xref: /linux/arch/powerpc/kernel/traps.c (revision beb4f4722cf60d9f0803054dec4eb5025f2cf594)
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 	printk("%s PAGE_SIZE=%luK%s%s%s%s%s%s%s %s\n",
261 	       IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE",
262 	       PAGE_SIZE / 1024,
263 	       early_radix_enabled() ? " MMU=Radix" : "",
264 	       early_mmu_has_feature(MMU_FTR_HPTE_TABLE) ? " MMU=Hash" : "",
265 	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
266 	       IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
267 	       IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "",
268 	       debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
269 	       IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "",
270 	       ppc_md.name ? ppc_md.name : "");
271 
272 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
273 		return 1;
274 
275 	print_modules();
276 	show_regs(regs);
277 
278 	return 0;
279 }
280 NOKPROBE_SYMBOL(__die);
281 
282 void die(const char *str, struct pt_regs *regs, long err)
283 {
284 	unsigned long flags;
285 
286 	/*
287 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
288 	 */
289 	if (TRAP(regs) != 0x100) {
290 		if (debugger(regs))
291 			return;
292 	}
293 
294 	flags = oops_begin(regs);
295 	if (__die(str, regs, err))
296 		err = 0;
297 	oops_end(flags, regs, err);
298 }
299 NOKPROBE_SYMBOL(die);
300 
301 void user_single_step_report(struct pt_regs *regs)
302 {
303 	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip, current);
304 }
305 
306 static void show_signal_msg(int signr, struct pt_regs *regs, int code,
307 			    unsigned long addr)
308 {
309 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
310 				      DEFAULT_RATELIMIT_BURST);
311 
312 	if (!show_unhandled_signals)
313 		return;
314 
315 	if (!unhandled_signal(current, signr))
316 		return;
317 
318 	if (!__ratelimit(&rs))
319 		return;
320 
321 	pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x",
322 		current->comm, current->pid, signame(signr), signr,
323 		addr, regs->nip, regs->link, code);
324 
325 	print_vma_addr(KERN_CONT " in ", regs->nip);
326 
327 	pr_cont("\n");
328 
329 	show_user_instructions(regs);
330 }
331 
332 static bool exception_common(int signr, struct pt_regs *regs, int code,
333 			      unsigned long addr)
334 {
335 	if (!user_mode(regs)) {
336 		die("Exception in kernel mode", regs, signr);
337 		return false;
338 	}
339 
340 	show_signal_msg(signr, regs, code, addr);
341 
342 	if (arch_irqs_disabled() && !arch_irq_disabled_regs(regs))
343 		local_irq_enable();
344 
345 	current->thread.trap_nr = code;
346 
347 	/*
348 	 * Save all the pkey registers AMR/IAMR/UAMOR. Eg: Core dumps need
349 	 * to capture the content, if the task gets killed.
350 	 */
351 	thread_pkey_regs_save(&current->thread);
352 
353 	return true;
354 }
355 
356 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key)
357 {
358 	if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr))
359 		return;
360 
361 	force_sig_pkuerr((void __user *) addr, key);
362 }
363 
364 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
365 {
366 	if (!exception_common(signr, regs, code, addr))
367 		return;
368 
369 	force_sig_fault(signr, code, (void __user *)addr, current);
370 }
371 
372 void system_reset_exception(struct pt_regs *regs)
373 {
374 	/*
375 	 * Avoid crashes in case of nested NMI exceptions. Recoverability
376 	 * is determined by RI and in_nmi
377 	 */
378 	bool nested = in_nmi();
379 	if (!nested)
380 		nmi_enter();
381 
382 	__this_cpu_inc(irq_stat.sreset_irqs);
383 
384 	/* See if any machine dependent calls */
385 	if (ppc_md.system_reset_exception) {
386 		if (ppc_md.system_reset_exception(regs))
387 			goto out;
388 	}
389 
390 	if (debugger(regs))
391 		goto out;
392 
393 	/*
394 	 * A system reset is a request to dump, so we always send
395 	 * it through the crashdump code (if fadump or kdump are
396 	 * registered).
397 	 */
398 	crash_fadump(regs, "System Reset");
399 
400 	crash_kexec(regs);
401 
402 	/*
403 	 * We aren't the primary crash CPU. We need to send it
404 	 * to a holding pattern to avoid it ending up in the panic
405 	 * code.
406 	 */
407 	crash_kexec_secondary(regs);
408 
409 	/*
410 	 * No debugger or crash dump registered, print logs then
411 	 * panic.
412 	 */
413 	die("System Reset", regs, SIGABRT);
414 
415 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
416 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
417 	nmi_panic(regs, "System Reset");
418 
419 out:
420 #ifdef CONFIG_PPC_BOOK3S_64
421 	BUG_ON(get_paca()->in_nmi == 0);
422 	if (get_paca()->in_nmi > 1)
423 		nmi_panic(regs, "Unrecoverable nested System Reset");
424 #endif
425 	/* Must die if the interrupt is not recoverable */
426 	if (!(regs->msr & MSR_RI))
427 		nmi_panic(regs, "Unrecoverable System Reset");
428 
429 	if (!nested)
430 		nmi_exit();
431 
432 	/* What should we do here? We could issue a shutdown or hard reset. */
433 }
434 
435 /*
436  * I/O accesses can cause machine checks on powermacs.
437  * Check if the NIP corresponds to the address of a sync
438  * instruction for which there is an entry in the exception
439  * table.
440  * Note that the 601 only takes a machine check on TEA
441  * (transfer error ack) signal assertion, and does not
442  * set any of the top 16 bits of SRR1.
443  *  -- paulus.
444  */
445 static inline int check_io_access(struct pt_regs *regs)
446 {
447 #ifdef CONFIG_PPC32
448 	unsigned long msr = regs->msr;
449 	const struct exception_table_entry *entry;
450 	unsigned int *nip = (unsigned int *)regs->nip;
451 
452 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
453 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
454 		/*
455 		 * Check that it's a sync instruction, or somewhere
456 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
457 		 * As the address is in the exception table
458 		 * we should be able to read the instr there.
459 		 * For the debug message, we look at the preceding
460 		 * load or store.
461 		 */
462 		if (*nip == PPC_INST_NOP)
463 			nip -= 2;
464 		else if (*nip == PPC_INST_ISYNC)
465 			--nip;
466 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
467 			unsigned int rb;
468 
469 			--nip;
470 			rb = (*nip >> 11) & 0x1f;
471 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
472 			       (*nip & 0x100)? "OUT to": "IN from",
473 			       regs->gpr[rb] - _IO_BASE, nip);
474 			regs->msr |= MSR_RI;
475 			regs->nip = extable_fixup(entry);
476 			return 1;
477 		}
478 	}
479 #endif /* CONFIG_PPC32 */
480 	return 0;
481 }
482 
483 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
484 /* On 4xx, the reason for the machine check or program exception
485    is in the ESR. */
486 #define get_reason(regs)	((regs)->dsisr)
487 #define REASON_FP		ESR_FP
488 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
489 #define REASON_PRIVILEGED	ESR_PPR
490 #define REASON_TRAP		ESR_PTR
491 
492 /* single-step stuff */
493 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
494 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
495 #define clear_br_trace(regs)	do {} while(0)
496 #else
497 /* On non-4xx, the reason for the machine check or program
498    exception is in the MSR. */
499 #define get_reason(regs)	((regs)->msr)
500 #define REASON_TM		SRR1_PROGTM
501 #define REASON_FP		SRR1_PROGFPE
502 #define REASON_ILLEGAL		SRR1_PROGILL
503 #define REASON_PRIVILEGED	SRR1_PROGPRIV
504 #define REASON_TRAP		SRR1_PROGTRAP
505 
506 #define single_stepping(regs)	((regs)->msr & MSR_SE)
507 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
508 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
509 #endif
510 
511 #if defined(CONFIG_E500)
512 int machine_check_e500mc(struct pt_regs *regs)
513 {
514 	unsigned long mcsr = mfspr(SPRN_MCSR);
515 	unsigned long pvr = mfspr(SPRN_PVR);
516 	unsigned long reason = mcsr;
517 	int recoverable = 1;
518 
519 	if (reason & MCSR_LD) {
520 		recoverable = fsl_rio_mcheck_exception(regs);
521 		if (recoverable == 1)
522 			goto silent_out;
523 	}
524 
525 	printk("Machine check in kernel mode.\n");
526 	printk("Caused by (from MCSR=%lx): ", reason);
527 
528 	if (reason & MCSR_MCP)
529 		pr_cont("Machine Check Signal\n");
530 
531 	if (reason & MCSR_ICPERR) {
532 		pr_cont("Instruction Cache Parity Error\n");
533 
534 		/*
535 		 * This is recoverable by invalidating the i-cache.
536 		 */
537 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
538 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
539 			;
540 
541 		/*
542 		 * This will generally be accompanied by an instruction
543 		 * fetch error report -- only treat MCSR_IF as fatal
544 		 * if it wasn't due to an L1 parity error.
545 		 */
546 		reason &= ~MCSR_IF;
547 	}
548 
549 	if (reason & MCSR_DCPERR_MC) {
550 		pr_cont("Data Cache Parity Error\n");
551 
552 		/*
553 		 * In write shadow mode we auto-recover from the error, but it
554 		 * may still get logged and cause a machine check.  We should
555 		 * only treat the non-write shadow case as non-recoverable.
556 		 */
557 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
558 		 * is not implemented but L1 data cache always runs in write
559 		 * shadow mode. Hence on data cache parity errors HW will
560 		 * automatically invalidate the L1 Data Cache.
561 		 */
562 		if (PVR_VER(pvr) != PVR_VER_E6500) {
563 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
564 				recoverable = 0;
565 		}
566 	}
567 
568 	if (reason & MCSR_L2MMU_MHIT) {
569 		pr_cont("Hit on multiple TLB entries\n");
570 		recoverable = 0;
571 	}
572 
573 	if (reason & MCSR_NMI)
574 		pr_cont("Non-maskable interrupt\n");
575 
576 	if (reason & MCSR_IF) {
577 		pr_cont("Instruction Fetch Error Report\n");
578 		recoverable = 0;
579 	}
580 
581 	if (reason & MCSR_LD) {
582 		pr_cont("Load Error Report\n");
583 		recoverable = 0;
584 	}
585 
586 	if (reason & MCSR_ST) {
587 		pr_cont("Store Error Report\n");
588 		recoverable = 0;
589 	}
590 
591 	if (reason & MCSR_LDG) {
592 		pr_cont("Guarded Load Error Report\n");
593 		recoverable = 0;
594 	}
595 
596 	if (reason & MCSR_TLBSYNC)
597 		pr_cont("Simultaneous tlbsync operations\n");
598 
599 	if (reason & MCSR_BSL2_ERR) {
600 		pr_cont("Level 2 Cache Error\n");
601 		recoverable = 0;
602 	}
603 
604 	if (reason & MCSR_MAV) {
605 		u64 addr;
606 
607 		addr = mfspr(SPRN_MCAR);
608 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
609 
610 		pr_cont("Machine Check %s Address: %#llx\n",
611 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
612 	}
613 
614 silent_out:
615 	mtspr(SPRN_MCSR, mcsr);
616 	return mfspr(SPRN_MCSR) == 0 && recoverable;
617 }
618 
619 int machine_check_e500(struct pt_regs *regs)
620 {
621 	unsigned long reason = mfspr(SPRN_MCSR);
622 
623 	if (reason & MCSR_BUS_RBERR) {
624 		if (fsl_rio_mcheck_exception(regs))
625 			return 1;
626 		if (fsl_pci_mcheck_exception(regs))
627 			return 1;
628 	}
629 
630 	printk("Machine check in kernel mode.\n");
631 	printk("Caused by (from MCSR=%lx): ", reason);
632 
633 	if (reason & MCSR_MCP)
634 		pr_cont("Machine Check Signal\n");
635 	if (reason & MCSR_ICPERR)
636 		pr_cont("Instruction Cache Parity Error\n");
637 	if (reason & MCSR_DCP_PERR)
638 		pr_cont("Data Cache Push Parity Error\n");
639 	if (reason & MCSR_DCPERR)
640 		pr_cont("Data Cache Parity Error\n");
641 	if (reason & MCSR_BUS_IAERR)
642 		pr_cont("Bus - Instruction Address Error\n");
643 	if (reason & MCSR_BUS_RAERR)
644 		pr_cont("Bus - Read Address Error\n");
645 	if (reason & MCSR_BUS_WAERR)
646 		pr_cont("Bus - Write Address Error\n");
647 	if (reason & MCSR_BUS_IBERR)
648 		pr_cont("Bus - Instruction Data Error\n");
649 	if (reason & MCSR_BUS_RBERR)
650 		pr_cont("Bus - Read Data Bus Error\n");
651 	if (reason & MCSR_BUS_WBERR)
652 		pr_cont("Bus - Write Data Bus Error\n");
653 	if (reason & MCSR_BUS_IPERR)
654 		pr_cont("Bus - Instruction Parity Error\n");
655 	if (reason & MCSR_BUS_RPERR)
656 		pr_cont("Bus - Read Parity Error\n");
657 
658 	return 0;
659 }
660 
661 int machine_check_generic(struct pt_regs *regs)
662 {
663 	return 0;
664 }
665 #elif defined(CONFIG_E200)
666 int machine_check_e200(struct pt_regs *regs)
667 {
668 	unsigned long reason = mfspr(SPRN_MCSR);
669 
670 	printk("Machine check in kernel mode.\n");
671 	printk("Caused by (from MCSR=%lx): ", reason);
672 
673 	if (reason & MCSR_MCP)
674 		pr_cont("Machine Check Signal\n");
675 	if (reason & MCSR_CP_PERR)
676 		pr_cont("Cache Push Parity Error\n");
677 	if (reason & MCSR_CPERR)
678 		pr_cont("Cache Parity Error\n");
679 	if (reason & MCSR_EXCP_ERR)
680 		pr_cont("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
681 	if (reason & MCSR_BUS_IRERR)
682 		pr_cont("Bus - Read Bus Error on instruction fetch\n");
683 	if (reason & MCSR_BUS_DRERR)
684 		pr_cont("Bus - Read Bus Error on data load\n");
685 	if (reason & MCSR_BUS_WRERR)
686 		pr_cont("Bus - Write Bus Error on buffered store or cache line push\n");
687 
688 	return 0;
689 }
690 #elif defined(CONFIG_PPC32)
691 int machine_check_generic(struct pt_regs *regs)
692 {
693 	unsigned long reason = regs->msr;
694 
695 	printk("Machine check in kernel mode.\n");
696 	printk("Caused by (from SRR1=%lx): ", reason);
697 	switch (reason & 0x601F0000) {
698 	case 0x80000:
699 		pr_cont("Machine check signal\n");
700 		break;
701 	case 0:		/* for 601 */
702 	case 0x40000:
703 	case 0x140000:	/* 7450 MSS error and TEA */
704 		pr_cont("Transfer error ack signal\n");
705 		break;
706 	case 0x20000:
707 		pr_cont("Data parity error signal\n");
708 		break;
709 	case 0x10000:
710 		pr_cont("Address parity error signal\n");
711 		break;
712 	case 0x20000000:
713 		pr_cont("L1 Data Cache error\n");
714 		break;
715 	case 0x40000000:
716 		pr_cont("L1 Instruction Cache error\n");
717 		break;
718 	case 0x00100000:
719 		pr_cont("L2 data cache parity error\n");
720 		break;
721 	default:
722 		pr_cont("Unknown values in msr\n");
723 	}
724 	return 0;
725 }
726 #endif /* everything else */
727 
728 void machine_check_exception(struct pt_regs *regs)
729 {
730 	int recover = 0;
731 	bool nested = in_nmi();
732 	if (!nested)
733 		nmi_enter();
734 
735 	__this_cpu_inc(irq_stat.mce_exceptions);
736 
737 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
738 
739 	/* See if any machine dependent calls. In theory, we would want
740 	 * to call the CPU first, and call the ppc_md. one if the CPU
741 	 * one returns a positive number. However there is existing code
742 	 * that assumes the board gets a first chance, so let's keep it
743 	 * that way for now and fix things later. --BenH.
744 	 */
745 	if (ppc_md.machine_check_exception)
746 		recover = ppc_md.machine_check_exception(regs);
747 	else if (cur_cpu_spec->machine_check)
748 		recover = cur_cpu_spec->machine_check(regs);
749 
750 	if (recover > 0)
751 		goto bail;
752 
753 	if (debugger_fault_handler(regs))
754 		goto bail;
755 
756 	if (check_io_access(regs))
757 		goto bail;
758 
759 	/* Must die if the interrupt is not recoverable */
760 	if (!(regs->msr & MSR_RI))
761 		nmi_panic(regs, "Unrecoverable Machine check");
762 
763 	if (!nested)
764 		nmi_exit();
765 
766 	die("Machine check", regs, SIGBUS);
767 
768 	return;
769 
770 bail:
771 	if (!nested)
772 		nmi_exit();
773 }
774 
775 void SMIException(struct pt_regs *regs)
776 {
777 	die("System Management Interrupt", regs, SIGABRT);
778 }
779 
780 #ifdef CONFIG_VSX
781 static void p9_hmi_special_emu(struct pt_regs *regs)
782 {
783 	unsigned int ra, rb, t, i, sel, instr, rc;
784 	const void __user *addr;
785 	u8 vbuf[16], *vdst;
786 	unsigned long ea, msr, msr_mask;
787 	bool swap;
788 
789 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
790 		return;
791 
792 	/*
793 	 * lxvb16x	opcode: 0x7c0006d8
794 	 * lxvd2x	opcode: 0x7c000698
795 	 * lxvh8x	opcode: 0x7c000658
796 	 * lxvw4x	opcode: 0x7c000618
797 	 */
798 	if ((instr & 0xfc00073e) != 0x7c000618) {
799 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
800 			 " instr=%08x\n",
801 			 smp_processor_id(), current->comm, current->pid,
802 			 regs->nip, instr);
803 		return;
804 	}
805 
806 	/* Grab vector registers into the task struct */
807 	msr = regs->msr; /* Grab msr before we flush the bits */
808 	flush_vsx_to_thread(current);
809 	enable_kernel_altivec();
810 
811 	/*
812 	 * Is userspace running with a different endian (this is rare but
813 	 * not impossible)
814 	 */
815 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
816 
817 	/* Decode the instruction */
818 	ra = (instr >> 16) & 0x1f;
819 	rb = (instr >> 11) & 0x1f;
820 	t = (instr >> 21) & 0x1f;
821 	if (instr & 1)
822 		vdst = (u8 *)&current->thread.vr_state.vr[t];
823 	else
824 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
825 
826 	/* Grab the vector address */
827 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
828 	if (is_32bit_task())
829 		ea &= 0xfffffffful;
830 	addr = (__force const void __user *)ea;
831 
832 	/* Check it */
833 	if (!access_ok(addr, 16)) {
834 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
835 			 " instr=%08x addr=%016lx\n",
836 			 smp_processor_id(), current->comm, current->pid,
837 			 regs->nip, instr, (unsigned long)addr);
838 		return;
839 	}
840 
841 	/* Read the vector */
842 	rc = 0;
843 	if ((unsigned long)addr & 0xfUL)
844 		/* unaligned case */
845 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
846 	else
847 		__get_user_atomic_128_aligned(vbuf, addr, rc);
848 	if (rc) {
849 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
850 			 " instr=%08x addr=%016lx\n",
851 			 smp_processor_id(), current->comm, current->pid,
852 			 regs->nip, instr, (unsigned long)addr);
853 		return;
854 	}
855 
856 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
857 		 " instr=%08x addr=%016lx\n",
858 		 smp_processor_id(), current->comm, current->pid, regs->nip,
859 		 instr, (unsigned long) addr);
860 
861 	/* Grab instruction "selector" */
862 	sel = (instr >> 6) & 3;
863 
864 	/*
865 	 * Check to make sure the facility is actually enabled. This
866 	 * could happen if we get a false positive hit.
867 	 *
868 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
869 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
870 	 */
871 	msr_mask = MSR_VSX;
872 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
873 		msr_mask = MSR_VEC;
874 	if (!(msr & msr_mask)) {
875 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
876 			 " instr=%08x msr:%016lx\n",
877 			 smp_processor_id(), current->comm, current->pid,
878 			 regs->nip, instr, msr);
879 		return;
880 	}
881 
882 	/* Do logging here before we modify sel based on endian */
883 	switch (sel) {
884 	case 0:	/* lxvw4x */
885 		PPC_WARN_EMULATED(lxvw4x, regs);
886 		break;
887 	case 1: /* lxvh8x */
888 		PPC_WARN_EMULATED(lxvh8x, regs);
889 		break;
890 	case 2: /* lxvd2x */
891 		PPC_WARN_EMULATED(lxvd2x, regs);
892 		break;
893 	case 3: /* lxvb16x */
894 		PPC_WARN_EMULATED(lxvb16x, regs);
895 		break;
896 	}
897 
898 #ifdef __LITTLE_ENDIAN__
899 	/*
900 	 * An LE kernel stores the vector in the task struct as an LE
901 	 * byte array (effectively swapping both the components and
902 	 * the content of the components). Those instructions expect
903 	 * the components to remain in ascending address order, so we
904 	 * swap them back.
905 	 *
906 	 * If we are running a BE user space, the expectation is that
907 	 * of a simple memcpy, so forcing the emulation to look like
908 	 * a lxvb16x should do the trick.
909 	 */
910 	if (swap)
911 		sel = 3;
912 
913 	switch (sel) {
914 	case 0:	/* lxvw4x */
915 		for (i = 0; i < 4; i++)
916 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
917 		break;
918 	case 1: /* lxvh8x */
919 		for (i = 0; i < 8; i++)
920 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
921 		break;
922 	case 2: /* lxvd2x */
923 		for (i = 0; i < 2; i++)
924 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
925 		break;
926 	case 3: /* lxvb16x */
927 		for (i = 0; i < 16; i++)
928 			vdst[i] = vbuf[15-i];
929 		break;
930 	}
931 #else /* __LITTLE_ENDIAN__ */
932 	/* On a big endian kernel, a BE userspace only needs a memcpy */
933 	if (!swap)
934 		sel = 3;
935 
936 	/* Otherwise, we need to swap the content of the components */
937 	switch (sel) {
938 	case 0:	/* lxvw4x */
939 		for (i = 0; i < 4; i++)
940 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
941 		break;
942 	case 1: /* lxvh8x */
943 		for (i = 0; i < 8; i++)
944 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
945 		break;
946 	case 2: /* lxvd2x */
947 		for (i = 0; i < 2; i++)
948 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
949 		break;
950 	case 3: /* lxvb16x */
951 		memcpy(vdst, vbuf, 16);
952 		break;
953 	}
954 #endif /* !__LITTLE_ENDIAN__ */
955 
956 	/* Go to next instruction */
957 	regs->nip += 4;
958 }
959 #endif /* CONFIG_VSX */
960 
961 void handle_hmi_exception(struct pt_regs *regs)
962 {
963 	struct pt_regs *old_regs;
964 
965 	old_regs = set_irq_regs(regs);
966 	irq_enter();
967 
968 #ifdef CONFIG_VSX
969 	/* Real mode flagged P9 special emu is needed */
970 	if (local_paca->hmi_p9_special_emu) {
971 		local_paca->hmi_p9_special_emu = 0;
972 
973 		/*
974 		 * We don't want to take page faults while doing the
975 		 * emulation, we just replay the instruction if necessary.
976 		 */
977 		pagefault_disable();
978 		p9_hmi_special_emu(regs);
979 		pagefault_enable();
980 	}
981 #endif /* CONFIG_VSX */
982 
983 	if (ppc_md.handle_hmi_exception)
984 		ppc_md.handle_hmi_exception(regs);
985 
986 	irq_exit();
987 	set_irq_regs(old_regs);
988 }
989 
990 void unknown_exception(struct pt_regs *regs)
991 {
992 	enum ctx_state prev_state = exception_enter();
993 
994 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
995 	       regs->nip, regs->msr, regs->trap);
996 
997 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
998 
999 	exception_exit(prev_state);
1000 }
1001 
1002 void instruction_breakpoint_exception(struct pt_regs *regs)
1003 {
1004 	enum ctx_state prev_state = exception_enter();
1005 
1006 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1007 					5, SIGTRAP) == NOTIFY_STOP)
1008 		goto bail;
1009 	if (debugger_iabr_match(regs))
1010 		goto bail;
1011 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1012 
1013 bail:
1014 	exception_exit(prev_state);
1015 }
1016 
1017 void RunModeException(struct pt_regs *regs)
1018 {
1019 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1020 }
1021 
1022 void single_step_exception(struct pt_regs *regs)
1023 {
1024 	enum ctx_state prev_state = exception_enter();
1025 
1026 	clear_single_step(regs);
1027 	clear_br_trace(regs);
1028 
1029 	if (kprobe_post_handler(regs))
1030 		return;
1031 
1032 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1033 					5, SIGTRAP) == NOTIFY_STOP)
1034 		goto bail;
1035 	if (debugger_sstep(regs))
1036 		goto bail;
1037 
1038 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1039 
1040 bail:
1041 	exception_exit(prev_state);
1042 }
1043 NOKPROBE_SYMBOL(single_step_exception);
1044 
1045 /*
1046  * After we have successfully emulated an instruction, we have to
1047  * check if the instruction was being single-stepped, and if so,
1048  * pretend we got a single-step exception.  This was pointed out
1049  * by Kumar Gala.  -- paulus
1050  */
1051 static void emulate_single_step(struct pt_regs *regs)
1052 {
1053 	if (single_stepping(regs))
1054 		single_step_exception(regs);
1055 }
1056 
1057 static inline int __parse_fpscr(unsigned long fpscr)
1058 {
1059 	int ret = FPE_FLTUNK;
1060 
1061 	/* Invalid operation */
1062 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1063 		ret = FPE_FLTINV;
1064 
1065 	/* Overflow */
1066 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1067 		ret = FPE_FLTOVF;
1068 
1069 	/* Underflow */
1070 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1071 		ret = FPE_FLTUND;
1072 
1073 	/* Divide by zero */
1074 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1075 		ret = FPE_FLTDIV;
1076 
1077 	/* Inexact result */
1078 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1079 		ret = FPE_FLTRES;
1080 
1081 	return ret;
1082 }
1083 
1084 static void parse_fpe(struct pt_regs *regs)
1085 {
1086 	int code = 0;
1087 
1088 	flush_fp_to_thread(current);
1089 
1090 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1091 
1092 	_exception(SIGFPE, regs, code, regs->nip);
1093 }
1094 
1095 /*
1096  * Illegal instruction emulation support.  Originally written to
1097  * provide the PVR to user applications using the mfspr rd, PVR.
1098  * Return non-zero if we can't emulate, or -EFAULT if the associated
1099  * memory access caused an access fault.  Return zero on success.
1100  *
1101  * There are a couple of ways to do this, either "decode" the instruction
1102  * or directly match lots of bits.  In this case, matching lots of
1103  * bits is faster and easier.
1104  *
1105  */
1106 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1107 {
1108 	u8 rT = (instword >> 21) & 0x1f;
1109 	u8 rA = (instword >> 16) & 0x1f;
1110 	u8 NB_RB = (instword >> 11) & 0x1f;
1111 	u32 num_bytes;
1112 	unsigned long EA;
1113 	int pos = 0;
1114 
1115 	/* Early out if we are an invalid form of lswx */
1116 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1117 		if ((rT == rA) || (rT == NB_RB))
1118 			return -EINVAL;
1119 
1120 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1121 
1122 	switch (instword & PPC_INST_STRING_MASK) {
1123 		case PPC_INST_LSWX:
1124 		case PPC_INST_STSWX:
1125 			EA += NB_RB;
1126 			num_bytes = regs->xer & 0x7f;
1127 			break;
1128 		case PPC_INST_LSWI:
1129 		case PPC_INST_STSWI:
1130 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1131 			break;
1132 		default:
1133 			return -EINVAL;
1134 	}
1135 
1136 	while (num_bytes != 0)
1137 	{
1138 		u8 val;
1139 		u32 shift = 8 * (3 - (pos & 0x3));
1140 
1141 		/* if process is 32-bit, clear upper 32 bits of EA */
1142 		if ((regs->msr & MSR_64BIT) == 0)
1143 			EA &= 0xFFFFFFFF;
1144 
1145 		switch ((instword & PPC_INST_STRING_MASK)) {
1146 			case PPC_INST_LSWX:
1147 			case PPC_INST_LSWI:
1148 				if (get_user(val, (u8 __user *)EA))
1149 					return -EFAULT;
1150 				/* first time updating this reg,
1151 				 * zero it out */
1152 				if (pos == 0)
1153 					regs->gpr[rT] = 0;
1154 				regs->gpr[rT] |= val << shift;
1155 				break;
1156 			case PPC_INST_STSWI:
1157 			case PPC_INST_STSWX:
1158 				val = regs->gpr[rT] >> shift;
1159 				if (put_user(val, (u8 __user *)EA))
1160 					return -EFAULT;
1161 				break;
1162 		}
1163 		/* move EA to next address */
1164 		EA += 1;
1165 		num_bytes--;
1166 
1167 		/* manage our position within the register */
1168 		if (++pos == 4) {
1169 			pos = 0;
1170 			if (++rT == 32)
1171 				rT = 0;
1172 		}
1173 	}
1174 
1175 	return 0;
1176 }
1177 
1178 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1179 {
1180 	u32 ra,rs;
1181 	unsigned long tmp;
1182 
1183 	ra = (instword >> 16) & 0x1f;
1184 	rs = (instword >> 21) & 0x1f;
1185 
1186 	tmp = regs->gpr[rs];
1187 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1188 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1189 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1190 	regs->gpr[ra] = tmp;
1191 
1192 	return 0;
1193 }
1194 
1195 static int emulate_isel(struct pt_regs *regs, u32 instword)
1196 {
1197 	u8 rT = (instword >> 21) & 0x1f;
1198 	u8 rA = (instword >> 16) & 0x1f;
1199 	u8 rB = (instword >> 11) & 0x1f;
1200 	u8 BC = (instword >> 6) & 0x1f;
1201 	u8 bit;
1202 	unsigned long tmp;
1203 
1204 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1205 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1206 
1207 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1208 
1209 	return 0;
1210 }
1211 
1212 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1213 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1214 {
1215         /* If we're emulating a load/store in an active transaction, we cannot
1216          * emulate it as the kernel operates in transaction suspended context.
1217          * We need to abort the transaction.  This creates a persistent TM
1218          * abort so tell the user what caused it with a new code.
1219 	 */
1220 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1221 		tm_enable();
1222 		tm_abort(cause);
1223 		return true;
1224 	}
1225 	return false;
1226 }
1227 #else
1228 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1229 {
1230 	return false;
1231 }
1232 #endif
1233 
1234 static int emulate_instruction(struct pt_regs *regs)
1235 {
1236 	u32 instword;
1237 	u32 rd;
1238 
1239 	if (!user_mode(regs))
1240 		return -EINVAL;
1241 	CHECK_FULL_REGS(regs);
1242 
1243 	if (get_user(instword, (u32 __user *)(regs->nip)))
1244 		return -EFAULT;
1245 
1246 	/* Emulate the mfspr rD, PVR. */
1247 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1248 		PPC_WARN_EMULATED(mfpvr, regs);
1249 		rd = (instword >> 21) & 0x1f;
1250 		regs->gpr[rd] = mfspr(SPRN_PVR);
1251 		return 0;
1252 	}
1253 
1254 	/* Emulating the dcba insn is just a no-op.  */
1255 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1256 		PPC_WARN_EMULATED(dcba, regs);
1257 		return 0;
1258 	}
1259 
1260 	/* Emulate the mcrxr insn.  */
1261 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1262 		int shift = (instword >> 21) & 0x1c;
1263 		unsigned long msk = 0xf0000000UL >> shift;
1264 
1265 		PPC_WARN_EMULATED(mcrxr, regs);
1266 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1267 		regs->xer &= ~0xf0000000UL;
1268 		return 0;
1269 	}
1270 
1271 	/* Emulate load/store string insn. */
1272 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1273 		if (tm_abort_check(regs,
1274 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1275 			return -EINVAL;
1276 		PPC_WARN_EMULATED(string, regs);
1277 		return emulate_string_inst(regs, instword);
1278 	}
1279 
1280 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1281 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1282 		PPC_WARN_EMULATED(popcntb, regs);
1283 		return emulate_popcntb_inst(regs, instword);
1284 	}
1285 
1286 	/* Emulate isel (Integer Select) instruction */
1287 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1288 		PPC_WARN_EMULATED(isel, regs);
1289 		return emulate_isel(regs, instword);
1290 	}
1291 
1292 	/* Emulate sync instruction variants */
1293 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1294 		PPC_WARN_EMULATED(sync, regs);
1295 		asm volatile("sync");
1296 		return 0;
1297 	}
1298 
1299 #ifdef CONFIG_PPC64
1300 	/* Emulate the mfspr rD, DSCR. */
1301 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1302 		PPC_INST_MFSPR_DSCR_USER) ||
1303 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1304 		PPC_INST_MFSPR_DSCR)) &&
1305 			cpu_has_feature(CPU_FTR_DSCR)) {
1306 		PPC_WARN_EMULATED(mfdscr, regs);
1307 		rd = (instword >> 21) & 0x1f;
1308 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1309 		return 0;
1310 	}
1311 	/* Emulate the mtspr DSCR, rD. */
1312 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1313 		PPC_INST_MTSPR_DSCR_USER) ||
1314 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1315 		PPC_INST_MTSPR_DSCR)) &&
1316 			cpu_has_feature(CPU_FTR_DSCR)) {
1317 		PPC_WARN_EMULATED(mtdscr, regs);
1318 		rd = (instword >> 21) & 0x1f;
1319 		current->thread.dscr = regs->gpr[rd];
1320 		current->thread.dscr_inherit = 1;
1321 		mtspr(SPRN_DSCR, current->thread.dscr);
1322 		return 0;
1323 	}
1324 #endif
1325 
1326 	return -EINVAL;
1327 }
1328 
1329 int is_valid_bugaddr(unsigned long addr)
1330 {
1331 	return is_kernel_addr(addr);
1332 }
1333 
1334 #ifdef CONFIG_MATH_EMULATION
1335 static int emulate_math(struct pt_regs *regs)
1336 {
1337 	int ret;
1338 	extern int do_mathemu(struct pt_regs *regs);
1339 
1340 	ret = do_mathemu(regs);
1341 	if (ret >= 0)
1342 		PPC_WARN_EMULATED(math, regs);
1343 
1344 	switch (ret) {
1345 	case 0:
1346 		emulate_single_step(regs);
1347 		return 0;
1348 	case 1: {
1349 			int code = 0;
1350 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1351 			_exception(SIGFPE, regs, code, regs->nip);
1352 			return 0;
1353 		}
1354 	case -EFAULT:
1355 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1356 		return 0;
1357 	}
1358 
1359 	return -1;
1360 }
1361 #else
1362 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1363 #endif
1364 
1365 void program_check_exception(struct pt_regs *regs)
1366 {
1367 	enum ctx_state prev_state = exception_enter();
1368 	unsigned int reason = get_reason(regs);
1369 
1370 	/* We can now get here via a FP Unavailable exception if the core
1371 	 * has no FPU, in that case the reason flags will be 0 */
1372 
1373 	if (reason & REASON_FP) {
1374 		/* IEEE FP exception */
1375 		parse_fpe(regs);
1376 		goto bail;
1377 	}
1378 	if (reason & REASON_TRAP) {
1379 		unsigned long bugaddr;
1380 		/* Debugger is first in line to stop recursive faults in
1381 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1382 		if (debugger_bpt(regs))
1383 			goto bail;
1384 
1385 		if (kprobe_handler(regs))
1386 			goto bail;
1387 
1388 		/* trap exception */
1389 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1390 				== NOTIFY_STOP)
1391 			goto bail;
1392 
1393 		bugaddr = regs->nip;
1394 		/*
1395 		 * Fixup bugaddr for BUG_ON() in real mode
1396 		 */
1397 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1398 			bugaddr += PAGE_OFFSET;
1399 
1400 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1401 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1402 			regs->nip += 4;
1403 			goto bail;
1404 		}
1405 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1406 		goto bail;
1407 	}
1408 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1409 	if (reason & REASON_TM) {
1410 		/* This is a TM "Bad Thing Exception" program check.
1411 		 * This occurs when:
1412 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1413 		 *    transition in TM states.
1414 		 * -  A trechkpt is attempted when transactional.
1415 		 * -  A treclaim is attempted when non transactional.
1416 		 * -  A tend is illegally attempted.
1417 		 * -  writing a TM SPR when transactional.
1418 		 *
1419 		 * If usermode caused this, it's done something illegal and
1420 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1421 		 * operand to distinguish from the instruction just being bad
1422 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1423 		 * illegal /placement/ of a valid instruction.
1424 		 */
1425 		if (user_mode(regs)) {
1426 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1427 			goto bail;
1428 		} else {
1429 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1430 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1431 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1432 			die("Unrecoverable exception", regs, SIGABRT);
1433 		}
1434 	}
1435 #endif
1436 
1437 	/*
1438 	 * If we took the program check in the kernel skip down to sending a
1439 	 * SIGILL. The subsequent cases all relate to emulating instructions
1440 	 * which we should only do for userspace. We also do not want to enable
1441 	 * interrupts for kernel faults because that might lead to further
1442 	 * faults, and loose the context of the original exception.
1443 	 */
1444 	if (!user_mode(regs))
1445 		goto sigill;
1446 
1447 	/* We restore the interrupt state now */
1448 	if (!arch_irq_disabled_regs(regs))
1449 		local_irq_enable();
1450 
1451 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1452 	 * but there seems to be a hardware bug on the 405GP (RevD)
1453 	 * that means ESR is sometimes set incorrectly - either to
1454 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1455 	 * hardware people - not sure if it can happen on any illegal
1456 	 * instruction or only on FP instructions, whether there is a
1457 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1458 	 */
1459 	if (!emulate_math(regs))
1460 		goto bail;
1461 
1462 	/* Try to emulate it if we should. */
1463 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1464 		switch (emulate_instruction(regs)) {
1465 		case 0:
1466 			regs->nip += 4;
1467 			emulate_single_step(regs);
1468 			goto bail;
1469 		case -EFAULT:
1470 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1471 			goto bail;
1472 		}
1473 	}
1474 
1475 sigill:
1476 	if (reason & REASON_PRIVILEGED)
1477 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1478 	else
1479 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1480 
1481 bail:
1482 	exception_exit(prev_state);
1483 }
1484 NOKPROBE_SYMBOL(program_check_exception);
1485 
1486 /*
1487  * This occurs when running in hypervisor mode on POWER6 or later
1488  * and an illegal instruction is encountered.
1489  */
1490 void emulation_assist_interrupt(struct pt_regs *regs)
1491 {
1492 	regs->msr |= REASON_ILLEGAL;
1493 	program_check_exception(regs);
1494 }
1495 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1496 
1497 void alignment_exception(struct pt_regs *regs)
1498 {
1499 	enum ctx_state prev_state = exception_enter();
1500 	int sig, code, fixed = 0;
1501 
1502 	/* We restore the interrupt state now */
1503 	if (!arch_irq_disabled_regs(regs))
1504 		local_irq_enable();
1505 
1506 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1507 		goto bail;
1508 
1509 	/* we don't implement logging of alignment exceptions */
1510 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1511 		fixed = fix_alignment(regs);
1512 
1513 	if (fixed == 1) {
1514 		regs->nip += 4;	/* skip over emulated instruction */
1515 		emulate_single_step(regs);
1516 		goto bail;
1517 	}
1518 
1519 	/* Operand address was bad */
1520 	if (fixed == -EFAULT) {
1521 		sig = SIGSEGV;
1522 		code = SEGV_ACCERR;
1523 	} else {
1524 		sig = SIGBUS;
1525 		code = BUS_ADRALN;
1526 	}
1527 	if (user_mode(regs))
1528 		_exception(sig, regs, code, regs->dar);
1529 	else
1530 		bad_page_fault(regs, regs->dar, sig);
1531 
1532 bail:
1533 	exception_exit(prev_state);
1534 }
1535 
1536 void StackOverflow(struct pt_regs *regs)
1537 {
1538 	pr_crit("Kernel stack overflow in process %s[%d], r1=%lx\n",
1539 		current->comm, task_pid_nr(current), regs->gpr[1]);
1540 	debugger(regs);
1541 	show_regs(regs);
1542 	panic("kernel stack overflow");
1543 }
1544 
1545 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1546 {
1547 	enum ctx_state prev_state = exception_enter();
1548 
1549 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1550 			  "%lx at %lx\n", regs->trap, regs->nip);
1551 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1552 
1553 	exception_exit(prev_state);
1554 }
1555 
1556 void altivec_unavailable_exception(struct pt_regs *regs)
1557 {
1558 	enum ctx_state prev_state = exception_enter();
1559 
1560 	if (user_mode(regs)) {
1561 		/* A user program has executed an altivec instruction,
1562 		   but this kernel doesn't support altivec. */
1563 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1564 		goto bail;
1565 	}
1566 
1567 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1568 			"%lx at %lx\n", regs->trap, regs->nip);
1569 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1570 
1571 bail:
1572 	exception_exit(prev_state);
1573 }
1574 
1575 void vsx_unavailable_exception(struct pt_regs *regs)
1576 {
1577 	if (user_mode(regs)) {
1578 		/* A user program has executed an vsx instruction,
1579 		   but this kernel doesn't support vsx. */
1580 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1581 		return;
1582 	}
1583 
1584 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1585 			"%lx at %lx\n", regs->trap, regs->nip);
1586 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1587 }
1588 
1589 #ifdef CONFIG_PPC64
1590 static void tm_unavailable(struct pt_regs *regs)
1591 {
1592 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1593 	if (user_mode(regs)) {
1594 		current->thread.load_tm++;
1595 		regs->msr |= MSR_TM;
1596 		tm_enable();
1597 		tm_restore_sprs(&current->thread);
1598 		return;
1599 	}
1600 #endif
1601 	pr_emerg("Unrecoverable TM Unavailable Exception "
1602 			"%lx at %lx\n", regs->trap, regs->nip);
1603 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1604 }
1605 
1606 void facility_unavailable_exception(struct pt_regs *regs)
1607 {
1608 	static char *facility_strings[] = {
1609 		[FSCR_FP_LG] = "FPU",
1610 		[FSCR_VECVSX_LG] = "VMX/VSX",
1611 		[FSCR_DSCR_LG] = "DSCR",
1612 		[FSCR_PM_LG] = "PMU SPRs",
1613 		[FSCR_BHRB_LG] = "BHRB",
1614 		[FSCR_TM_LG] = "TM",
1615 		[FSCR_EBB_LG] = "EBB",
1616 		[FSCR_TAR_LG] = "TAR",
1617 		[FSCR_MSGP_LG] = "MSGP",
1618 		[FSCR_SCV_LG] = "SCV",
1619 	};
1620 	char *facility = "unknown";
1621 	u64 value;
1622 	u32 instword, rd;
1623 	u8 status;
1624 	bool hv;
1625 
1626 	hv = (TRAP(regs) == 0xf80);
1627 	if (hv)
1628 		value = mfspr(SPRN_HFSCR);
1629 	else
1630 		value = mfspr(SPRN_FSCR);
1631 
1632 	status = value >> 56;
1633 	if ((hv || status >= 2) &&
1634 	    (status < ARRAY_SIZE(facility_strings)) &&
1635 	    facility_strings[status])
1636 		facility = facility_strings[status];
1637 
1638 	/* We should not have taken this interrupt in kernel */
1639 	if (!user_mode(regs)) {
1640 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1641 			 facility, status, regs->nip);
1642 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1643 	}
1644 
1645 	/* We restore the interrupt state now */
1646 	if (!arch_irq_disabled_regs(regs))
1647 		local_irq_enable();
1648 
1649 	if (status == FSCR_DSCR_LG) {
1650 		/*
1651 		 * User is accessing the DSCR register using the problem
1652 		 * state only SPR number (0x03) either through a mfspr or
1653 		 * a mtspr instruction. If it is a write attempt through
1654 		 * a mtspr, then we set the inherit bit. This also allows
1655 		 * the user to write or read the register directly in the
1656 		 * future by setting via the FSCR DSCR bit. But in case it
1657 		 * is a read DSCR attempt through a mfspr instruction, we
1658 		 * just emulate the instruction instead. This code path will
1659 		 * always emulate all the mfspr instructions till the user
1660 		 * has attempted at least one mtspr instruction. This way it
1661 		 * preserves the same behaviour when the user is accessing
1662 		 * the DSCR through privilege level only SPR number (0x11)
1663 		 * which is emulated through illegal instruction exception.
1664 		 * We always leave HFSCR DSCR set.
1665 		 */
1666 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1667 			pr_err("Failed to fetch the user instruction\n");
1668 			return;
1669 		}
1670 
1671 		/* Write into DSCR (mtspr 0x03, RS) */
1672 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1673 				== PPC_INST_MTSPR_DSCR_USER) {
1674 			rd = (instword >> 21) & 0x1f;
1675 			current->thread.dscr = regs->gpr[rd];
1676 			current->thread.dscr_inherit = 1;
1677 			current->thread.fscr |= FSCR_DSCR;
1678 			mtspr(SPRN_FSCR, current->thread.fscr);
1679 		}
1680 
1681 		/* Read from DSCR (mfspr RT, 0x03) */
1682 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1683 				== PPC_INST_MFSPR_DSCR_USER) {
1684 			if (emulate_instruction(regs)) {
1685 				pr_err("DSCR based mfspr emulation failed\n");
1686 				return;
1687 			}
1688 			regs->nip += 4;
1689 			emulate_single_step(regs);
1690 		}
1691 		return;
1692 	}
1693 
1694 	if (status == FSCR_TM_LG) {
1695 		/*
1696 		 * If we're here then the hardware is TM aware because it
1697 		 * generated an exception with FSRM_TM set.
1698 		 *
1699 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1700 		 * told us not to do TM, or the kernel is not built with TM
1701 		 * support.
1702 		 *
1703 		 * If both of those things are true, then userspace can spam the
1704 		 * console by triggering the printk() below just by continually
1705 		 * doing tbegin (or any TM instruction). So in that case just
1706 		 * send the process a SIGILL immediately.
1707 		 */
1708 		if (!cpu_has_feature(CPU_FTR_TM))
1709 			goto out;
1710 
1711 		tm_unavailable(regs);
1712 		return;
1713 	}
1714 
1715 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1716 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1717 
1718 out:
1719 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1720 }
1721 #endif
1722 
1723 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1724 
1725 void fp_unavailable_tm(struct pt_regs *regs)
1726 {
1727 	/* Note:  This does not handle any kind of FP laziness. */
1728 
1729 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1730 		 regs->nip, regs->msr);
1731 
1732         /* We can only have got here if the task started using FP after
1733          * beginning the transaction.  So, the transactional regs are just a
1734          * copy of the checkpointed ones.  But, we still need to recheckpoint
1735          * as we're enabling FP for the process; it will return, abort the
1736          * transaction, and probably retry but now with FP enabled.  So the
1737          * checkpointed FP registers need to be loaded.
1738 	 */
1739 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1740 
1741 	/*
1742 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1743 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1744 	 *
1745 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1746 	 * recheckpoint.
1747 	 */
1748 
1749 	/* Enable FP for the task: */
1750 	current->thread.load_fp = 1;
1751 
1752 	/*
1753 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1754 	 */
1755 	tm_recheckpoint(&current->thread);
1756 }
1757 
1758 void altivec_unavailable_tm(struct pt_regs *regs)
1759 {
1760 	/* See the comments in fp_unavailable_tm().  This function operates
1761 	 * the same way.
1762 	 */
1763 
1764 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1765 		 "MSR=%lx\n",
1766 		 regs->nip, regs->msr);
1767 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1768 	current->thread.load_vec = 1;
1769 	tm_recheckpoint(&current->thread);
1770 	current->thread.used_vr = 1;
1771 }
1772 
1773 void vsx_unavailable_tm(struct pt_regs *regs)
1774 {
1775 	/* See the comments in fp_unavailable_tm().  This works similarly,
1776 	 * though we're loading both FP and VEC registers in here.
1777 	 *
1778 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1779 	 * regs.  Either way, set MSR_VSX.
1780 	 */
1781 
1782 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1783 		 "MSR=%lx\n",
1784 		 regs->nip, regs->msr);
1785 
1786 	current->thread.used_vsr = 1;
1787 
1788 	/* This reclaims FP and/or VR regs if they're already enabled */
1789 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1790 
1791 	current->thread.load_vec = 1;
1792 	current->thread.load_fp = 1;
1793 
1794 	tm_recheckpoint(&current->thread);
1795 }
1796 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1797 
1798 void performance_monitor_exception(struct pt_regs *regs)
1799 {
1800 	__this_cpu_inc(irq_stat.pmu_irqs);
1801 
1802 	perf_irq(regs);
1803 }
1804 
1805 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1806 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1807 {
1808 	int changed = 0;
1809 	/*
1810 	 * Determine the cause of the debug event, clear the
1811 	 * event flags and send a trap to the handler. Torez
1812 	 */
1813 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1814 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1815 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1816 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1817 #endif
1818 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1819 			     5);
1820 		changed |= 0x01;
1821 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1822 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1823 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1824 			     6);
1825 		changed |= 0x01;
1826 	}  else if (debug_status & DBSR_IAC1) {
1827 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1828 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1829 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1830 			     1);
1831 		changed |= 0x01;
1832 	}  else if (debug_status & DBSR_IAC2) {
1833 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1834 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1835 			     2);
1836 		changed |= 0x01;
1837 	}  else if (debug_status & DBSR_IAC3) {
1838 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1839 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1840 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1841 			     3);
1842 		changed |= 0x01;
1843 	}  else if (debug_status & DBSR_IAC4) {
1844 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1845 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1846 			     4);
1847 		changed |= 0x01;
1848 	}
1849 	/*
1850 	 * At the point this routine was called, the MSR(DE) was turned off.
1851 	 * Check all other debug flags and see if that bit needs to be turned
1852 	 * back on or not.
1853 	 */
1854 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1855 			       current->thread.debug.dbcr1))
1856 		regs->msr |= MSR_DE;
1857 	else
1858 		/* Make sure the IDM flag is off */
1859 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1860 
1861 	if (changed & 0x01)
1862 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1863 }
1864 
1865 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1866 {
1867 	current->thread.debug.dbsr = debug_status;
1868 
1869 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1870 	 * on server, it stops on the target of the branch. In order to simulate
1871 	 * the server behaviour, we thus restart right away with a single step
1872 	 * instead of stopping here when hitting a BT
1873 	 */
1874 	if (debug_status & DBSR_BT) {
1875 		regs->msr &= ~MSR_DE;
1876 
1877 		/* Disable BT */
1878 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1879 		/* Clear the BT event */
1880 		mtspr(SPRN_DBSR, DBSR_BT);
1881 
1882 		/* Do the single step trick only when coming from userspace */
1883 		if (user_mode(regs)) {
1884 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1885 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1886 			regs->msr |= MSR_DE;
1887 			return;
1888 		}
1889 
1890 		if (kprobe_post_handler(regs))
1891 			return;
1892 
1893 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1894 			       5, SIGTRAP) == NOTIFY_STOP) {
1895 			return;
1896 		}
1897 		if (debugger_sstep(regs))
1898 			return;
1899 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1900 		regs->msr &= ~MSR_DE;
1901 
1902 		/* Disable instruction completion */
1903 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
1904 		/* Clear the instruction completion event */
1905 		mtspr(SPRN_DBSR, DBSR_IC);
1906 
1907 		if (kprobe_post_handler(regs))
1908 			return;
1909 
1910 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1911 			       5, SIGTRAP) == NOTIFY_STOP) {
1912 			return;
1913 		}
1914 
1915 		if (debugger_sstep(regs))
1916 			return;
1917 
1918 		if (user_mode(regs)) {
1919 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
1920 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1921 					       current->thread.debug.dbcr1))
1922 				regs->msr |= MSR_DE;
1923 			else
1924 				/* Make sure the IDM bit is off */
1925 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1926 		}
1927 
1928 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1929 	} else
1930 		handle_debug(regs, debug_status);
1931 }
1932 NOKPROBE_SYMBOL(DebugException);
1933 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
1934 
1935 #if !defined(CONFIG_TAU_INT)
1936 void TAUException(struct pt_regs *regs)
1937 {
1938 	printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
1939 	       regs->nip, regs->msr, regs->trap, print_tainted());
1940 }
1941 #endif /* CONFIG_INT_TAU */
1942 
1943 #ifdef CONFIG_ALTIVEC
1944 void altivec_assist_exception(struct pt_regs *regs)
1945 {
1946 	int err;
1947 
1948 	if (!user_mode(regs)) {
1949 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
1950 		       " at %lx\n", regs->nip);
1951 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
1952 	}
1953 
1954 	flush_altivec_to_thread(current);
1955 
1956 	PPC_WARN_EMULATED(altivec, regs);
1957 	err = emulate_altivec(regs);
1958 	if (err == 0) {
1959 		regs->nip += 4;		/* skip emulated instruction */
1960 		emulate_single_step(regs);
1961 		return;
1962 	}
1963 
1964 	if (err == -EFAULT) {
1965 		/* got an error reading the instruction */
1966 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
1967 	} else {
1968 		/* didn't recognize the instruction */
1969 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
1970 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
1971 				   "in %s at %lx\n", current->comm, regs->nip);
1972 		current->thread.vr_state.vscr.u[3] |= 0x10000;
1973 	}
1974 }
1975 #endif /* CONFIG_ALTIVEC */
1976 
1977 #ifdef CONFIG_FSL_BOOKE
1978 void CacheLockingException(struct pt_regs *regs, unsigned long address,
1979 			   unsigned long error_code)
1980 {
1981 	/* We treat cache locking instructions from the user
1982 	 * as priv ops, in the future we could try to do
1983 	 * something smarter
1984 	 */
1985 	if (error_code & (ESR_DLK|ESR_ILK))
1986 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1987 	return;
1988 }
1989 #endif /* CONFIG_FSL_BOOKE */
1990 
1991 #ifdef CONFIG_SPE
1992 void SPEFloatingPointException(struct pt_regs *regs)
1993 {
1994 	extern int do_spe_mathemu(struct pt_regs *regs);
1995 	unsigned long spefscr;
1996 	int fpexc_mode;
1997 	int code = FPE_FLTUNK;
1998 	int err;
1999 
2000 	flush_spe_to_thread(current);
2001 
2002 	spefscr = current->thread.spefscr;
2003 	fpexc_mode = current->thread.fpexc_mode;
2004 
2005 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2006 		code = FPE_FLTOVF;
2007 	}
2008 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2009 		code = FPE_FLTUND;
2010 	}
2011 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2012 		code = FPE_FLTDIV;
2013 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2014 		code = FPE_FLTINV;
2015 	}
2016 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2017 		code = FPE_FLTRES;
2018 
2019 	err = do_spe_mathemu(regs);
2020 	if (err == 0) {
2021 		regs->nip += 4;		/* skip emulated instruction */
2022 		emulate_single_step(regs);
2023 		return;
2024 	}
2025 
2026 	if (err == -EFAULT) {
2027 		/* got an error reading the instruction */
2028 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2029 	} else if (err == -EINVAL) {
2030 		/* didn't recognize the instruction */
2031 		printk(KERN_ERR "unrecognized spe instruction "
2032 		       "in %s at %lx\n", current->comm, regs->nip);
2033 	} else {
2034 		_exception(SIGFPE, regs, code, regs->nip);
2035 	}
2036 
2037 	return;
2038 }
2039 
2040 void SPEFloatingPointRoundException(struct pt_regs *regs)
2041 {
2042 	extern int speround_handler(struct pt_regs *regs);
2043 	int err;
2044 
2045 	preempt_disable();
2046 	if (regs->msr & MSR_SPE)
2047 		giveup_spe(current);
2048 	preempt_enable();
2049 
2050 	regs->nip -= 4;
2051 	err = speround_handler(regs);
2052 	if (err == 0) {
2053 		regs->nip += 4;		/* skip emulated instruction */
2054 		emulate_single_step(regs);
2055 		return;
2056 	}
2057 
2058 	if (err == -EFAULT) {
2059 		/* got an error reading the instruction */
2060 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2061 	} else if (err == -EINVAL) {
2062 		/* didn't recognize the instruction */
2063 		printk(KERN_ERR "unrecognized spe instruction "
2064 		       "in %s at %lx\n", current->comm, regs->nip);
2065 	} else {
2066 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2067 		return;
2068 	}
2069 }
2070 #endif
2071 
2072 /*
2073  * We enter here if we get an unrecoverable exception, that is, one
2074  * that happened at a point where the RI (recoverable interrupt) bit
2075  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2076  * we therefore lost state by taking this exception.
2077  */
2078 void unrecoverable_exception(struct pt_regs *regs)
2079 {
2080 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2081 		 regs->trap, regs->nip, regs->msr);
2082 	die("Unrecoverable exception", regs, SIGABRT);
2083 }
2084 NOKPROBE_SYMBOL(unrecoverable_exception);
2085 
2086 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2087 /*
2088  * Default handler for a Watchdog exception,
2089  * spins until a reboot occurs
2090  */
2091 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2092 {
2093 	/* Generic WatchdogHandler, implement your own */
2094 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2095 	return;
2096 }
2097 
2098 void WatchdogException(struct pt_regs *regs)
2099 {
2100 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2101 	WatchdogHandler(regs);
2102 }
2103 #endif
2104 
2105 /*
2106  * We enter here if we discover during exception entry that we are
2107  * running in supervisor mode with a userspace value in the stack pointer.
2108  */
2109 void kernel_bad_stack(struct pt_regs *regs)
2110 {
2111 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2112 	       regs->gpr[1], regs->nip);
2113 	die("Bad kernel stack pointer", regs, SIGABRT);
2114 }
2115 NOKPROBE_SYMBOL(kernel_bad_stack);
2116 
2117 void __init trap_init(void)
2118 {
2119 }
2120 
2121 
2122 #ifdef CONFIG_PPC_EMULATED_STATS
2123 
2124 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2125 
2126 struct ppc_emulated ppc_emulated = {
2127 #ifdef CONFIG_ALTIVEC
2128 	WARN_EMULATED_SETUP(altivec),
2129 #endif
2130 	WARN_EMULATED_SETUP(dcba),
2131 	WARN_EMULATED_SETUP(dcbz),
2132 	WARN_EMULATED_SETUP(fp_pair),
2133 	WARN_EMULATED_SETUP(isel),
2134 	WARN_EMULATED_SETUP(mcrxr),
2135 	WARN_EMULATED_SETUP(mfpvr),
2136 	WARN_EMULATED_SETUP(multiple),
2137 	WARN_EMULATED_SETUP(popcntb),
2138 	WARN_EMULATED_SETUP(spe),
2139 	WARN_EMULATED_SETUP(string),
2140 	WARN_EMULATED_SETUP(sync),
2141 	WARN_EMULATED_SETUP(unaligned),
2142 #ifdef CONFIG_MATH_EMULATION
2143 	WARN_EMULATED_SETUP(math),
2144 #endif
2145 #ifdef CONFIG_VSX
2146 	WARN_EMULATED_SETUP(vsx),
2147 #endif
2148 #ifdef CONFIG_PPC64
2149 	WARN_EMULATED_SETUP(mfdscr),
2150 	WARN_EMULATED_SETUP(mtdscr),
2151 	WARN_EMULATED_SETUP(lq_stq),
2152 	WARN_EMULATED_SETUP(lxvw4x),
2153 	WARN_EMULATED_SETUP(lxvh8x),
2154 	WARN_EMULATED_SETUP(lxvd2x),
2155 	WARN_EMULATED_SETUP(lxvb16x),
2156 #endif
2157 };
2158 
2159 u32 ppc_warn_emulated;
2160 
2161 void ppc_warn_emulated_print(const char *type)
2162 {
2163 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2164 			    type);
2165 }
2166 
2167 static int __init ppc_warn_emulated_init(void)
2168 {
2169 	struct dentry *dir, *d;
2170 	unsigned int i;
2171 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2172 
2173 	if (!powerpc_debugfs_root)
2174 		return -ENODEV;
2175 
2176 	dir = debugfs_create_dir("emulated_instructions",
2177 				 powerpc_debugfs_root);
2178 	if (!dir)
2179 		return -ENOMEM;
2180 
2181 	d = debugfs_create_u32("do_warn", 0644, dir,
2182 			       &ppc_warn_emulated);
2183 	if (!d)
2184 		goto fail;
2185 
2186 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) {
2187 		d = debugfs_create_u32(entries[i].name, 0644, dir,
2188 				       (u32 *)&entries[i].val.counter);
2189 		if (!d)
2190 			goto fail;
2191 	}
2192 
2193 	return 0;
2194 
2195 fail:
2196 	debugfs_remove_recursive(dir);
2197 	return -ENOMEM;
2198 }
2199 
2200 device_initcall(ppc_warn_emulated_init);
2201 
2202 #endif /* CONFIG_PPC_EMULATED_STATS */
2203