESR_PTR
498 
499 /* single-step stuff */
500 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
501 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
502 #define clear_br_trace(regs)	do {} while(0)
503 #else
504 /* On non-4xx, the reason for the machine check or program
505    exception is in the MSR. */
506 #define get_reason(regs)	((regs)->msr)
507 #define REASON_TM		SRR1_PROGTM
508 #define REASON_FP		SRR1_PROGFPE
509 #define REASON_ILLEGAL		SRR1_PROGILL
510 #define REASON_PRIVILEGED	SRR1_PROGPRIV
511 #define REASON_TRAP		SRR1_PROGTRAP
512 
513 #define single_stepping(regs)	((regs)->msr & MSR_SE)
514 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
515 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
516 #endif
517 
518 #if defined(CONFIG_E500)
519 int machine_check_e500mc(struct pt_regs *regs)
520 {
521 	unsigned long mcsr = mfspr(SPRN_MCSR);
522 	unsigned long pvr = mfspr(SPRN_PVR);
523 	unsigned long reason = mcsr;
524 	int recoverable = 1;
525 
526 	if (reason & MCSR_LD) {
527 		recoverable = fsl_rio_mcheck_exception(regs);
528 		if (recoverable == 1)
529 			goto silent_out;
530 	}
531 
532 	printk("Machine check in kernel mode.\n");
533 	printk("Caused by (from MCSR=%lx): ", reason);
534 
535 	if (reason & MCSR_MCP)
536 		pr_cont("Machine Check Signal\n");
537 
538 	if (reason & MCSR_ICPERR) {
539 		pr_cont("Instruction Cache Parity Error\n");
540 
541 		/*
542 		 * This is recoverable by invalidating the i-cache.
543 		 */
544 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
545 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
546 			;
547 
548 		/*
549 		 * This will generally be accompanied by an instruction
550 		 * fetch error report -- only treat MCSR_IF as fatal
551 		 * if it wasn't due to an L1 parity error.
552 		 */
553 		reason &= ~MCSR_IF;
554 	}
555 
556 	if (reason & MCSR_DCPERR_MC) {
557 		pr_cont("Data Cache Parity Error\n");
558 
559 		/*
560 		 * In write shadow mode we auto-recover from the error, but it
561 		 * may still get logged and cause a machine check.  We should
562 		 * only treat the non-write shadow case as non-recoverable.
563 		 */
564 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
565 		 * is not implemented but L1 data cache always runs in write
566 		 * shadow mode. Hence on data cache parity errors HW will
567 		 * automatically invalidate the L1 Data Cache.
568 		 */
569 		if (PVR_VER(pvr) != PVR_VER_E6500) {
570 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
571 				recoverable = 0;
572 		}
573 	}
574 
575 	if (reason & MCSR_L2MMU_MHIT) {
576 		pr_cont("Hit on multiple TLB entries\n");
577 		recoverable = 0;
578 	}
579 
580 	if (reason & MCSR_NMI)
581 		pr_cont("Non-maskable interrupt\n");
582 
583 	if (reason & MCSR_IF) {
584 		pr_cont("Instruction Fetch Error Report\n");
585 		recoverable = 0;
586 	}
587 
588 	if (reason & MCSR_LD) {
589 		pr_cont("Load Error Report\n");
590 		recoverable = 0;
591 	}
592 
593 	if (reason & MCSR_ST) {
594 		pr_cont("Store Error Report\n");
595 		recoverable = 0;
596 	}
597 
598 	if (reason & MCSR_LDG) {
599 		pr_cont("Guarded Load Error Report\n");
600 		recoverable = 0;
601 	}
602 
603 	if (reason & MCSR_TLBSYNC)
604 		pr_cont("Simultaneous tlbsync operations\n");
605 
606 	if (reason & MCSR_BSL2_ERR) {
607 		pr_cont("Level 2 Cache Error\n");
608 		recoverable = 0;
609 	}
610 
611 	if (reason & MCSR_MAV) {
612 		u64 addr;
613 
614 		addr = mfspr(SPRN_MCAR);
615 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
616 
617 		pr_cont("Machine Check %s Address: %#llx\n",
618 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
619 	}
620 
621 silent_out:
622 	mtspr(SPRN_MCSR, mcsr);
623 	return mfspr(SPRN_MCSR) == 0 && recoverable;
624 }
625 
626 int machine_check_e500(struct pt_regs *regs)
627 {
628 	unsigned long reason = mfspr(SPRN_MCSR);
629 
630 	if (reason & MCSR_BUS_RBERR) {
631 		if (fsl_rio_mcheck_exception(regs))
632 			return 1;
633 		if (fsl_pci_mcheck_exception(regs))
634 			return 1;
635 	}
636 
637 	printk("Machine check in kernel mode.\n");
638 	printk("Caused by (from MCSR=%lx): ", reason);
639 
640 	if (reason & MCSR_MCP)
641 		pr_cont("Machine Check Signal\n");
642 	if (reason & MCSR_ICPERR)
643 		pr_cont("Instruction Cache Parity Error\n");
644 	if (reason & MCSR_DCP_PERR)
645 		pr_cont("Data Cache Push Parity Error\n");
646 	if (reason & MCSR_DCPERR)
647 		pr_cont("Data Cache Parity Error\n");
648 	if (reason & MCSR_BUS_IAERR)
649 		pr_cont("Bus - Instruction Address Error\n");
650 	if (reason & MCSR_BUS_RAERR)
651 		pr_cont("Bus - Read Address Error\n");
652 	if (reason & MCSR_BUS_WAERR)
653 		pr_cont("Bus - Write Address Error\n");
654 	if (reason & MCSR_BUS_IBERR)
655 		pr_cont("Bus - Instruction Data Error\n");
656 	if (reason & MCSR_BUS_RBERR)
657 		pr_cont("Bus - Read Data Bus Error\n");
658 	if (reason & MCSR_BUS_WBERR)
659 		pr_cont("Bus - Write Data Bus Error\n");
660 	if (reason & MCSR_BUS_IPERR)
661 		pr_cont("Bus - Instruction Parity Error\n");
662 	if (reason & MCSR_BUS_RPERR)
663 		pr_cont("Bus - Read Parity Error\n");
664 
665 	return 0;
666 }
667 
668 int machine_check_generic(struct pt_regs *regs)
669 {
670 	return 0;
671 }
672 #elif defined(CONFIG_E200)
673 int machine_check_e200(struct pt_regs *regs)
674 {
675 	unsigned long reason = mfspr(SPRN_MCSR);
676 
677 	printk("Machine check in kernel mode.\n");
678 	printk("Caused by (from MCSR=%lx): ", reason);
679 
680 	if (reason & MCSR_MCP)
681 		pr_cont("Machine Check Signal\n");
682 	if (reason & MCSR_CP_PERR)
683 		pr_cont("Cache Push Parity Error\n");
684 	if (reason & MCSR_CPERR)
685 		pr_cont("Cache Parity Error\n");
686 	if (reason & MCSR_EXCP_ERR)
687 		pr_cont("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
688 	if (reason & MCSR_BUS_IRERR)
689 		pr_cont("Bus - Read Bus Error on instruction fetch\n");
690 	if (reason & MCSR_BUS_DRERR)
691 		pr_cont("Bus - Read Bus Error on data load\n");
692 	if (reason & MCSR_BUS_WRERR)
693 		pr_cont("Bus - Write Bus Error on buffered store or cache line push\n");
694 
695 	return 0;
696 }
697 #elif defined(CONFIG_PPC32)
698 int machine_check_generic(struct pt_regs *regs)
699 {
700 	unsigned long reason = regs->msr;
701 
702 	printk("Machine check in kernel mode.\n");
703 	printk("Caused by (from SRR1=%lx): ", reason);
704 	switch (reason & 0x601F0000) {
705 	case 0x80000:
706 		pr_cont("Machine check signal\n");
707 		break;
708 	case 0:		/* for 601 */
709 	case 0x40000:
710 	case 0x140000:	/* 7450 MSS error and TEA */
711 		pr_cont("Transfer error ack signal\n");
712 		break;
713 	case 0x20000:
714 		pr_cont("Data parity error signal\n");
715 		break;
716 	case 0x10000:
717 		pr_cont("Address parity error signal\n");
718 		break;
719 	case 0x20000000:
720 		pr_cont("L1 Data Cache error\n");
721 		break;
722 	case 0x40000000:
723 		pr_cont("L1 Instruction Cache error\n");
724 		break;
725 	case 0x00100000:
726 		pr_cont("L2 data cache parity error\n");
727 		break;
728 	default:
729 		pr_cont("Unknown values in msr\n");
730 	}
731 	return 0;
732 }
733 #endif /* everything else */
734 
735 void machine_check_exception(struct pt_regs *regs)
736 {
737 	int recover = 0;
738 	bool nested = in_nmi();
739 	if (!nested)
740 		nmi_enter();
741 
742 	__this_cpu_inc(irq_stat.mce_exceptions);
743 
744 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
745 
746 	/* See if any machine dependent calls. In theory, we would want
747 	 * to call the CPU first, and call the ppc_md. one if the CPU
748 	 * one returns a positive number. However there is existing code
749 	 * that assumes the board gets a first chance, so let's keep it
750 	 * that way for now and fix things later. --BenH.
751 	 */
752 	if (ppc_md.machine_check_exception)
753 		recover = ppc_md.machine_check_exception(regs);
754 	else if (cur_cpu_spec->machine_check)
755 		recover = cur_cpu_spec->machine_check(regs);
756 
757 	if (recover > 0)
758 		goto bail;
759 
760 	if (debugger_fault_handler(regs))
761 		goto bail;
762 
763 	if (check_io_access(regs))
764 		goto bail;
765 
766 	/* Must die if the interrupt is not recoverable */
767 	if (!(regs->msr & MSR_RI))
768 		nmi_panic(regs, "Unrecoverable Machine check");
769 
770 	if (!nested)
771 		nmi_exit();
772 
773 	die("Machine check", regs, SIGBUS);
774 
775 	return;
776 
777 bail:
778 	if (!nested)
779 		nmi_exit();
780 }
781 
782 void SMIException(struct pt_regs *regs)
783 {
784 	die("System Management Interrupt", regs, SIGABRT);
785 }
786 
787 #ifdef CONFIG_VSX
788 static void p9_hmi_special_emu(struct pt_regs *regs)
789 {
790 	unsigned int ra, rb, t, i, sel, instr, rc;
791 	const void __user *addr;
792 	u8 vbuf[16], *vdst;
793 	unsigned long ea, msr, msr_mask;
794 	bool swap;
795 
796 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
797 		return;
798 
799 	/*
800 	 * lxvb16x	opcode: 0x7c0006d8
801 	 * lxvd2x	opcode: 0x7c000698
802 	 * lxvh8x	opcode: 0x7c000658
803 	 * lxvw4x	opcode: 0x7c000618
804 	 */
805 	if ((instr & 0xfc00073e) != 0x7c000618) {
806 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
807 			 " instr=%08x\n",
808 			 smp_processor_id(), current->comm, current->pid,
809 			 regs->nip, instr);
810 		return;
811 	}
812 
813 	/* Grab vector registers into the task struct */
814 	msr = regs->msr; /* Grab msr before we flush the bits */
815 	flush_vsx_to_thread(current);
816 	enable_kernel_altivec();
817 
818 	/*
819 	 * Is userspace running with a different endian (this is rare but
820 	 * not impossible)
821 	 */
822 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
823 
824 	/* Decode the instruction */
825 	ra = (instr >> 16) & 0x1f;
826 	rb = (instr >> 11) & 0x1f;
827 	t = (instr >> 21) & 0x1f;
828 	if (instr & 1)
829 		vdst = (u8 *)&current->thread.vr_state.vr[t];
830 	else
831 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
832 
833 	/* Grab the vector address */
834 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
835 	if (is_32bit_task())
836 		ea &= 0xfffffffful;
837 	addr = (__force const void __user *)ea;
838 
839 	/* Check it */
840 	if (!access_ok(addr, 16)) {
841 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
842 			 " instr=%08x addr=%016lx\n",
843 			 smp_processor_id(), current->comm, current->pid,
844 			 regs->nip, instr, (unsigned long)addr);
845 		return;
846 	}
847 
848 	/* Read the vector */
849 	rc = 0;
850 	if ((unsigned long)addr & 0xfUL)
851 		/* unaligned case */
852 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
853 	else
854 		__get_user_atomic_128_aligned(vbuf, addr, rc);
855 	if (rc) {
856 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
857 			 " instr=%08x addr=%016lx\n",
858 			 smp_processor_id(), current->comm, current->pid,
859 			 regs->nip, instr, (unsigned long)addr);
860 		return;
861 	}
862 
863 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
864 		 " instr=%08x addr=%016lx\n",
865 		 smp_processor_id(), current->comm, current->pid, regs->nip,
866 		 instr, (unsigned long) addr);
867 
868 	/* Grab instruction "selector" */
869 	sel = (instr >> 6) & 3;
870 
871 	/*
872 	 * Check to make sure the facility is actually enabled. This
873 	 * could happen if we get a false positive hit.
874 	 *
875 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
876 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
877 	 */
878 	msr_mask = MSR_VSX;
879 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
880 		msr_mask = MSR_VEC;
881 	if (!(msr & msr_mask)) {
882 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
883 			 " instr=%08x msr:%016lx\n",
884 			 smp_processor_id(), current->comm, current->pid,
885 			 regs->nip, instr, msr);
886 		return;
887 	}
888 
889 	/* Do logging here before we modify sel based on endian */
890 	switch (sel) {
891 	case 0:	/* lxvw4x */
892 		PPC_WARN_EMULATED(lxvw4x, regs);
893 		break;
894 	case 1: /* lxvh8x */
895 		PPC_WARN_EMULATED(lxvh8x, regs);
896 		break;
897 	case 2: /* lxvd2x */
898 		PPC_WARN_EMULATED(lxvd2x, regs);
899 		break;
900 	case 3: /* lxvb16x */
901 		PPC_WARN_EMULATED(lxvb16x, regs);
902 		break;
903 	}
904 
905 #ifdef __LITTLE_ENDIAN__
906 	/*
907 	 * An LE kernel stores the vector in the task struct as an LE
908 	 * byte array (effectively swapping both the components and
909 	 * the content of the components). Those instructions expect
910 	 * the components to remain in ascending address order, so we
911 	 * swap them back.
912 	 *
913 	 * If we are running a BE user space, the expectation is that
914 	 * of a simple memcpy, so forcing the emulation to look like
915 	 * a lxvb16x should do the trick.
916 	 */
917 	if (swap)
918 		sel = 3;
919 
920 	switch (sel) {
921 	case 0:	/* lxvw4x */
922 		for (i = 0; i < 4; i++)
923 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
924 		break;
925 	case 1: /* lxvh8x */
926 		for (i = 0; i < 8; i++)
927 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
928 		break;
929 	case 2: /* lxvd2x */
930 		for (i = 0; i < 2; i++)
931 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
932 		break;
933 	case 3: /* lxvb16x */
934 		for (i = 0; i < 16; i++)
935 			vdst[i] = vbuf[15-i];
936 		break;
937 	}
938 #else /* __LITTLE_ENDIAN__ */
939 	/* On a big endian kernel, a BE userspace only needs a memcpy */
940 	if (!swap)
941 		sel = 3;
942 
943 	/* Otherwise, we need to swap the content of the components */
944 	switch (sel) {
945 	case 0:	/* lxvw4x */
946 		for (i = 0; i < 4; i++)
947 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
948 		break;
949 	case 1: /* lxvh8x */
950 		for (i = 0; i < 8; i++)
951 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
952 		break;
953 	case 2: /* lxvd2x */
954 		for (i = 0; i < 2; i++)
955 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
956 		break;
957 	case 3: /* lxvb16x */
958 		memcpy(vdst, vbuf, 16);
959 		break;
960 	}
961 #endif /* !__LITTLE_ENDIAN__ */
962 
963 	/* Go to next instruction */
964 	regs->nip += 4;
965 }
966 #endif /* CONFIG_VSX */
967 
968 void handle_hmi_exception(struct pt_regs *regs)
969 {
970 	struct pt_regs *old_regs;
971 
972 	old_regs = set_irq_regs(regs);
973 	irq_enter();
974 
975 #ifdef CONFIG_VSX
976 	/* Real mode flagged P9 special emu is needed */
977 	if (local_paca->hmi_p9_special_emu) {
978 		local_paca->hmi_p9_special_emu = 0;
979 
980 		/*
981 		 * We don't want to take page faults while doing the
982 		 * emulation, we just replay the instruction if necessary.
983 		 */
984 		pagefault_disable();
985 		p9_hmi_special_emu(regs);
986 		pagefault_enable();
987 	}
988 #endif /* CONFIG_VSX */
989 
990 	if (ppc_md.handle_hmi_exception)
991 		ppc_md.handle_hmi_exception(regs);
992 
993 	irq_exit();
994 	set_irq_regs(old_regs);
995 }
996 
997 void unknown_exception(struct pt_regs *regs)
998 {
999 	enum ctx_state prev_state = exception_enter();
1000 
1001 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1002 	       regs->nip, regs->msr, regs->trap);
1003 
1004 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1005 
1006 	exception_exit(prev_state);
1007 }
1008 
1009 void instruction_breakpoint_exception(struct pt_regs *regs)
1010 {
1011 	enum ctx_state prev_state = exception_enter();
1012 
1013 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1014 					5, SIGTRAP) == NOTIFY_STOP)
1015 		goto bail;
1016 	if (debugger_iabr_match(regs))
1017 		goto bail;
1018 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1019 
1020 bail:
1021 	exception_exit(prev_state);
1022 }
1023 
1024 void RunModeException(struct pt_regs *regs)
1025 {
1026 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1027 }
1028 
1029 void single_step_exception(struct pt_regs *regs)
1030 {
1031 	enum ctx_state prev_state = exception_enter();
1032 
1033 	clear_single_step(regs);
1034 	clear_br_trace(regs);
1035 
1036 	if (kprobe_post_handler(regs))
1037 		return;
1038 
1039 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1040 					5, SIGTRAP) == NOTIFY_STOP)
1041 		goto bail;
1042 	if (debugger_sstep(regs))
1043 		goto bail;
1044 
1045 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1046 
1047 bail:
1048 	exception_exit(prev_state);
1049 }
1050 NOKPROBE_SYMBOL(single_step_exception);
1051 
1052 /*
1053  * After we have successfully emulated an instruction, we have to
1054  * check if the instruction was being single-stepped, and if so,
1055  * pretend we got a single-step exception.  This was pointed out
1056  * by Kumar Gala.  -- paulus
1057  */
1058 static void emulate_single_step(struct pt_regs *regs)
1059 {
1060 	if (single_stepping(regs))
1061 		single_step_exception(regs);
1062 }
1063 
1064 static inline int __parse_fpscr(unsigned long fpscr)
1065 {
1066 	int ret = FPE_FLTUNK;
1067 
1068 	/* Invalid operation */
1069 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1070 		ret = FPE_FLTINV;
1071 
1072 	/* Overflow */
1073 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1074 		ret = FPE_FLTOVF;
1075 
1076 	/* Underflow */
1077 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1078 		ret = FPE_FLTUND;
1079 
1080 	/* Divide by zero */
1081 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1082 		ret = FPE_FLTDIV;
1083 
1084 	/* Inexact result */
1085 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1086 		ret = FPE_FLTRES;
1087 
1088 	return ret;
1089 }
1090 
1091 static void parse_fpe(struct pt_regs *regs)
1092 {
1093 	int code = 0;
1094 
1095 	flush_fp_to_thread(current);
1096 
1097 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1098 
1099 	_exception(SIGFPE, regs, code, regs->nip);
1100 }
1101 
1102 /*
1103  * Illegal instruction emulation support.  Originally written to
1104  * provide the PVR to user applications using the mfspr rd, PVR.
1105  * Return non-zero if we can't emulate, or -EFAULT if the associated
1106  * memory access caused an access fault.  Return zero on success.
1107  *
1108  * There are a couple of ways to do this, either "decode" the instruction
1109  * or directly match lots of bits.  In this case, matching lots of
1110  * bits is faster and easier.
1111  *
1112  */
1113 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1114 {
1115 	u8 rT = (instword >> 21) & 0x1f;
1116 	u8 rA = (instword >> 16) & 0x1f;
1117 	u8 NB_RB = (instword >> 11) & 0x1f;
1118 	u32 num_bytes;
1119 	unsigned long EA;
1120 	int pos = 0;
1121 
1122 	/* Early out if we are an invalid form of lswx */
1123 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1124 		if ((rT == rA) || (rT == NB_RB))
1125 			return -EINVAL;
1126 
1127 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1128 
1129 	switch (instword & PPC_INST_STRING_MASK) {
1130 		case PPC_INST_LSWX:
1131 		case PPC_INST_STSWX:
1132 			EA += NB_RB;
1133 			num_bytes = regs->xer & 0x7f;
1134 			break;
1135 		case PPC_INST_LSWI:
1136 		case PPC_INST_STSWI:
1137 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1138 			break;
1139 		default:
1140 			return -EINVAL;
1141 	}
1142 
1143 	while (num_bytes != 0)
1144 	{
1145 		u8 val;
1146 		u32 shift = 8 * (3 - (pos & 0x3));
1147 
1148 		/* if process is 32-bit, clear upper 32 bits of EA */
1149 		if ((regs->msr & MSR_64BIT) == 0)
1150 			EA &= 0xFFFFFFFF;
1151 
1152 		switch ((instword & PPC_INST_STRING_MASK)) {
1153 			case PPC_INST_LSWX:
1154 			case PPC_INST_LSWI:
1155 				if (get_user(val, (u8 __user *)EA))
1156 					return -EFAULT;
1157 				/* first time updating this reg,
1158 				 * zero it out */
1159 				if (pos == 0)
1160 					regs->gpr[rT] = 0;
1161 				regs->gpr[rT] |= val << shift;
1162 				break;
1163 			case PPC_INST_STSWI:
1164 			case PPC_INST_STSWX:
1165 				val = regs->gpr[rT] >> shift;
1166 				if (put_user(val, (u8 __user *)EA))
1167 					return -EFAULT;
1168 				break;
1169 		}
1170 		/* move EA to next address */
1171 		EA += 1;
1172 		num_bytes--;
1173 
1174 		/* manage our position within the register */
1175 		if (++pos == 4) {
1176 			pos = 0;
1177 			if (++rT == 32)
1178 				rT = 0;
1179 		}
1180 	}
1181 
1182 	return 0;
1183 }
1184 
1185 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1186 {
1187 	u32 ra,rs;
1188 	unsigned long tmp;
1189 
1190 	ra = (instword >> 16) & 0x1f;
1191 	rs = (instword >> 21) & 0x1f;
1192 
1193 	tmp = regs->gpr[rs];
1194 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1195 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1196 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1197 	regs->gpr[ra] = tmp;
1198 
1199 	return 0;
1200 }
1201 
1202 static int emulate_isel(struct pt_regs *regs, u32 instword)
1203 {
1204 	u8 rT = (instword >> 21) & 0x1f;
1205 	u8 rA = (instword >> 16) & 0x1f;
1206 	u8 rB = (instword >> 11) & 0x1f;
1207 	u8 BC = (instword >> 6) & 0x1f;
1208 	u8 bit;
1209 	unsigned long tmp;
1210 
1211 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1212 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1213 
1214 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1215 
1216 	return 0;
1217 }
1218 
1219 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1220 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1221 {
1222         /* If we're emulating a load/store in an active transaction, we cannot
1223          * emulate it as the kernel operates in transaction suspended context.
1224          * We need to abort the transaction.  This creates a persistent TM
1225          * abort so tell the user what caused it with a new code.
1226 	 */
1227 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1228 		tm_enable();
1229 		tm_abort(cause);
1230 		return true;
1231 	}
1232 	return false;
1233 }
1234 #else
1235 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1236 {
1237 	return false;
1238 }
1239 #endif
1240 
1241 static int emulate_instruction(struct pt_regs *regs)
1242 {
1243 	u32 instword;
1244 	u32 rd;
1245 
1246 	if (!user_mode(regs))
1247 		return -EINVAL;
1248 	CHECK_FULL_REGS(regs);
1249 
1250 	if (get_user(instword, (u32 __user *)(regs->nip)))
1251 		return -EFAULT;
1252 
1253 	/* Emulate the mfspr rD, PVR. */
1254 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1255 		PPC_WARN_EMULATED(mfpvr, regs);
1256 		rd = (instword >> 21) & 0x1f;
1257 		regs->gpr[rd] = mfspr(SPRN_PVR);
1258 		return 0;
1259 	}
1260 
1261 	/* Emulating the dcba insn is just a no-op.  */
1262 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1263 		PPC_WARN_EMULATED(dcba, regs);
1264 		return 0;
1265 	}
1266 
1267 	/* Emulate the mcrxr insn.  */
1268 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1269 		int shift = (instword >> 21) & 0x1c;
1270 		unsigned long msk = 0xf0000000UL >> shift;
1271 
1272 		PPC_WARN_EMULATED(mcrxr, regs);
1273 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1274 		regs->xer &= ~0xf0000000UL;
1275 		return 0;
1276 	}
1277 
1278 	/* Emulate load/store string insn. */
1279 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1280 		if (tm_abort_check(regs,
1281 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1282 			return -EINVAL;
1283 		PPC_WARN_EMULATED(string, regs);
1284 		return emulate_string_inst(regs, instword);
1285 	}
1286 
1287 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1288 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1289 		PPC_WARN_EMULATED(popcntb, regs);
1290 		return emulate_popcntb_inst(regs, instword);
1291 	}
1292 
1293 	/* Emulate isel (Integer Select) instruction */
1294 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1295 		PPC_WARN_EMULATED(isel, regs);
1296 		return emulate_isel(regs, instword);
1297 	}
1298 
1299 	/* Emulate sync instruction variants */
1300 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1301 		PPC_WARN_EMULATED(sync, regs);
1302 		asm volatile("sync");
1303 		return 0;
1304 	}
1305 
1306 #ifdef CONFIG_PPC64
1307 	/* Emulate the mfspr rD, DSCR. */
1308 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1309 		PPC_INST_MFSPR_DSCR_USER) ||
1310 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1311 		PPC_INST_MFSPR_DSCR)) &&
1312 			cpu_has_feature(CPU_FTR_DSCR)) {
1313 		PPC_WARN_EMULATED(mfdscr, regs);
1314 		rd = (instword >> 21) & 0x1f;
1315 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1316 		return 0;
1317 	}
1318 	/* Emulate the mtspr DSCR, rD. */
1319 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1320 		PPC_INST_MTSPR_DSCR_USER) ||
1321 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1322 		PPC_INST_MTSPR_DSCR)) &&
1323 			cpu_has_feature(CPU_FTR_DSCR)) {
1324 		PPC_WARN_EMULATED(mtdscr, regs);
1325 		rd = (instword >> 21) & 0x1f;
1326 		current->thread.dscr = regs->gpr[rd];
1327 		current->thread.dscr_inherit = 1;
1328 		mtspr(SPRN_DSCR, current->thread.dscr);
1329 		return 0;
1330 	}
1331 #endif
1332 
1333 	return -EINVAL;
1334 }
1335 
1336 int is_valid_bugaddr(unsigned long addr)
1337 {
1338 	return is_kernel_addr(addr);
1339 }
1340 
1341 #ifdef CONFIG_MATH_EMULATION
1342 static int emulate_math(struct pt_regs *regs)
1343 {
1344 	int ret;
1345 	extern int do_mathemu(struct pt_regs *regs);
1346 
1347 	ret = do_mathemu(regs);
1348 	if (ret >= 0)
1349 		PPC_WARN_EMULATED(math, regs);
1350 
1351 	switch (ret) {
1352 	case 0:
1353 		emulate_single_step(regs);
1354 		return 0;
1355 	case 1: {
1356 			int code = 0;
1357 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1358 			_exception(SIGFPE, regs, code, regs->nip);
1359 			return 0;
1360 		}
1361 	case -EFAULT:
1362 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1363 		return 0;
1364 	}
1365 
1366 	return -1;
1367 }
1368 #else
1369 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1370 #endif
1371 
1372 void program_check_exception(struct pt_regs *regs)
1373 {
1374 	enum ctx_state prev_state = exception_enter();
1375 	unsigned int reason = get_reason(regs);
1376 
1377 	/* We can now get here via a FP Unavailable exception if the core
1378 	 * has no FPU, in that case the reason flags will be 0 */
1379 
1380 	if (reason & REASON_FP) {
1381 		/* IEEE FP exception */
1382 		parse_fpe(regs);
1383 		goto bail;
1384 	}
1385 	if (reason & REASON_TRAP) {
1386 		unsigned long bugaddr;
1387 		/* Debugger is first in line to stop recursive faults in
1388 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1389 		if (debugger_bpt(regs))
1390 			goto bail;
1391 
1392 		if (kprobe_handler(regs))
1393 			goto bail;
1394 
1395 		/* trap exception */
1396 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1397 				== NOTIFY_STOP)
1398 			goto bail;
1399 
1400 		bugaddr = regs->nip;
1401 		/*
1402 		 * Fixup bugaddr for BUG_ON() in real mode
1403 		 */
1404 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1405 			bugaddr += PAGE_OFFSET;
1406 
1407 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1408 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1409 			regs->nip += 4;
1410 			goto bail;
1411 		}
1412 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1413 		goto bail;
1414 	}
1415 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1416 	if (reason & REASON_TM) {
1417 		/* This is a TM "Bad Thing Exception" program check.
1418 		 * This occurs when:
1419 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1420 		 *    transition in TM states.
1421 		 * -  A trechkpt is attempted when transactional.
1422 		 * -  A treclaim is attempted when non transactional.
1423 		 * -  A tend is illegally attempted.
1424 		 * -  writing a TM SPR when transactional.
1425 		 *
1426 		 * If usermode caused this, it's done something illegal and
1427 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1428 		 * operand to distinguish from the instruction just being bad
1429 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1430 		 * illegal /placement/ of a valid instruction.
1431 		 */
1432 		if (user_mode(regs)) {
1433 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1434 			goto bail;
1435 		} else {
1436 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1437 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1438 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1439 			die("Unrecoverable exception", regs, SIGABRT);
1440 		}
1441 	}
1442 #endif
1443 
1444 	/*
1445 	 * If we took the program check in the kernel skip down to sending a
1446 	 * SIGILL. The subsequent cases all relate to emulating instructions
1447 	 * which we should only do for userspace. We also do not want to enable
1448 	 * interrupts for kernel faults because that might lead to further
1449 	 * faults, and loose the context of the original exception.
1450 	 */
1451 	if (!user_mode(regs))
1452 		goto sigill;
1453 
1454 	/* We restore the interrupt state now */
1455 	if (!arch_irq_disabled_regs(regs))
1456 		local_irq_enable();
1457 
1458 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1459 	 * but there seems to be a hardware bug on the 405GP (RevD)
1460 	 * that means ESR is sometimes set incorrectly - either to
1461 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1462 	 * hardware people - not sure if it can happen on any illegal
1463 	 * instruction or only on FP instructions, whether there is a
1464 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1465 	 */
1466 	if (!emulate_math(regs))
1467 		goto bail;
1468 
1469 	/* Try to emulate it if we should. */
1470 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1471 		switch (emulate_instruction(regs)) {
1472 		case 0:
1473 			regs->nip += 4;
1474 			emulate_single_step(regs);
1475 			goto bail;
1476 		case -EFAULT:
1477 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1478 			goto bail;
1479 		}
1480 	}
1481 
1482 sigill:
1483 	if (reason & REASON_PRIVILEGED)
1484 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1485 	else
1486 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1487 
1488 bail:
1489 	exception_exit(prev_state);
1490 }
1491 NOKPROBE_SYMBOL(program_check_exception);
1492 
1493 /*
1494  * This occurs when running in hypervisor mode on POWER6 or later
1495  * and an illegal instruction is encountered.
1496  */
1497 void emulation_assist_interrupt(struct pt_regs *regs)
1498 {
1499 	regs->msr |= REASON_ILLEGAL;
1500 	program_check_exception(regs);
1501 }
1502 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1503 
1504 void alignment_exception(struct pt_regs *regs)
1505 {
1506 	enum ctx_state prev_state = exception_enter();
1507 	int sig, code, fixed = 0;
1508 
1509 	/* We restore the interrupt state now */
1510 	if (!arch_irq_disabled_regs(regs))
1511 		local_irq_enable();
1512 
1513 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1514 		goto bail;
1515 
1516 	/* we don't implement logging of alignment exceptions */
1517 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1518 		fixed = fix_alignment(regs);
1519 
1520 	if (fixed == 1) {
1521 		regs->nip += 4;	/* skip over emulated instruction */
1522 		emulate_single_step(regs);
1523 		goto bail;
1524 	}
1525 
1526 	/* Operand address was bad */
1527 	if (fixed == -EFAULT) {
1528 		sig = SIGSEGV;
1529 		code = SEGV_ACCERR;
1530 	} else {
1531 		sig = SIGBUS;
1532 		code = BUS_ADRALN;
1533 	}
1534 	if (user_mode(regs))
1535 		_exception(sig, regs, code, regs->dar);
1536 	else
1537 		bad_page_fault(regs, regs->dar, sig);
1538 
1539 bail:
1540 	exception_exit(prev_state);
1541 }
1542 
1543 void StackOverflow(struct pt_regs *regs)
1544 {
1545 	printk(KERN_CRIT "Kernel stack overflow in process %p, r1=%lx\n",
1546 	       current, regs->gpr[1]);
1547 	debugger(regs);
1548 	show_regs(regs);
1549 	panic("kernel stack overflow");
1550 }
1551 
1552 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1553 {
1554 	enum ctx_state prev_state = exception_enter();
1555 
1556 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1557 			  "%lx at %lx\n", regs->trap, regs->nip);
1558 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1559 
1560 	exception_exit(prev_state);
1561 }
1562 
1563 void altivec_unavailable_exception(struct pt_regs *regs)
1564 {
1565 	enum ctx_state prev_state = exception_enter();
1566 
1567 	if (user_mode(regs)) {
1568 		/* A user program has executed an altivec instruction,
1569 		   but this kernel doesn't support altivec. */
1570 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1571 		goto bail;
1572 	}
1573 
1574 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1575 			"%lx at %lx\n", regs->trap, regs->nip);
1576 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1577 
1578 bail:
1579 	exception_exit(prev_state);
1580 }
1581 
1582 void vsx_unavailable_exception(struct pt_regs *regs)
1583 {
1584 	if (user_mode(regs)) {
1585 		/* A user program has executed an vsx instruction,
1586 		   but this kernel doesn't support vsx. */
1587 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1588 		return;
1589 	}
1590 
1591 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1592 			"%lx at %lx\n", regs->trap, regs->nip);
1593 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1594 }
1595 
1596 #ifdef CONFIG_PPC64
1597 static void tm_unavailable(struct pt_regs *regs)
1598 {
1599 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1600 	if (user_mode(regs)) {
1601 		current->thread.load_tm++;
1602 		regs->msr |= MSR_TM;
1603 		tm_enable();
1604 		tm_restore_sprs(&current->thread);
1605 		return;
1606 	}
1607 #endif
1608 	pr_emerg("Unrecoverable TM Unavailable Exception "
1609 			"%lx at %lx\n", regs->trap, regs->nip);
1610 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1611 }
1612 
1613 void facility_unavailable_exception(struct pt_regs *regs)
1614 {
1615 	static char *facility_strings[] = {
1616 		[FSCR_FP_LG] = "FPU",
1617 		[FSCR_VECVSX_LG] = "VMX/VSX",
1618 		[FSCR_DSCR_LG] = "DSCR",
1619 		[FSCR_PM_LG] = "PMU SPRs",
1620 		[FSCR_BHRB_LG] = "BHRB",
1621 		[FSCR_TM_LG] = "TM",
1622 		[FSCR_EBB_LG] = "EBB",
1623 		[FSCR_TAR_LG] = "TAR",
1624 		[FSCR_MSGP_LG] = "MSGP",
1625 		[FSCR_SCV_LG] = "SCV",
1626 	};
1627 	char *facility = "unknown";
1628 	u64 value;
1629 	u32 instword, rd;
1630 	u8 status;
1631 	bool hv;
1632 
1633 	hv = (TRAP(regs) == 0xf80);
1634 	if (hv)
1635 		value = mfspr(SPRN_HFSCR);
1636 	else
1637 		value = mfspr(SPRN_FSCR);
1638 
1639 	status = value >> 56;
1640 	if ((hv || status >= 2) &&
1641 	    (status < ARRAY_SIZE(facility_strings)) &&
1642 	    facility_strings[status])
1643 		facility = facility_strings[status];
1644 
1645 	/* We should not have taken this interrupt in kernel */
1646 	if (!user_mode(regs)) {
1647 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1648 			 facility, status, regs->nip);
1649 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1650 	}
1651 
1652 	/* We restore the interrupt state now */
1653 	if (!arch_irq_disabled_regs(regs))
1654 		local_irq_enable();
1655 
1656 	if (status == FSCR_DSCR_LG) {
1657 		/*
1658 		 * User is accessing the DSCR register using the problem
1659 		 * state only SPR number (0x03) either through a mfspr or
1660 		 * a mtspr instruction. If it is a write attempt through
1661 		 * a mtspr, then we set the inherit bit. This also allows
1662 		 * the user to write or read the register directly in the
1663 		 * future by setting via the FSCR DSCR bit. But in case it
1664 		 * is a read DSCR attempt through a mfspr instruction, we
1665 		 * just emulate the instruction instead. This code path will
1666 		 * always emulate all the mfspr instructions till the user
1667 		 * has attempted at least one mtspr instruction. This way it
1668 		 * preserves the same behaviour when the user is accessing
1669 		 * the DSCR through privilege level only SPR number (0x11)
1670 		 * which is emulated through illegal instruction exception.
1671 		 * We always leave HFSCR DSCR set.
1672 		 */
1673 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1674 			pr_err("Failed to fetch the user instruction\n");
1675 			return;
1676 		}
1677 
1678 		/* Write into DSCR (mtspr 0x03, RS) */
1679 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1680 				== PPC_INST_MTSPR_DSCR_USER) {
1681 			rd = (instword >> 21) & 0x1f;
1682 			current->thread.dscr = regs->gpr[rd];
1683 			current->thread.dscr_inherit = 1;
1684 			current->thread.fscr |= FSCR_DSCR;
1685 			mtspr(SPRN_FSCR, current->thread.fscr);
1686 		}
1687 
1688 		/* Read from DSCR (mfspr RT, 0x03) */
1689 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1690 				== PPC_INST_MFSPR_DSCR_USER) {
1691 			if (emulate_instruction(regs)) {
1692 				pr_err("DSCR based mfspr emulation failed\n");
1693 				return;
1694 			}
1695 			regs->nip += 4;
1696 			emulate_single_step(regs);
1697 		}
1698 		return;
1699 	}
1700 
1701 	if (status == FSCR_TM_LG) {
1702 		/*
1703 		 * If we're here then the hardware is TM aware because it
1704 		 * generated an exception with FSRM_TM set.
1705 		 *
1706 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1707 		 * told us not to do TM, or the kernel is not built with TM
1708 		 * support.
1709 		 *
1710 		 * If both of those things are true, then userspace can spam the
1711 		 * console by triggering the printk() below just by continually
1712 		 * doing tbegin (or any TM instruction). So in that case just
1713 		 * send the process a SIGILL immediately.
1714 		 */
1715 		if (!cpu_has_feature(CPU_FTR_TM))
1716 			goto out;
1717 
1718 		tm_unavailable(regs);
1719 		return;
1720 	}
1721 
1722 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1723 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1724 
1725 out:
1726 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1727 }
1728 #endif
1729 
1730 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1731 
1732 void fp_unavailable_tm(struct pt_regs *regs)
1733 {
1734 	/* Note:  This does not handle any kind of FP laziness. */
1735 
1736 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1737 		 regs->nip, regs->msr);
1738 
1739         /* We can only have got here if the task started using FP after
1740          * beginning the transaction.  So, the transactional regs are just a
1741          * copy of the checkpointed ones.  But, we still need to recheckpoint
1742          * as we're enabling FP for the process; it will return, abort the
1743          * transaction, and probably retry but now with FP enabled.  So the
1744          * checkpointed FP registers need to be loaded.
1745 	 */
1746 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1747 
1748 	/*
1749 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1750 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1751 	 *
1752 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1753 	 * recheckpoint.
1754 	 */
1755 
1756 	/* Enable FP for the task: */
1757 	current->thread.load_fp = 1;
1758 
1759 	/*
1760 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1761 	 */
1762 	tm_recheckpoint(&current->thread);
1763 }
1764 
1765 void altivec_unavailable_tm(struct pt_regs *regs)
1766 {
1767 	/* See the comments in fp_unavailable_tm().  This function operates
1768 	 * the same way.
1769 	 */
1770 
1771 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1772 		 "MSR=%lx\n",
1773 		 regs->nip, regs->msr);
1774 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1775 	current->thread.load_vec = 1;
1776 	tm_recheckpoint(&current->thread);
1777 	current->thread.used_vr = 1;
1778 }
1779 
1780 void vsx_unavailable_tm(struct pt_regs *regs)
1781 {
1782 	/* See the comments in fp_unavailable_tm().  This works similarly,
1783 	 * though we're loading both FP and VEC registers in here.
1784 	 *
1785 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1786 	 * regs.  Either way, set MSR_VSX.
1787 	 */
1788 
1789 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1790 		 "MSR=%lx\n",
1791 		 regs->nip, regs->msr);
1792 
1793 	current->thread.used_vsr = 1;
1794 
1795 	/* This reclaims FP and/or VR regs if they're already enabled */
1796 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1797 
1798 	current->thread.load_vec = 1;
1799 	current->thread.load_fp = 1;
1800 
1801 	tm_recheckpoint(&current->thread);
1802 }
1803 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1804 
1805 void performance_monitor_exception(struct pt_regs *regs)
1806 {
1807 	__this_cpu_inc(irq_stat.pmu_irqs);
1808 
1809 	perf_irq(regs);
1810 }
1811 
1812 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1813 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1814 {
1815 	int changed = 0;
1816 	/*
1817 	 * Determine the cause of the debug event, clear the
1818 	 * event flags and send a trap to the handler. Torez
1819 	 */
1820 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1821 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1822 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1823 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1824 #endif
1825 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1826 			     5);
1827 		changed |= 0x01;
1828 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1829 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1830 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1831 			     6);
1832 		changed |= 0x01;
1833 	}  else if (debug_status & DBSR_IAC1) {
1834 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1835 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1836 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1837 			     1);
1838 		changed |= 0x01;
1839 	}  else if (debug_status & DBSR_IAC2) {
1840 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1841 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1842 			     2);
1843 		changed |= 0x01;
1844 	}  else if (debug_status & DBSR_IAC3) {
1845 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1846 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1847 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1848 			     3);
1849 		changed |= 0x01;
1850 	}  else if (debug_status & DBSR_IAC4) {
1851 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1852 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1853 			     4);
1854 		changed |= 0x01;
1855 	}
1856 	/*
1857 	 * At the point this routine was called, the MSR(DE) was turned off.
1858 	 * Check all other debug flags and see if that bit needs to be turned
1859 	 * back on or not.
1860 	 */
1861 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1862 			       current->thread.debug.dbcr1))
1863 		regs->msr |= MSR_DE;
1864 	else
1865 		/* Make sure the IDM flag is off */
1866 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1867 
1868 	if (changed & 0x01)
1869 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1870 }
1871 
1872 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1873 {
1874 	current->thread.debug.dbsr = debug_status;
1875 
1876 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1877 	 * on server, it stops on the target of the branch. In order to simulate
1878 	 * the server behaviour, we thus restart right away with a single step
1879 	 * instead of stopping here when hitting a BT
1880 	 */
1881 	if (debug_status & DBSR_BT) {
1882 		regs->msr &= ~MSR_DE;
1883 
1884 		/* Disable BT */
1885 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1886 		/* Clear the BT event */
1887 		mtspr(SPRN_DBSR, DBSR_BT);
1888 
1889 		/* Do the single step trick only when coming from userspace */
1890 		if (user_mode(regs)) {
1891 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1892 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1893 			regs->msr |= MSR_DE;
1894 			return;
1895 		}
1896 
1897 		if (kprobe_post_handler(regs))
1898 			return;
1899 
1900 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1901 			       5, SIGTRAP) == NOTIFY_STOP) {
1902 			return;
1903 		}
1904 		if (debugger_sstep(regs))
1905 			return;
1906 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1907 		regs->msr &= ~MSR_DE;
1908 
1909 		/* Disable instruction completion */
1910 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
1911 		/* Clear the instruction completion event */
1912 		mtspr(SPRN_DBSR, DBSR_IC);
1913 
1914 		if (kprobe_post_handler(regs))
1915 			return;
1916 
1917 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1918 			       5, SIGTRAP) == NOTIFY_STOP) {
1919 			return;
1920 		}
1921 
1922 		if (debugger_sstep(regs))
1923 			return;
1924 
1925 		if (user_mode(regs)) {
1926 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
1927 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1928 					       current->thread.debug.dbcr1))
1929 				regs->msr |= MSR_DE;
1930 			else
1931 				/* Make sure the IDM bit is off */
1932 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1933 		}
1934 
1935 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1936 	} else
1937 		handle_debug(regs, debug_status);
1938 }
1939 NOKPROBE_SYMBOL(DebugException);
1940 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
1941 
1942 #if !defined(CONFIG_TAU_INT)
1943 void TAUException(struct pt_regs *regs)
1944 {
1945 	printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
1946 	       regs->nip, regs->msr, regs->trap, print_tainted());
1947 }
1948 #endif /* CONFIG_INT_TAU */
1949 
1950 #ifdef CONFIG_ALTIVEC
1951 void altivec_assist_exception(struct pt_regs *regs)
1952 {
1953 	int err;
1954 
1955 	if (!user_mode(regs)) {
1956 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
1957 		       " at %lx\n", regs->nip);
1958 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
1959 	}
1960 
1961 	flush_altivec_to_thread(current);
1962 
1963 	PPC_WARN_EMULATED(altivec, regs);
1964 	err = emulate_altivec(regs);
1965 	if (err == 0) {
1966 		regs->nip += 4;		/* skip emulated instruction */
1967 		emulate_single_step(regs);
1968 		return;
1969 	}
1970 
1971 	if (err == -EFAULT) {
1972 		/* got an error reading the instruction */
1973 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
1974 	} else {
1975 		/* didn't recognize the instruction */
1976 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
1977 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
1978 				   "in %s at %lx\n", current->comm, regs->nip);
1979 		current->thread.vr_state.vscr.u[3] |= 0x10000;
1980 	}
1981 }
1982 #endif /* CONFIG_ALTIVEC */
1983 
1984 #ifdef CONFIG_FSL_BOOKE
1985 void CacheLockingException(struct pt_regs *regs, unsigned long address,
1986 			   unsigned long error_code)
1987 {
1988 	/* We treat cache locking instructions from the user
1989 	 * as priv ops, in the future we could try to do
1990 	 * something smarter
1991 	 */
1992 	if (error_code & (ESR_DLK|ESR_ILK))
1993 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1994 	return;
1995 }
1996 #endif /* CONFIG_FSL_BOOKE */
1997 
1998 #ifdef CONFIG_SPE
1999 void SPEFloatingPointException(struct pt_regs *regs)
2000 {
2001 	extern int do_spe_mathemu(struct pt_regs *regs);
2002 	unsigned long spefscr;
2003 	int fpexc_mode;
2004 	int code = FPE_FLTUNK;
2005 	int err;
2006 
2007 	flush_spe_to_thread(current);
2008 
2009 	spefscr = current->thread.spefscr;
2010 	fpexc_mode = current->thread.fpexc_mode;
2011 
2012 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2013 		code = FPE_FLTOVF;
2014 	}
2015 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2016 		code = FPE_FLTUND;
2017 	}
2018 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2019 		code = FPE_FLTDIV;
2020 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2021 		code = FPE_FLTINV;
2022 	}
2023 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2024 		code = FPE_FLTRES;
2025 
2026 	err = do_spe_mathemu(regs);
2027 	if (err == 0) {
2028 		regs->nip += 4;		/* skip emulated instruction */
2029 		emulate_single_step(regs);
2030 		return;
2031 	}
2032 
2033 	if (err == -EFAULT) {
2034 		/* got an error reading the instruction */
2035 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2036 	} else if (err == -EINVAL) {
2037 		/* didn't recognize the instruction */
2038 		printk(KERN_ERR "unrecognized spe instruction "
2039 		       "in %s at %lx\n", current->comm, regs->nip);
2040 	} else {
2041 		_exception(SIGFPE, regs, code, regs->nip);
2042 	}
2043 
2044 	return;
2045 }
2046 
2047 void SPEFloatingPointRoundException(struct pt_regs *regs)
2048 {
2049 	extern int speround_handler(struct pt_regs *regs);
2050 	int err;
2051 
2052 	preempt_disable();
2053 	if (regs->msr & MSR_SPE)
2054 		giveup_spe(current);
2055 	preempt_enable();
2056 
2057 	regs->nip -= 4;
2058 	err = speround_handler(regs);
2059 	if (err == 0) {
2060 		regs->nip += 4;		/* skip emulated instruction */
2061 		emulate_single_step(regs);
2062 		return;
2063 	}
2064 
2065 	if (err == -EFAULT) {
2066 		/* got an error reading the instruction */
2067 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2068 	} else if (err == -EINVAL) {
2069 		/* didn't recognize the instruction */
2070 		printk(KERN_ERR "unrecognized spe instruction "
2071 		       "in %s at %lx\n", current->comm, regs->nip);
2072 	} else {
2073 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2074 		return;
2075 	}
2076 }
2077 #endif
2078 
2079 /*
2080  * We enter here if we get an unrecoverable exception, that is, one
2081  * that happened at a point where the RI (recoverable interrupt) bit
2082  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2083  * we therefore lost state by taking this exception.
2084  */
2085 void unrecoverable_exception(struct pt_regs *regs)
2086 {
2087 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2088 		 regs->trap, regs->nip, regs->msr);
2089 	die("Unrecoverable exception", regs, SIGABRT);
2090 }
2091 NOKPROBE_SYMBOL(unrecoverable_exception);
2092 
2093 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2094 /*
2095  * Default handler for a Watchdog exception,
2096  * spins until a reboot occurs
2097  */
2098 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2099 {
2100 	/* Generic WatchdogHandler, implement your own */
2101 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2102 	return;
2103 }
2104 
2105 void WatchdogException(struct pt_regs *regs)
2106 {
2107 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2108 	WatchdogHandler(regs);
2109 }
2110 #endif
2111 
2112 /*
2113  * We enter here if we discover during exception entry that we are
2114  * running in supervisor mode with a userspace value in the stack pointer.
2115  */
2116 void kernel_bad_stack(struct pt_regs *regs)
2117 {
2118 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2119 	       regs->gpr[1], regs->nip);
2120 	die("Bad kernel stack pointer", regs, SIGABRT);
2121 }
2122 NOKPROBE_SYMBOL(kernel_bad_stack);
2123 
2124 void __init trap_init(void)
2125 {
2126 }
2127 
2128 
2129 #ifdef CONFIG_PPC_EMULATED_STATS
2130 
2131 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2132 
2133 struct ppc_emulated ppc_emulated = {
2134 #ifdef CONFIG_ALTIVEC
2135 	WARN_EMULATED_SETUP(altivec),
2136 #endif
2137 	WARN_EMULATED_SETUP(dcba),
2138 	WARN_EMULATED_SETUP(dcbz),
2139 	WARN_EMULATED_SETUP(fp_pair),
2140 	WARN_EMULATED_SETUP(isel),
2141 	WARN_EMULATED_SETUP(mcrxr),
2142 	WARN_EMULATED_SETUP(mfpvr),
2143 	WARN_EMULATED_SETUP(multiple),
2144 	WARN_EMULATED_SETUP(popcntb),
2145 	WARN_EMULATED_SETUP(spe),
2146 	WARN_EMULATED_SETUP(string),
2147 	WARN_EMULATED_SETUP(sync),
2148 	WARN_EMULATED_SETUP(unaligned),
2149 #ifdef CONFIG_MATH_EMULATION
2150 	WARN_EMULATED_SETUP(math),
2151 #endif
2152 #ifdef CONFIG_VSX
2153 	WARN_EMULATED_SETUP(vsx),
2154 #endif
2155 #ifdef CONFIG_PPC64
2156 	WARN_EMULATED_SETUP(mfdscr),
2157 	WARN_EMULATED_SETUP(mtdscr),
2158 	WARN_EMULATED_SETUP(lq_stq),
2159 	WARN_EMULATED_SETUP(lxvw4x),
2160 	WARN_EMULATED_SETUP(lxvh8x),
2161 	WARN_EMULATED_SETUP(lxvd2x),
2162 	WARN_EMULATED_SETUP(lxvb16x),
2163 #endif
2164 };
2165 
2166 u32 ppc_warn_emulated;
2167 
2168 void ppc_warn_emulated_print(const char *type)
2169 {
2170 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2171 			    type);
2172 }
2173 
2174 static int __init ppc_warn_emulated_init(void)
2175 {
2176 	struct dentry *dir, *d;
2177 	unsigned int i;
2178 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2179 
2180 	if (!powerpc_debugfs_root)
2181 		return -ENODEV;
2182 
2183 	dir = debugfs_create_dir("emulated_instructions",
2184 				 powerpc_debugfs_root);
2185 	if (!dir)
2186 		return -ENOMEM;
2187 
2188 	d = debugfs_create_u32("do_warn", 0644, dir,
2189 			       &ppc_warn_emulated);
2190 	if (!d)
2191 		goto fail;
2192 
2193 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) {
2194 		d = debugfs_create_u32(entries[i].name, 0644, dir,
2195 				       (u32 *)&entries[i].val.counter);
2196 		if (!d)
2197 			goto fail;
2198 	}
2199 
2200 	return 0;
2201 
2202 fail:
2203 	debugfs_remove_recursive(dir);
2204 	return -ENOMEM;
2205 }
2206 
2207 device_initcall(ppc_warn_emulated_init);
2208 
2209 #endif /* CONFIG_PPC_EMULATED_STATS */
2210