xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision ba61bb17)
1 /*
2  *  Copyright (C) 1995-1996  Gary Thomas (gdt@linuxppc.org)
3  *  Copyright 2007-2010 Freescale Semiconductor, Inc.
4  *
5  *  This program is free software; you can redistribute it and/or
6  *  modify it under the terms of the GNU General Public License
7  *  as published by the Free Software Foundation; either version
8  *  2 of the License, or (at your option) any later version.
9  *
10  *  Modified by Cort Dougan (cort@cs.nmt.edu)
11  *  and Paul Mackerras (paulus@samba.org)
12  */
13 
14 /*
15  * This file handles the architecture-dependent parts of hardware exceptions
16  */
17 
18 #include <linux/errno.h>
19 #include <linux/sched.h>
20 #include <linux/sched/debug.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/pkeys.h>
24 #include <linux/stddef.h>
25 #include <linux/unistd.h>
26 #include <linux/ptrace.h>
27 #include <linux/user.h>
28 #include <linux/interrupt.h>
29 #include <linux/init.h>
30 #include <linux/extable.h>
31 #include <linux/module.h>	/* print_modules */
32 #include <linux/prctl.h>
33 #include <linux/delay.h>
34 #include <linux/kprobes.h>
35 #include <linux/kexec.h>
36 #include <linux/backlight.h>
37 #include <linux/bug.h>
38 #include <linux/kdebug.h>
39 #include <linux/ratelimit.h>
40 #include <linux/context_tracking.h>
41 #include <linux/smp.h>
42 #include <linux/console.h>
43 #include <linux/kmsg_dump.h>
44 
45 #include <asm/emulated_ops.h>
46 #include <asm/pgtable.h>
47 #include <linux/uaccess.h>
48 #include <asm/debugfs.h>
49 #include <asm/io.h>
50 #include <asm/machdep.h>
51 #include <asm/rtas.h>
52 #include <asm/pmc.h>
53 #include <asm/reg.h>
54 #ifdef CONFIG_PMAC_BACKLIGHT
55 #include <asm/backlight.h>
56 #endif
57 #ifdef CONFIG_PPC64
58 #include <asm/firmware.h>
59 #include <asm/processor.h>
60 #include <asm/tm.h>
61 #endif
62 #include <asm/kexec.h>
63 #include <asm/ppc-opcode.h>
64 #include <asm/rio.h>
65 #include <asm/fadump.h>
66 #include <asm/switch_to.h>
67 #include <asm/tm.h>
68 #include <asm/debug.h>
69 #include <asm/asm-prototypes.h>
70 #include <asm/hmi.h>
71 #include <sysdev/fsl_pci.h>
72 #include <asm/kprobes.h>
73 
74 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
75 int (*__debugger)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
79 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
80 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
81 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
82 
83 EXPORT_SYMBOL(__debugger);
84 EXPORT_SYMBOL(__debugger_ipi);
85 EXPORT_SYMBOL(__debugger_bpt);
86 EXPORT_SYMBOL(__debugger_sstep);
87 EXPORT_SYMBOL(__debugger_iabr_match);
88 EXPORT_SYMBOL(__debugger_break_match);
89 EXPORT_SYMBOL(__debugger_fault_handler);
90 #endif
91 
92 /* Transactional Memory trap debug */
93 #ifdef TM_DEBUG_SW
94 #define TM_DEBUG(x...) printk(KERN_INFO x)
95 #else
96 #define TM_DEBUG(x...) do { } while(0)
97 #endif
98 
99 /*
100  * Trap & Exception support
101  */
102 
103 #ifdef CONFIG_PMAC_BACKLIGHT
104 static void pmac_backlight_unblank(void)
105 {
106 	mutex_lock(&pmac_backlight_mutex);
107 	if (pmac_backlight) {
108 		struct backlight_properties *props;
109 
110 		props = &pmac_backlight->props;
111 		props->brightness = props->max_brightness;
112 		props->power = FB_BLANK_UNBLANK;
113 		backlight_update_status(pmac_backlight);
114 	}
115 	mutex_unlock(&pmac_backlight_mutex);
116 }
117 #else
118 static inline void pmac_backlight_unblank(void) { }
119 #endif
120 
121 /*
122  * If oops/die is expected to crash the machine, return true here.
123  *
124  * This should not be expected to be 100% accurate, there may be
125  * notifiers registered or other unexpected conditions that may bring
126  * down the kernel. Or if the current process in the kernel is holding
127  * locks or has other critical state, the kernel may become effectively
128  * unusable anyway.
129  */
130 bool die_will_crash(void)
131 {
132 	if (should_fadump_crash())
133 		return true;
134 	if (kexec_should_crash(current))
135 		return true;
136 	if (in_interrupt() || panic_on_oops ||
137 			!current->pid || is_global_init(current))
138 		return true;
139 
140 	return false;
141 }
142 
143 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
144 static int die_owner = -1;
145 static unsigned int die_nest_count;
146 static int die_counter;
147 
148 extern void panic_flush_kmsg_start(void)
149 {
150 	/*
151 	 * These are mostly taken from kernel/panic.c, but tries to do
152 	 * relatively minimal work. Don't use delay functions (TB may
153 	 * be broken), don't crash dump (need to set a firmware log),
154 	 * don't run notifiers. We do want to get some information to
155 	 * Linux console.
156 	 */
157 	console_verbose();
158 	bust_spinlocks(1);
159 }
160 
161 extern void panic_flush_kmsg_end(void)
162 {
163 	printk_safe_flush_on_panic();
164 	kmsg_dump(KMSG_DUMP_PANIC);
165 	bust_spinlocks(0);
166 	debug_locks_off();
167 	console_flush_on_panic();
168 }
169 
170 static unsigned long oops_begin(struct pt_regs *regs)
171 {
172 	int cpu;
173 	unsigned long flags;
174 
175 	oops_enter();
176 
177 	/* racy, but better than risking deadlock. */
178 	raw_local_irq_save(flags);
179 	cpu = smp_processor_id();
180 	if (!arch_spin_trylock(&die_lock)) {
181 		if (cpu == die_owner)
182 			/* nested oops. should stop eventually */;
183 		else
184 			arch_spin_lock(&die_lock);
185 	}
186 	die_nest_count++;
187 	die_owner = cpu;
188 	console_verbose();
189 	bust_spinlocks(1);
190 	if (machine_is(powermac))
191 		pmac_backlight_unblank();
192 	return flags;
193 }
194 NOKPROBE_SYMBOL(oops_begin);
195 
196 static void oops_end(unsigned long flags, struct pt_regs *regs,
197 			       int signr)
198 {
199 	bust_spinlocks(0);
200 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
201 	die_nest_count--;
202 	oops_exit();
203 	printk("\n");
204 	if (!die_nest_count) {
205 		/* Nest count reaches zero, release the lock. */
206 		die_owner = -1;
207 		arch_spin_unlock(&die_lock);
208 	}
209 	raw_local_irq_restore(flags);
210 
211 	/*
212 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
213 	 */
214 	if (TRAP(regs) == 0x100)
215 		return;
216 
217 	crash_fadump(regs, "die oops");
218 
219 	if (kexec_should_crash(current))
220 		crash_kexec(regs);
221 
222 	if (!signr)
223 		return;
224 
225 	/*
226 	 * While our oops output is serialised by a spinlock, output
227 	 * from panic() called below can race and corrupt it. If we
228 	 * know we are going to panic, delay for 1 second so we have a
229 	 * chance to get clean backtraces from all CPUs that are oopsing.
230 	 */
231 	if (in_interrupt() || panic_on_oops || !current->pid ||
232 	    is_global_init(current)) {
233 		mdelay(MSEC_PER_SEC);
234 	}
235 
236 	if (in_interrupt())
237 		panic("Fatal exception in interrupt");
238 	if (panic_on_oops)
239 		panic("Fatal exception");
240 	do_exit(signr);
241 }
242 NOKPROBE_SYMBOL(oops_end);
243 
244 static int __die(const char *str, struct pt_regs *regs, long err)
245 {
246 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
247 
248 	if (IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN))
249 		printk("LE ");
250 	else
251 		printk("BE ");
252 
253 	if (IS_ENABLED(CONFIG_PREEMPT))
254 		pr_cont("PREEMPT ");
255 
256 	if (IS_ENABLED(CONFIG_SMP))
257 		pr_cont("SMP NR_CPUS=%d ", NR_CPUS);
258 
259 	if (debug_pagealloc_enabled())
260 		pr_cont("DEBUG_PAGEALLOC ");
261 
262 	if (IS_ENABLED(CONFIG_NUMA))
263 		pr_cont("NUMA ");
264 
265 	pr_cont("%s\n", ppc_md.name ? ppc_md.name : "");
266 
267 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
268 		return 1;
269 
270 	print_modules();
271 	show_regs(regs);
272 
273 	return 0;
274 }
275 NOKPROBE_SYMBOL(__die);
276 
277 void die(const char *str, struct pt_regs *regs, long err)
278 {
279 	unsigned long flags;
280 
281 	/*
282 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
283 	 */
284 	if (TRAP(regs) != 0x100) {
285 		if (debugger(regs))
286 			return;
287 	}
288 
289 	flags = oops_begin(regs);
290 	if (__die(str, regs, err))
291 		err = 0;
292 	oops_end(flags, regs, err);
293 }
294 NOKPROBE_SYMBOL(die);
295 
296 void user_single_step_siginfo(struct task_struct *tsk,
297 				struct pt_regs *regs, siginfo_t *info)
298 {
299 	info->si_signo = SIGTRAP;
300 	info->si_code = TRAP_TRACE;
301 	info->si_addr = (void __user *)regs->nip;
302 }
303 
304 
305 void _exception_pkey(int signr, struct pt_regs *regs, int code,
306 		unsigned long addr, int key)
307 {
308 	siginfo_t info;
309 	const char fmt32[] = KERN_INFO "%s[%d]: unhandled signal %d " \
310 			"at %08lx nip %08lx lr %08lx code %x\n";
311 	const char fmt64[] = KERN_INFO "%s[%d]: unhandled signal %d " \
312 			"at %016lx nip %016lx lr %016lx code %x\n";
313 
314 	if (!user_mode(regs)) {
315 		die("Exception in kernel mode", regs, signr);
316 		return;
317 	}
318 
319 	if (show_unhandled_signals && unhandled_signal(current, signr)) {
320 		printk_ratelimited(regs->msr & MSR_64BIT ? fmt64 : fmt32,
321 				   current->comm, current->pid, signr,
322 				   addr, regs->nip, regs->link, code);
323 	}
324 
325 	if (arch_irqs_disabled() && !arch_irq_disabled_regs(regs))
326 		local_irq_enable();
327 
328 	current->thread.trap_nr = code;
329 
330 	/*
331 	 * Save all the pkey registers AMR/IAMR/UAMOR. Eg: Core dumps need
332 	 * to capture the content, if the task gets killed.
333 	 */
334 	thread_pkey_regs_save(&current->thread);
335 
336 	clear_siginfo(&info);
337 	info.si_signo = signr;
338 	info.si_code = code;
339 	info.si_addr = (void __user *) addr;
340 	info.si_pkey = key;
341 
342 	force_sig_info(signr, &info, current);
343 }
344 
345 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
346 {
347 	_exception_pkey(signr, regs, code, addr, 0);
348 }
349 
350 void system_reset_exception(struct pt_regs *regs)
351 {
352 	/*
353 	 * Avoid crashes in case of nested NMI exceptions. Recoverability
354 	 * is determined by RI and in_nmi
355 	 */
356 	bool nested = in_nmi();
357 	if (!nested)
358 		nmi_enter();
359 
360 	__this_cpu_inc(irq_stat.sreset_irqs);
361 
362 	/* See if any machine dependent calls */
363 	if (ppc_md.system_reset_exception) {
364 		if (ppc_md.system_reset_exception(regs))
365 			goto out;
366 	}
367 
368 	if (debugger(regs))
369 		goto out;
370 
371 	/*
372 	 * A system reset is a request to dump, so we always send
373 	 * it through the crashdump code (if fadump or kdump are
374 	 * registered).
375 	 */
376 	crash_fadump(regs, "System Reset");
377 
378 	crash_kexec(regs);
379 
380 	/*
381 	 * We aren't the primary crash CPU. We need to send it
382 	 * to a holding pattern to avoid it ending up in the panic
383 	 * code.
384 	 */
385 	crash_kexec_secondary(regs);
386 
387 	/*
388 	 * No debugger or crash dump registered, print logs then
389 	 * panic.
390 	 */
391 	die("System Reset", regs, SIGABRT);
392 
393 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
394 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
395 	nmi_panic(regs, "System Reset");
396 
397 out:
398 #ifdef CONFIG_PPC_BOOK3S_64
399 	BUG_ON(get_paca()->in_nmi == 0);
400 	if (get_paca()->in_nmi > 1)
401 		nmi_panic(regs, "Unrecoverable nested System Reset");
402 #endif
403 	/* Must die if the interrupt is not recoverable */
404 	if (!(regs->msr & MSR_RI))
405 		nmi_panic(regs, "Unrecoverable System Reset");
406 
407 	if (!nested)
408 		nmi_exit();
409 
410 	/* What should we do here? We could issue a shutdown or hard reset. */
411 }
412 
413 /*
414  * I/O accesses can cause machine checks on powermacs.
415  * Check if the NIP corresponds to the address of a sync
416  * instruction for which there is an entry in the exception
417  * table.
418  * Note that the 601 only takes a machine check on TEA
419  * (transfer error ack) signal assertion, and does not
420  * set any of the top 16 bits of SRR1.
421  *  -- paulus.
422  */
423 static inline int check_io_access(struct pt_regs *regs)
424 {
425 #ifdef CONFIG_PPC32
426 	unsigned long msr = regs->msr;
427 	const struct exception_table_entry *entry;
428 	unsigned int *nip = (unsigned int *)regs->nip;
429 
430 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
431 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
432 		/*
433 		 * Check that it's a sync instruction, or somewhere
434 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
435 		 * As the address is in the exception table
436 		 * we should be able to read the instr there.
437 		 * For the debug message, we look at the preceding
438 		 * load or store.
439 		 */
440 		if (*nip == PPC_INST_NOP)
441 			nip -= 2;
442 		else if (*nip == PPC_INST_ISYNC)
443 			--nip;
444 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
445 			unsigned int rb;
446 
447 			--nip;
448 			rb = (*nip >> 11) & 0x1f;
449 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
450 			       (*nip & 0x100)? "OUT to": "IN from",
451 			       regs->gpr[rb] - _IO_BASE, nip);
452 			regs->msr |= MSR_RI;
453 			regs->nip = extable_fixup(entry);
454 			return 1;
455 		}
456 	}
457 #endif /* CONFIG_PPC32 */
458 	return 0;
459 }
460 
461 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
462 /* On 4xx, the reason for the machine check or program exception
463    is in the ESR. */
464 #define get_reason(regs)	((regs)->dsisr)
465 #define REASON_FP		ESR_FP
466 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
467 #define REASON_PRIVILEGED	ESR_PPR
468 #define REASON_TRAP		ESR_PTR
469 
470 /* single-step stuff */
471 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
472 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
473 #define clear_br_trace(regs)	do {} while(0)
474 #else
475 /* On non-4xx, the reason for the machine check or program
476    exception is in the MSR. */
477 #define get_reason(regs)	((regs)->msr)
478 #define REASON_TM		SRR1_PROGTM
479 #define REASON_FP		SRR1_PROGFPE
480 #define REASON_ILLEGAL		SRR1_PROGILL
481 #define REASON_PRIVILEGED	SRR1_PROGPRIV
482 #define REASON_TRAP		SRR1_PROGTRAP
483 
484 #define single_stepping(regs)	((regs)->msr & MSR_SE)
485 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
486 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
487 #endif
488 
489 #if defined(CONFIG_E500)
490 int machine_check_e500mc(struct pt_regs *regs)
491 {
492 	unsigned long mcsr = mfspr(SPRN_MCSR);
493 	unsigned long pvr = mfspr(SPRN_PVR);
494 	unsigned long reason = mcsr;
495 	int recoverable = 1;
496 
497 	if (reason & MCSR_LD) {
498 		recoverable = fsl_rio_mcheck_exception(regs);
499 		if (recoverable == 1)
500 			goto silent_out;
501 	}
502 
503 	printk("Machine check in kernel mode.\n");
504 	printk("Caused by (from MCSR=%lx): ", reason);
505 
506 	if (reason & MCSR_MCP)
507 		printk("Machine Check Signal\n");
508 
509 	if (reason & MCSR_ICPERR) {
510 		printk("Instruction Cache Parity Error\n");
511 
512 		/*
513 		 * This is recoverable by invalidating the i-cache.
514 		 */
515 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
516 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
517 			;
518 
519 		/*
520 		 * This will generally be accompanied by an instruction
521 		 * fetch error report -- only treat MCSR_IF as fatal
522 		 * if it wasn't due to an L1 parity error.
523 		 */
524 		reason &= ~MCSR_IF;
525 	}
526 
527 	if (reason & MCSR_DCPERR_MC) {
528 		printk("Data Cache Parity Error\n");
529 
530 		/*
531 		 * In write shadow mode we auto-recover from the error, but it
532 		 * may still get logged and cause a machine check.  We should
533 		 * only treat the non-write shadow case as non-recoverable.
534 		 */
535 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
536 		 * is not implemented but L1 data cache always runs in write
537 		 * shadow mode. Hence on data cache parity errors HW will
538 		 * automatically invalidate the L1 Data Cache.
539 		 */
540 		if (PVR_VER(pvr) != PVR_VER_E6500) {
541 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
542 				recoverable = 0;
543 		}
544 	}
545 
546 	if (reason & MCSR_L2MMU_MHIT) {
547 		printk("Hit on multiple TLB entries\n");
548 		recoverable = 0;
549 	}
550 
551 	if (reason & MCSR_NMI)
552 		printk("Non-maskable interrupt\n");
553 
554 	if (reason & MCSR_IF) {
555 		printk("Instruction Fetch Error Report\n");
556 		recoverable = 0;
557 	}
558 
559 	if (reason & MCSR_LD) {
560 		printk("Load Error Report\n");
561 		recoverable = 0;
562 	}
563 
564 	if (reason & MCSR_ST) {
565 		printk("Store Error Report\n");
566 		recoverable = 0;
567 	}
568 
569 	if (reason & MCSR_LDG) {
570 		printk("Guarded Load Error Report\n");
571 		recoverable = 0;
572 	}
573 
574 	if (reason & MCSR_TLBSYNC)
575 		printk("Simultaneous tlbsync operations\n");
576 
577 	if (reason & MCSR_BSL2_ERR) {
578 		printk("Level 2 Cache Error\n");
579 		recoverable = 0;
580 	}
581 
582 	if (reason & MCSR_MAV) {
583 		u64 addr;
584 
585 		addr = mfspr(SPRN_MCAR);
586 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
587 
588 		printk("Machine Check %s Address: %#llx\n",
589 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
590 	}
591 
592 silent_out:
593 	mtspr(SPRN_MCSR, mcsr);
594 	return mfspr(SPRN_MCSR) == 0 && recoverable;
595 }
596 
597 int machine_check_e500(struct pt_regs *regs)
598 {
599 	unsigned long reason = mfspr(SPRN_MCSR);
600 
601 	if (reason & MCSR_BUS_RBERR) {
602 		if (fsl_rio_mcheck_exception(regs))
603 			return 1;
604 		if (fsl_pci_mcheck_exception(regs))
605 			return 1;
606 	}
607 
608 	printk("Machine check in kernel mode.\n");
609 	printk("Caused by (from MCSR=%lx): ", reason);
610 
611 	if (reason & MCSR_MCP)
612 		printk("Machine Check Signal\n");
613 	if (reason & MCSR_ICPERR)
614 		printk("Instruction Cache Parity Error\n");
615 	if (reason & MCSR_DCP_PERR)
616 		printk("Data Cache Push Parity Error\n");
617 	if (reason & MCSR_DCPERR)
618 		printk("Data Cache Parity Error\n");
619 	if (reason & MCSR_BUS_IAERR)
620 		printk("Bus - Instruction Address Error\n");
621 	if (reason & MCSR_BUS_RAERR)
622 		printk("Bus - Read Address Error\n");
623 	if (reason & MCSR_BUS_WAERR)
624 		printk("Bus - Write Address Error\n");
625 	if (reason & MCSR_BUS_IBERR)
626 		printk("Bus - Instruction Data Error\n");
627 	if (reason & MCSR_BUS_RBERR)
628 		printk("Bus - Read Data Bus Error\n");
629 	if (reason & MCSR_BUS_WBERR)
630 		printk("Bus - Write Data Bus Error\n");
631 	if (reason & MCSR_BUS_IPERR)
632 		printk("Bus - Instruction Parity Error\n");
633 	if (reason & MCSR_BUS_RPERR)
634 		printk("Bus - Read Parity Error\n");
635 
636 	return 0;
637 }
638 
639 int machine_check_generic(struct pt_regs *regs)
640 {
641 	return 0;
642 }
643 #elif defined(CONFIG_E200)
644 int machine_check_e200(struct pt_regs *regs)
645 {
646 	unsigned long reason = mfspr(SPRN_MCSR);
647 
648 	printk("Machine check in kernel mode.\n");
649 	printk("Caused by (from MCSR=%lx): ", reason);
650 
651 	if (reason & MCSR_MCP)
652 		printk("Machine Check Signal\n");
653 	if (reason & MCSR_CP_PERR)
654 		printk("Cache Push Parity Error\n");
655 	if (reason & MCSR_CPERR)
656 		printk("Cache Parity Error\n");
657 	if (reason & MCSR_EXCP_ERR)
658 		printk("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
659 	if (reason & MCSR_BUS_IRERR)
660 		printk("Bus - Read Bus Error on instruction fetch\n");
661 	if (reason & MCSR_BUS_DRERR)
662 		printk("Bus - Read Bus Error on data load\n");
663 	if (reason & MCSR_BUS_WRERR)
664 		printk("Bus - Write Bus Error on buffered store or cache line push\n");
665 
666 	return 0;
667 }
668 #elif defined(CONFIG_PPC32)
669 int machine_check_generic(struct pt_regs *regs)
670 {
671 	unsigned long reason = regs->msr;
672 
673 	printk("Machine check in kernel mode.\n");
674 	printk("Caused by (from SRR1=%lx): ", reason);
675 	switch (reason & 0x601F0000) {
676 	case 0x80000:
677 		printk("Machine check signal\n");
678 		break;
679 	case 0:		/* for 601 */
680 	case 0x40000:
681 	case 0x140000:	/* 7450 MSS error and TEA */
682 		printk("Transfer error ack signal\n");
683 		break;
684 	case 0x20000:
685 		printk("Data parity error signal\n");
686 		break;
687 	case 0x10000:
688 		printk("Address parity error signal\n");
689 		break;
690 	case 0x20000000:
691 		printk("L1 Data Cache error\n");
692 		break;
693 	case 0x40000000:
694 		printk("L1 Instruction Cache error\n");
695 		break;
696 	case 0x00100000:
697 		printk("L2 data cache parity error\n");
698 		break;
699 	default:
700 		printk("Unknown values in msr\n");
701 	}
702 	return 0;
703 }
704 #endif /* everything else */
705 
706 void machine_check_exception(struct pt_regs *regs)
707 {
708 	int recover = 0;
709 	bool nested = in_nmi();
710 	if (!nested)
711 		nmi_enter();
712 
713 	/* 64s accounts the mce in machine_check_early when in HVMODE */
714 	if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64) || !cpu_has_feature(CPU_FTR_HVMODE))
715 		__this_cpu_inc(irq_stat.mce_exceptions);
716 
717 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
718 
719 	/* See if any machine dependent calls. In theory, we would want
720 	 * to call the CPU first, and call the ppc_md. one if the CPU
721 	 * one returns a positive number. However there is existing code
722 	 * that assumes the board gets a first chance, so let's keep it
723 	 * that way for now and fix things later. --BenH.
724 	 */
725 	if (ppc_md.machine_check_exception)
726 		recover = ppc_md.machine_check_exception(regs);
727 	else if (cur_cpu_spec->machine_check)
728 		recover = cur_cpu_spec->machine_check(regs);
729 
730 	if (recover > 0)
731 		goto bail;
732 
733 	if (debugger_fault_handler(regs))
734 		goto bail;
735 
736 	if (check_io_access(regs))
737 		goto bail;
738 
739 	die("Machine check", regs, SIGBUS);
740 
741 	/* Must die if the interrupt is not recoverable */
742 	if (!(regs->msr & MSR_RI))
743 		nmi_panic(regs, "Unrecoverable Machine check");
744 
745 bail:
746 	if (!nested)
747 		nmi_exit();
748 }
749 
750 void SMIException(struct pt_regs *regs)
751 {
752 	die("System Management Interrupt", regs, SIGABRT);
753 }
754 
755 #ifdef CONFIG_VSX
756 static void p9_hmi_special_emu(struct pt_regs *regs)
757 {
758 	unsigned int ra, rb, t, i, sel, instr, rc;
759 	const void __user *addr;
760 	u8 vbuf[16], *vdst;
761 	unsigned long ea, msr, msr_mask;
762 	bool swap;
763 
764 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
765 		return;
766 
767 	/*
768 	 * lxvb16x	opcode: 0x7c0006d8
769 	 * lxvd2x	opcode: 0x7c000698
770 	 * lxvh8x	opcode: 0x7c000658
771 	 * lxvw4x	opcode: 0x7c000618
772 	 */
773 	if ((instr & 0xfc00073e) != 0x7c000618) {
774 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
775 			 " instr=%08x\n",
776 			 smp_processor_id(), current->comm, current->pid,
777 			 regs->nip, instr);
778 		return;
779 	}
780 
781 	/* Grab vector registers into the task struct */
782 	msr = regs->msr; /* Grab msr before we flush the bits */
783 	flush_vsx_to_thread(current);
784 	enable_kernel_altivec();
785 
786 	/*
787 	 * Is userspace running with a different endian (this is rare but
788 	 * not impossible)
789 	 */
790 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
791 
792 	/* Decode the instruction */
793 	ra = (instr >> 16) & 0x1f;
794 	rb = (instr >> 11) & 0x1f;
795 	t = (instr >> 21) & 0x1f;
796 	if (instr & 1)
797 		vdst = (u8 *)&current->thread.vr_state.vr[t];
798 	else
799 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
800 
801 	/* Grab the vector address */
802 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
803 	if (is_32bit_task())
804 		ea &= 0xfffffffful;
805 	addr = (__force const void __user *)ea;
806 
807 	/* Check it */
808 	if (!access_ok(VERIFY_READ, addr, 16)) {
809 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
810 			 " instr=%08x addr=%016lx\n",
811 			 smp_processor_id(), current->comm, current->pid,
812 			 regs->nip, instr, (unsigned long)addr);
813 		return;
814 	}
815 
816 	/* Read the vector */
817 	rc = 0;
818 	if ((unsigned long)addr & 0xfUL)
819 		/* unaligned case */
820 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
821 	else
822 		__get_user_atomic_128_aligned(vbuf, addr, rc);
823 	if (rc) {
824 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
825 			 " instr=%08x addr=%016lx\n",
826 			 smp_processor_id(), current->comm, current->pid,
827 			 regs->nip, instr, (unsigned long)addr);
828 		return;
829 	}
830 
831 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
832 		 " instr=%08x addr=%016lx\n",
833 		 smp_processor_id(), current->comm, current->pid, regs->nip,
834 		 instr, (unsigned long) addr);
835 
836 	/* Grab instruction "selector" */
837 	sel = (instr >> 6) & 3;
838 
839 	/*
840 	 * Check to make sure the facility is actually enabled. This
841 	 * could happen if we get a false positive hit.
842 	 *
843 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
844 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
845 	 */
846 	msr_mask = MSR_VSX;
847 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
848 		msr_mask = MSR_VEC;
849 	if (!(msr & msr_mask)) {
850 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
851 			 " instr=%08x msr:%016lx\n",
852 			 smp_processor_id(), current->comm, current->pid,
853 			 regs->nip, instr, msr);
854 		return;
855 	}
856 
857 	/* Do logging here before we modify sel based on endian */
858 	switch (sel) {
859 	case 0:	/* lxvw4x */
860 		PPC_WARN_EMULATED(lxvw4x, regs);
861 		break;
862 	case 1: /* lxvh8x */
863 		PPC_WARN_EMULATED(lxvh8x, regs);
864 		break;
865 	case 2: /* lxvd2x */
866 		PPC_WARN_EMULATED(lxvd2x, regs);
867 		break;
868 	case 3: /* lxvb16x */
869 		PPC_WARN_EMULATED(lxvb16x, regs);
870 		break;
871 	}
872 
873 #ifdef __LITTLE_ENDIAN__
874 	/*
875 	 * An LE kernel stores the vector in the task struct as an LE
876 	 * byte array (effectively swapping both the components and
877 	 * the content of the components). Those instructions expect
878 	 * the components to remain in ascending address order, so we
879 	 * swap them back.
880 	 *
881 	 * If we are running a BE user space, the expectation is that
882 	 * of a simple memcpy, so forcing the emulation to look like
883 	 * a lxvb16x should do the trick.
884 	 */
885 	if (swap)
886 		sel = 3;
887 
888 	switch (sel) {
889 	case 0:	/* lxvw4x */
890 		for (i = 0; i < 4; i++)
891 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
892 		break;
893 	case 1: /* lxvh8x */
894 		for (i = 0; i < 8; i++)
895 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
896 		break;
897 	case 2: /* lxvd2x */
898 		for (i = 0; i < 2; i++)
899 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
900 		break;
901 	case 3: /* lxvb16x */
902 		for (i = 0; i < 16; i++)
903 			vdst[i] = vbuf[15-i];
904 		break;
905 	}
906 #else /* __LITTLE_ENDIAN__ */
907 	/* On a big endian kernel, a BE userspace only needs a memcpy */
908 	if (!swap)
909 		sel = 3;
910 
911 	/* Otherwise, we need to swap the content of the components */
912 	switch (sel) {
913 	case 0:	/* lxvw4x */
914 		for (i = 0; i < 4; i++)
915 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
916 		break;
917 	case 1: /* lxvh8x */
918 		for (i = 0; i < 8; i++)
919 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
920 		break;
921 	case 2: /* lxvd2x */
922 		for (i = 0; i < 2; i++)
923 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
924 		break;
925 	case 3: /* lxvb16x */
926 		memcpy(vdst, vbuf, 16);
927 		break;
928 	}
929 #endif /* !__LITTLE_ENDIAN__ */
930 
931 	/* Go to next instruction */
932 	regs->nip += 4;
933 }
934 #endif /* CONFIG_VSX */
935 
936 void handle_hmi_exception(struct pt_regs *regs)
937 {
938 	struct pt_regs *old_regs;
939 
940 	old_regs = set_irq_regs(regs);
941 	irq_enter();
942 
943 #ifdef CONFIG_VSX
944 	/* Real mode flagged P9 special emu is needed */
945 	if (local_paca->hmi_p9_special_emu) {
946 		local_paca->hmi_p9_special_emu = 0;
947 
948 		/*
949 		 * We don't want to take page faults while doing the
950 		 * emulation, we just replay the instruction if necessary.
951 		 */
952 		pagefault_disable();
953 		p9_hmi_special_emu(regs);
954 		pagefault_enable();
955 	}
956 #endif /* CONFIG_VSX */
957 
958 	if (ppc_md.handle_hmi_exception)
959 		ppc_md.handle_hmi_exception(regs);
960 
961 	irq_exit();
962 	set_irq_regs(old_regs);
963 }
964 
965 void unknown_exception(struct pt_regs *regs)
966 {
967 	enum ctx_state prev_state = exception_enter();
968 
969 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
970 	       regs->nip, regs->msr, regs->trap);
971 
972 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
973 
974 	exception_exit(prev_state);
975 }
976 
977 void instruction_breakpoint_exception(struct pt_regs *regs)
978 {
979 	enum ctx_state prev_state = exception_enter();
980 
981 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
982 					5, SIGTRAP) == NOTIFY_STOP)
983 		goto bail;
984 	if (debugger_iabr_match(regs))
985 		goto bail;
986 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
987 
988 bail:
989 	exception_exit(prev_state);
990 }
991 
992 void RunModeException(struct pt_regs *regs)
993 {
994 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
995 }
996 
997 void single_step_exception(struct pt_regs *regs)
998 {
999 	enum ctx_state prev_state = exception_enter();
1000 
1001 	clear_single_step(regs);
1002 	clear_br_trace(regs);
1003 
1004 	if (kprobe_post_handler(regs))
1005 		return;
1006 
1007 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1008 					5, SIGTRAP) == NOTIFY_STOP)
1009 		goto bail;
1010 	if (debugger_sstep(regs))
1011 		goto bail;
1012 
1013 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1014 
1015 bail:
1016 	exception_exit(prev_state);
1017 }
1018 NOKPROBE_SYMBOL(single_step_exception);
1019 
1020 /*
1021  * After we have successfully emulated an instruction, we have to
1022  * check if the instruction was being single-stepped, and if so,
1023  * pretend we got a single-step exception.  This was pointed out
1024  * by Kumar Gala.  -- paulus
1025  */
1026 static void emulate_single_step(struct pt_regs *regs)
1027 {
1028 	if (single_stepping(regs))
1029 		single_step_exception(regs);
1030 }
1031 
1032 static inline int __parse_fpscr(unsigned long fpscr)
1033 {
1034 	int ret = FPE_FLTUNK;
1035 
1036 	/* Invalid operation */
1037 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1038 		ret = FPE_FLTINV;
1039 
1040 	/* Overflow */
1041 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1042 		ret = FPE_FLTOVF;
1043 
1044 	/* Underflow */
1045 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1046 		ret = FPE_FLTUND;
1047 
1048 	/* Divide by zero */
1049 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1050 		ret = FPE_FLTDIV;
1051 
1052 	/* Inexact result */
1053 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1054 		ret = FPE_FLTRES;
1055 
1056 	return ret;
1057 }
1058 
1059 static void parse_fpe(struct pt_regs *regs)
1060 {
1061 	int code = 0;
1062 
1063 	flush_fp_to_thread(current);
1064 
1065 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1066 
1067 	_exception(SIGFPE, regs, code, regs->nip);
1068 }
1069 
1070 /*
1071  * Illegal instruction emulation support.  Originally written to
1072  * provide the PVR to user applications using the mfspr rd, PVR.
1073  * Return non-zero if we can't emulate, or -EFAULT if the associated
1074  * memory access caused an access fault.  Return zero on success.
1075  *
1076  * There are a couple of ways to do this, either "decode" the instruction
1077  * or directly match lots of bits.  In this case, matching lots of
1078  * bits is faster and easier.
1079  *
1080  */
1081 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1082 {
1083 	u8 rT = (instword >> 21) & 0x1f;
1084 	u8 rA = (instword >> 16) & 0x1f;
1085 	u8 NB_RB = (instword >> 11) & 0x1f;
1086 	u32 num_bytes;
1087 	unsigned long EA;
1088 	int pos = 0;
1089 
1090 	/* Early out if we are an invalid form of lswx */
1091 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1092 		if ((rT == rA) || (rT == NB_RB))
1093 			return -EINVAL;
1094 
1095 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1096 
1097 	switch (instword & PPC_INST_STRING_MASK) {
1098 		case PPC_INST_LSWX:
1099 		case PPC_INST_STSWX:
1100 			EA += NB_RB;
1101 			num_bytes = regs->xer & 0x7f;
1102 			break;
1103 		case PPC_INST_LSWI:
1104 		case PPC_INST_STSWI:
1105 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1106 			break;
1107 		default:
1108 			return -EINVAL;
1109 	}
1110 
1111 	while (num_bytes != 0)
1112 	{
1113 		u8 val;
1114 		u32 shift = 8 * (3 - (pos & 0x3));
1115 
1116 		/* if process is 32-bit, clear upper 32 bits of EA */
1117 		if ((regs->msr & MSR_64BIT) == 0)
1118 			EA &= 0xFFFFFFFF;
1119 
1120 		switch ((instword & PPC_INST_STRING_MASK)) {
1121 			case PPC_INST_LSWX:
1122 			case PPC_INST_LSWI:
1123 				if (get_user(val, (u8 __user *)EA))
1124 					return -EFAULT;
1125 				/* first time updating this reg,
1126 				 * zero it out */
1127 				if (pos == 0)
1128 					regs->gpr[rT] = 0;
1129 				regs->gpr[rT] |= val << shift;
1130 				break;
1131 			case PPC_INST_STSWI:
1132 			case PPC_INST_STSWX:
1133 				val = regs->gpr[rT] >> shift;
1134 				if (put_user(val, (u8 __user *)EA))
1135 					return -EFAULT;
1136 				break;
1137 		}
1138 		/* move EA to next address */
1139 		EA += 1;
1140 		num_bytes--;
1141 
1142 		/* manage our position within the register */
1143 		if (++pos == 4) {
1144 			pos = 0;
1145 			if (++rT == 32)
1146 				rT = 0;
1147 		}
1148 	}
1149 
1150 	return 0;
1151 }
1152 
1153 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1154 {
1155 	u32 ra,rs;
1156 	unsigned long tmp;
1157 
1158 	ra = (instword >> 16) & 0x1f;
1159 	rs = (instword >> 21) & 0x1f;
1160 
1161 	tmp = regs->gpr[rs];
1162 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1163 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1164 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1165 	regs->gpr[ra] = tmp;
1166 
1167 	return 0;
1168 }
1169 
1170 static int emulate_isel(struct pt_regs *regs, u32 instword)
1171 {
1172 	u8 rT = (instword >> 21) & 0x1f;
1173 	u8 rA = (instword >> 16) & 0x1f;
1174 	u8 rB = (instword >> 11) & 0x1f;
1175 	u8 BC = (instword >> 6) & 0x1f;
1176 	u8 bit;
1177 	unsigned long tmp;
1178 
1179 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1180 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1181 
1182 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1183 
1184 	return 0;
1185 }
1186 
1187 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1188 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1189 {
1190         /* If we're emulating a load/store in an active transaction, we cannot
1191          * emulate it as the kernel operates in transaction suspended context.
1192          * We need to abort the transaction.  This creates a persistent TM
1193          * abort so tell the user what caused it with a new code.
1194 	 */
1195 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1196 		tm_enable();
1197 		tm_abort(cause);
1198 		return true;
1199 	}
1200 	return false;
1201 }
1202 #else
1203 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1204 {
1205 	return false;
1206 }
1207 #endif
1208 
1209 static int emulate_instruction(struct pt_regs *regs)
1210 {
1211 	u32 instword;
1212 	u32 rd;
1213 
1214 	if (!user_mode(regs))
1215 		return -EINVAL;
1216 	CHECK_FULL_REGS(regs);
1217 
1218 	if (get_user(instword, (u32 __user *)(regs->nip)))
1219 		return -EFAULT;
1220 
1221 	/* Emulate the mfspr rD, PVR. */
1222 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1223 		PPC_WARN_EMULATED(mfpvr, regs);
1224 		rd = (instword >> 21) & 0x1f;
1225 		regs->gpr[rd] = mfspr(SPRN_PVR);
1226 		return 0;
1227 	}
1228 
1229 	/* Emulating the dcba insn is just a no-op.  */
1230 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1231 		PPC_WARN_EMULATED(dcba, regs);
1232 		return 0;
1233 	}
1234 
1235 	/* Emulate the mcrxr insn.  */
1236 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1237 		int shift = (instword >> 21) & 0x1c;
1238 		unsigned long msk = 0xf0000000UL >> shift;
1239 
1240 		PPC_WARN_EMULATED(mcrxr, regs);
1241 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1242 		regs->xer &= ~0xf0000000UL;
1243 		return 0;
1244 	}
1245 
1246 	/* Emulate load/store string insn. */
1247 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1248 		if (tm_abort_check(regs,
1249 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1250 			return -EINVAL;
1251 		PPC_WARN_EMULATED(string, regs);
1252 		return emulate_string_inst(regs, instword);
1253 	}
1254 
1255 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1256 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1257 		PPC_WARN_EMULATED(popcntb, regs);
1258 		return emulate_popcntb_inst(regs, instword);
1259 	}
1260 
1261 	/* Emulate isel (Integer Select) instruction */
1262 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1263 		PPC_WARN_EMULATED(isel, regs);
1264 		return emulate_isel(regs, instword);
1265 	}
1266 
1267 	/* Emulate sync instruction variants */
1268 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1269 		PPC_WARN_EMULATED(sync, regs);
1270 		asm volatile("sync");
1271 		return 0;
1272 	}
1273 
1274 #ifdef CONFIG_PPC64
1275 	/* Emulate the mfspr rD, DSCR. */
1276 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1277 		PPC_INST_MFSPR_DSCR_USER) ||
1278 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1279 		PPC_INST_MFSPR_DSCR)) &&
1280 			cpu_has_feature(CPU_FTR_DSCR)) {
1281 		PPC_WARN_EMULATED(mfdscr, regs);
1282 		rd = (instword >> 21) & 0x1f;
1283 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1284 		return 0;
1285 	}
1286 	/* Emulate the mtspr DSCR, rD. */
1287 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1288 		PPC_INST_MTSPR_DSCR_USER) ||
1289 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1290 		PPC_INST_MTSPR_DSCR)) &&
1291 			cpu_has_feature(CPU_FTR_DSCR)) {
1292 		PPC_WARN_EMULATED(mtdscr, regs);
1293 		rd = (instword >> 21) & 0x1f;
1294 		current->thread.dscr = regs->gpr[rd];
1295 		current->thread.dscr_inherit = 1;
1296 		mtspr(SPRN_DSCR, current->thread.dscr);
1297 		return 0;
1298 	}
1299 #endif
1300 
1301 	return -EINVAL;
1302 }
1303 
1304 int is_valid_bugaddr(unsigned long addr)
1305 {
1306 	return is_kernel_addr(addr);
1307 }
1308 
1309 #ifdef CONFIG_MATH_EMULATION
1310 static int emulate_math(struct pt_regs *regs)
1311 {
1312 	int ret;
1313 	extern int do_mathemu(struct pt_regs *regs);
1314 
1315 	ret = do_mathemu(regs);
1316 	if (ret >= 0)
1317 		PPC_WARN_EMULATED(math, regs);
1318 
1319 	switch (ret) {
1320 	case 0:
1321 		emulate_single_step(regs);
1322 		return 0;
1323 	case 1: {
1324 			int code = 0;
1325 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1326 			_exception(SIGFPE, regs, code, regs->nip);
1327 			return 0;
1328 		}
1329 	case -EFAULT:
1330 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1331 		return 0;
1332 	}
1333 
1334 	return -1;
1335 }
1336 #else
1337 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1338 #endif
1339 
1340 void program_check_exception(struct pt_regs *regs)
1341 {
1342 	enum ctx_state prev_state = exception_enter();
1343 	unsigned int reason = get_reason(regs);
1344 
1345 	/* We can now get here via a FP Unavailable exception if the core
1346 	 * has no FPU, in that case the reason flags will be 0 */
1347 
1348 	if (reason & REASON_FP) {
1349 		/* IEEE FP exception */
1350 		parse_fpe(regs);
1351 		goto bail;
1352 	}
1353 	if (reason & REASON_TRAP) {
1354 		unsigned long bugaddr;
1355 		/* Debugger is first in line to stop recursive faults in
1356 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1357 		if (debugger_bpt(regs))
1358 			goto bail;
1359 
1360 		if (kprobe_handler(regs))
1361 			goto bail;
1362 
1363 		/* trap exception */
1364 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1365 				== NOTIFY_STOP)
1366 			goto bail;
1367 
1368 		bugaddr = regs->nip;
1369 		/*
1370 		 * Fixup bugaddr for BUG_ON() in real mode
1371 		 */
1372 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1373 			bugaddr += PAGE_OFFSET;
1374 
1375 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1376 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1377 			regs->nip += 4;
1378 			goto bail;
1379 		}
1380 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1381 		goto bail;
1382 	}
1383 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1384 	if (reason & REASON_TM) {
1385 		/* This is a TM "Bad Thing Exception" program check.
1386 		 * This occurs when:
1387 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1388 		 *    transition in TM states.
1389 		 * -  A trechkpt is attempted when transactional.
1390 		 * -  A treclaim is attempted when non transactional.
1391 		 * -  A tend is illegally attempted.
1392 		 * -  writing a TM SPR when transactional.
1393 		 *
1394 		 * If usermode caused this, it's done something illegal and
1395 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1396 		 * operand to distinguish from the instruction just being bad
1397 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1398 		 * illegal /placement/ of a valid instruction.
1399 		 */
1400 		if (user_mode(regs)) {
1401 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1402 			goto bail;
1403 		} else {
1404 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1405 			       "at %lx (msr 0x%x)\n", regs->nip, reason);
1406 			die("Unrecoverable exception", regs, SIGABRT);
1407 		}
1408 	}
1409 #endif
1410 
1411 	/*
1412 	 * If we took the program check in the kernel skip down to sending a
1413 	 * SIGILL. The subsequent cases all relate to emulating instructions
1414 	 * which we should only do for userspace. We also do not want to enable
1415 	 * interrupts for kernel faults because that might lead to further
1416 	 * faults, and loose the context of the original exception.
1417 	 */
1418 	if (!user_mode(regs))
1419 		goto sigill;
1420 
1421 	/* We restore the interrupt state now */
1422 	if (!arch_irq_disabled_regs(regs))
1423 		local_irq_enable();
1424 
1425 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1426 	 * but there seems to be a hardware bug on the 405GP (RevD)
1427 	 * that means ESR is sometimes set incorrectly - either to
1428 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1429 	 * hardware people - not sure if it can happen on any illegal
1430 	 * instruction or only on FP instructions, whether there is a
1431 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1432 	 */
1433 	if (!emulate_math(regs))
1434 		goto bail;
1435 
1436 	/* Try to emulate it if we should. */
1437 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1438 		switch (emulate_instruction(regs)) {
1439 		case 0:
1440 			regs->nip += 4;
1441 			emulate_single_step(regs);
1442 			goto bail;
1443 		case -EFAULT:
1444 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1445 			goto bail;
1446 		}
1447 	}
1448 
1449 sigill:
1450 	if (reason & REASON_PRIVILEGED)
1451 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1452 	else
1453 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1454 
1455 bail:
1456 	exception_exit(prev_state);
1457 }
1458 NOKPROBE_SYMBOL(program_check_exception);
1459 
1460 /*
1461  * This occurs when running in hypervisor mode on POWER6 or later
1462  * and an illegal instruction is encountered.
1463  */
1464 void emulation_assist_interrupt(struct pt_regs *regs)
1465 {
1466 	regs->msr |= REASON_ILLEGAL;
1467 	program_check_exception(regs);
1468 }
1469 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1470 
1471 void alignment_exception(struct pt_regs *regs)
1472 {
1473 	enum ctx_state prev_state = exception_enter();
1474 	int sig, code, fixed = 0;
1475 
1476 	/* We restore the interrupt state now */
1477 	if (!arch_irq_disabled_regs(regs))
1478 		local_irq_enable();
1479 
1480 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1481 		goto bail;
1482 
1483 	/* we don't implement logging of alignment exceptions */
1484 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1485 		fixed = fix_alignment(regs);
1486 
1487 	if (fixed == 1) {
1488 		regs->nip += 4;	/* skip over emulated instruction */
1489 		emulate_single_step(regs);
1490 		goto bail;
1491 	}
1492 
1493 	/* Operand address was bad */
1494 	if (fixed == -EFAULT) {
1495 		sig = SIGSEGV;
1496 		code = SEGV_ACCERR;
1497 	} else {
1498 		sig = SIGBUS;
1499 		code = BUS_ADRALN;
1500 	}
1501 	if (user_mode(regs))
1502 		_exception(sig, regs, code, regs->dar);
1503 	else
1504 		bad_page_fault(regs, regs->dar, sig);
1505 
1506 bail:
1507 	exception_exit(prev_state);
1508 }
1509 
1510 void StackOverflow(struct pt_regs *regs)
1511 {
1512 	printk(KERN_CRIT "Kernel stack overflow in process %p, r1=%lx\n",
1513 	       current, regs->gpr[1]);
1514 	debugger(regs);
1515 	show_regs(regs);
1516 	panic("kernel stack overflow");
1517 }
1518 
1519 void nonrecoverable_exception(struct pt_regs *regs)
1520 {
1521 	printk(KERN_ERR "Non-recoverable exception at PC=%lx MSR=%lx\n",
1522 	       regs->nip, regs->msr);
1523 	debugger(regs);
1524 	die("nonrecoverable exception", regs, SIGKILL);
1525 }
1526 
1527 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1528 {
1529 	enum ctx_state prev_state = exception_enter();
1530 
1531 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1532 			  "%lx at %lx\n", regs->trap, regs->nip);
1533 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1534 
1535 	exception_exit(prev_state);
1536 }
1537 
1538 void altivec_unavailable_exception(struct pt_regs *regs)
1539 {
1540 	enum ctx_state prev_state = exception_enter();
1541 
1542 	if (user_mode(regs)) {
1543 		/* A user program has executed an altivec instruction,
1544 		   but this kernel doesn't support altivec. */
1545 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1546 		goto bail;
1547 	}
1548 
1549 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1550 			"%lx at %lx\n", regs->trap, regs->nip);
1551 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1552 
1553 bail:
1554 	exception_exit(prev_state);
1555 }
1556 
1557 void vsx_unavailable_exception(struct pt_regs *regs)
1558 {
1559 	if (user_mode(regs)) {
1560 		/* A user program has executed an vsx instruction,
1561 		   but this kernel doesn't support vsx. */
1562 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1563 		return;
1564 	}
1565 
1566 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1567 			"%lx at %lx\n", regs->trap, regs->nip);
1568 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1569 }
1570 
1571 #ifdef CONFIG_PPC64
1572 static void tm_unavailable(struct pt_regs *regs)
1573 {
1574 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1575 	if (user_mode(regs)) {
1576 		current->thread.load_tm++;
1577 		regs->msr |= MSR_TM;
1578 		tm_enable();
1579 		tm_restore_sprs(&current->thread);
1580 		return;
1581 	}
1582 #endif
1583 	pr_emerg("Unrecoverable TM Unavailable Exception "
1584 			"%lx at %lx\n", regs->trap, regs->nip);
1585 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1586 }
1587 
1588 void facility_unavailable_exception(struct pt_regs *regs)
1589 {
1590 	static char *facility_strings[] = {
1591 		[FSCR_FP_LG] = "FPU",
1592 		[FSCR_VECVSX_LG] = "VMX/VSX",
1593 		[FSCR_DSCR_LG] = "DSCR",
1594 		[FSCR_PM_LG] = "PMU SPRs",
1595 		[FSCR_BHRB_LG] = "BHRB",
1596 		[FSCR_TM_LG] = "TM",
1597 		[FSCR_EBB_LG] = "EBB",
1598 		[FSCR_TAR_LG] = "TAR",
1599 		[FSCR_MSGP_LG] = "MSGP",
1600 		[FSCR_SCV_LG] = "SCV",
1601 	};
1602 	char *facility = "unknown";
1603 	u64 value;
1604 	u32 instword, rd;
1605 	u8 status;
1606 	bool hv;
1607 
1608 	hv = (TRAP(regs) == 0xf80);
1609 	if (hv)
1610 		value = mfspr(SPRN_HFSCR);
1611 	else
1612 		value = mfspr(SPRN_FSCR);
1613 
1614 	status = value >> 56;
1615 	if ((hv || status >= 2) &&
1616 	    (status < ARRAY_SIZE(facility_strings)) &&
1617 	    facility_strings[status])
1618 		facility = facility_strings[status];
1619 
1620 	/* We should not have taken this interrupt in kernel */
1621 	if (!user_mode(regs)) {
1622 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1623 			 facility, status, regs->nip);
1624 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1625 	}
1626 
1627 	/* We restore the interrupt state now */
1628 	if (!arch_irq_disabled_regs(regs))
1629 		local_irq_enable();
1630 
1631 	if (status == FSCR_DSCR_LG) {
1632 		/*
1633 		 * User is accessing the DSCR register using the problem
1634 		 * state only SPR number (0x03) either through a mfspr or
1635 		 * a mtspr instruction. If it is a write attempt through
1636 		 * a mtspr, then we set the inherit bit. This also allows
1637 		 * the user to write or read the register directly in the
1638 		 * future by setting via the FSCR DSCR bit. But in case it
1639 		 * is a read DSCR attempt through a mfspr instruction, we
1640 		 * just emulate the instruction instead. This code path will
1641 		 * always emulate all the mfspr instructions till the user
1642 		 * has attempted at least one mtspr instruction. This way it
1643 		 * preserves the same behaviour when the user is accessing
1644 		 * the DSCR through privilege level only SPR number (0x11)
1645 		 * which is emulated through illegal instruction exception.
1646 		 * We always leave HFSCR DSCR set.
1647 		 */
1648 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1649 			pr_err("Failed to fetch the user instruction\n");
1650 			return;
1651 		}
1652 
1653 		/* Write into DSCR (mtspr 0x03, RS) */
1654 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1655 				== PPC_INST_MTSPR_DSCR_USER) {
1656 			rd = (instword >> 21) & 0x1f;
1657 			current->thread.dscr = regs->gpr[rd];
1658 			current->thread.dscr_inherit = 1;
1659 			current->thread.fscr |= FSCR_DSCR;
1660 			mtspr(SPRN_FSCR, current->thread.fscr);
1661 		}
1662 
1663 		/* Read from DSCR (mfspr RT, 0x03) */
1664 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1665 				== PPC_INST_MFSPR_DSCR_USER) {
1666 			if (emulate_instruction(regs)) {
1667 				pr_err("DSCR based mfspr emulation failed\n");
1668 				return;
1669 			}
1670 			regs->nip += 4;
1671 			emulate_single_step(regs);
1672 		}
1673 		return;
1674 	}
1675 
1676 	if (status == FSCR_TM_LG) {
1677 		/*
1678 		 * If we're here then the hardware is TM aware because it
1679 		 * generated an exception with FSRM_TM set.
1680 		 *
1681 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1682 		 * told us not to do TM, or the kernel is not built with TM
1683 		 * support.
1684 		 *
1685 		 * If both of those things are true, then userspace can spam the
1686 		 * console by triggering the printk() below just by continually
1687 		 * doing tbegin (or any TM instruction). So in that case just
1688 		 * send the process a SIGILL immediately.
1689 		 */
1690 		if (!cpu_has_feature(CPU_FTR_TM))
1691 			goto out;
1692 
1693 		tm_unavailable(regs);
1694 		return;
1695 	}
1696 
1697 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1698 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1699 
1700 out:
1701 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1702 }
1703 #endif
1704 
1705 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1706 
1707 void fp_unavailable_tm(struct pt_regs *regs)
1708 {
1709 	/* Note:  This does not handle any kind of FP laziness. */
1710 
1711 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1712 		 regs->nip, regs->msr);
1713 
1714         /* We can only have got here if the task started using FP after
1715          * beginning the transaction.  So, the transactional regs are just a
1716          * copy of the checkpointed ones.  But, we still need to recheckpoint
1717          * as we're enabling FP for the process; it will return, abort the
1718          * transaction, and probably retry but now with FP enabled.  So the
1719          * checkpointed FP registers need to be loaded.
1720 	 */
1721 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1722 	/* Reclaim didn't save out any FPRs to transact_fprs. */
1723 
1724 	/* Enable FP for the task: */
1725 	current->thread.load_fp = 1;
1726 
1727 	/* This loads and recheckpoints the FP registers from
1728 	 * thread.fpr[].  They will remain in registers after the
1729 	 * checkpoint so we don't need to reload them after.
1730 	 * If VMX is in use, the VRs now hold checkpointed values,
1731 	 * so we don't want to load the VRs from the thread_struct.
1732 	 */
1733 	tm_recheckpoint(&current->thread);
1734 }
1735 
1736 void altivec_unavailable_tm(struct pt_regs *regs)
1737 {
1738 	/* See the comments in fp_unavailable_tm().  This function operates
1739 	 * the same way.
1740 	 */
1741 
1742 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1743 		 "MSR=%lx\n",
1744 		 regs->nip, regs->msr);
1745 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1746 	current->thread.load_vec = 1;
1747 	tm_recheckpoint(&current->thread);
1748 	current->thread.used_vr = 1;
1749 }
1750 
1751 void vsx_unavailable_tm(struct pt_regs *regs)
1752 {
1753 	/* See the comments in fp_unavailable_tm().  This works similarly,
1754 	 * though we're loading both FP and VEC registers in here.
1755 	 *
1756 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1757 	 * regs.  Either way, set MSR_VSX.
1758 	 */
1759 
1760 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1761 		 "MSR=%lx\n",
1762 		 regs->nip, regs->msr);
1763 
1764 	current->thread.used_vsr = 1;
1765 
1766 	/* This reclaims FP and/or VR regs if they're already enabled */
1767 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1768 
1769 	current->thread.load_vec = 1;
1770 	current->thread.load_fp = 1;
1771 
1772 	tm_recheckpoint(&current->thread);
1773 }
1774 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1775 
1776 void performance_monitor_exception(struct pt_regs *regs)
1777 {
1778 	__this_cpu_inc(irq_stat.pmu_irqs);
1779 
1780 	perf_irq(regs);
1781 }
1782 
1783 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1784 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1785 {
1786 	int changed = 0;
1787 	/*
1788 	 * Determine the cause of the debug event, clear the
1789 	 * event flags and send a trap to the handler. Torez
1790 	 */
1791 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1792 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1793 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1794 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1795 #endif
1796 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1797 			     5);
1798 		changed |= 0x01;
1799 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1800 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1801 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1802 			     6);
1803 		changed |= 0x01;
1804 	}  else if (debug_status & DBSR_IAC1) {
1805 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1806 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1807 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1808 			     1);
1809 		changed |= 0x01;
1810 	}  else if (debug_status & DBSR_IAC2) {
1811 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1812 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1813 			     2);
1814 		changed |= 0x01;
1815 	}  else if (debug_status & DBSR_IAC3) {
1816 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1817 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1818 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1819 			     3);
1820 		changed |= 0x01;
1821 	}  else if (debug_status & DBSR_IAC4) {
1822 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1823 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1824 			     4);
1825 		changed |= 0x01;
1826 	}
1827 	/*
1828 	 * At the point this routine was called, the MSR(DE) was turned off.
1829 	 * Check all other debug flags and see if that bit needs to be turned
1830 	 * back on or not.
1831 	 */
1832 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1833 			       current->thread.debug.dbcr1))
1834 		regs->msr |= MSR_DE;
1835 	else
1836 		/* Make sure the IDM flag is off */
1837 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1838 
1839 	if (changed & 0x01)
1840 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1841 }
1842 
1843 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1844 {
1845 	current->thread.debug.dbsr = debug_status;
1846 
1847 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1848 	 * on server, it stops on the target of the branch. In order to simulate
1849 	 * the server behaviour, we thus restart right away with a single step
1850 	 * instead of stopping here when hitting a BT
1851 	 */
1852 	if (debug_status & DBSR_BT) {
1853 		regs->msr &= ~MSR_DE;
1854 
1855 		/* Disable BT */
1856 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1857 		/* Clear the BT event */
1858 		mtspr(SPRN_DBSR, DBSR_BT);
1859 
1860 		/* Do the single step trick only when coming from userspace */
1861 		if (user_mode(regs)) {
1862 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1863 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1864 			regs->msr |= MSR_DE;
1865 			return;
1866 		}
1867 
1868 		if (kprobe_post_handler(regs))
1869 			return;
1870 
1871 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1872 			       5, SIGTRAP) == NOTIFY_STOP) {
1873 			return;
1874 		}
1875 		if (debugger_sstep(regs))
1876 			return;
1877 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1878 		regs->msr &= ~MSR_DE;
1879 
1880 		/* Disable instruction completion */
1881 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
1882 		/* Clear the instruction completion event */
1883 		mtspr(SPRN_DBSR, DBSR_IC);
1884 
1885 		if (kprobe_post_handler(regs))
1886 			return;
1887 
1888 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1889 			       5, SIGTRAP) == NOTIFY_STOP) {
1890 			return;
1891 		}
1892 
1893 		if (debugger_sstep(regs))
1894 			return;
1895 
1896 		if (user_mode(regs)) {
1897 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
1898 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1899 					       current->thread.debug.dbcr1))
1900 				regs->msr |= MSR_DE;
1901 			else
1902 				/* Make sure the IDM bit is off */
1903 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1904 		}
1905 
1906 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1907 	} else
1908 		handle_debug(regs, debug_status);
1909 }
1910 NOKPROBE_SYMBOL(DebugException);
1911 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
1912 
1913 #if !defined(CONFIG_TAU_INT)
1914 void TAUException(struct pt_regs *regs)
1915 {
1916 	printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
1917 	       regs->nip, regs->msr, regs->trap, print_tainted());
1918 }
1919 #endif /* CONFIG_INT_TAU */
1920 
1921 #ifdef CONFIG_ALTIVEC
1922 void altivec_assist_exception(struct pt_regs *regs)
1923 {
1924 	int err;
1925 
1926 	if (!user_mode(regs)) {
1927 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
1928 		       " at %lx\n", regs->nip);
1929 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
1930 	}
1931 
1932 	flush_altivec_to_thread(current);
1933 
1934 	PPC_WARN_EMULATED(altivec, regs);
1935 	err = emulate_altivec(regs);
1936 	if (err == 0) {
1937 		regs->nip += 4;		/* skip emulated instruction */
1938 		emulate_single_step(regs);
1939 		return;
1940 	}
1941 
1942 	if (err == -EFAULT) {
1943 		/* got an error reading the instruction */
1944 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
1945 	} else {
1946 		/* didn't recognize the instruction */
1947 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
1948 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
1949 				   "in %s at %lx\n", current->comm, regs->nip);
1950 		current->thread.vr_state.vscr.u[3] |= 0x10000;
1951 	}
1952 }
1953 #endif /* CONFIG_ALTIVEC */
1954 
1955 #ifdef CONFIG_FSL_BOOKE
1956 void CacheLockingException(struct pt_regs *regs, unsigned long address,
1957 			   unsigned long error_code)
1958 {
1959 	/* We treat cache locking instructions from the user
1960 	 * as priv ops, in the future we could try to do
1961 	 * something smarter
1962 	 */
1963 	if (error_code & (ESR_DLK|ESR_ILK))
1964 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1965 	return;
1966 }
1967 #endif /* CONFIG_FSL_BOOKE */
1968 
1969 #ifdef CONFIG_SPE
1970 void SPEFloatingPointException(struct pt_regs *regs)
1971 {
1972 	extern int do_spe_mathemu(struct pt_regs *regs);
1973 	unsigned long spefscr;
1974 	int fpexc_mode;
1975 	int code = FPE_FLTUNK;
1976 	int err;
1977 
1978 	flush_spe_to_thread(current);
1979 
1980 	spefscr = current->thread.spefscr;
1981 	fpexc_mode = current->thread.fpexc_mode;
1982 
1983 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
1984 		code = FPE_FLTOVF;
1985 	}
1986 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
1987 		code = FPE_FLTUND;
1988 	}
1989 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
1990 		code = FPE_FLTDIV;
1991 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
1992 		code = FPE_FLTINV;
1993 	}
1994 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
1995 		code = FPE_FLTRES;
1996 
1997 	err = do_spe_mathemu(regs);
1998 	if (err == 0) {
1999 		regs->nip += 4;		/* skip emulated instruction */
2000 		emulate_single_step(regs);
2001 		return;
2002 	}
2003 
2004 	if (err == -EFAULT) {
2005 		/* got an error reading the instruction */
2006 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2007 	} else if (err == -EINVAL) {
2008 		/* didn't recognize the instruction */
2009 		printk(KERN_ERR "unrecognized spe instruction "
2010 		       "in %s at %lx\n", current->comm, regs->nip);
2011 	} else {
2012 		_exception(SIGFPE, regs, code, regs->nip);
2013 	}
2014 
2015 	return;
2016 }
2017 
2018 void SPEFloatingPointRoundException(struct pt_regs *regs)
2019 {
2020 	extern int speround_handler(struct pt_regs *regs);
2021 	int err;
2022 
2023 	preempt_disable();
2024 	if (regs->msr & MSR_SPE)
2025 		giveup_spe(current);
2026 	preempt_enable();
2027 
2028 	regs->nip -= 4;
2029 	err = speround_handler(regs);
2030 	if (err == 0) {
2031 		regs->nip += 4;		/* skip emulated instruction */
2032 		emulate_single_step(regs);
2033 		return;
2034 	}
2035 
2036 	if (err == -EFAULT) {
2037 		/* got an error reading the instruction */
2038 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2039 	} else if (err == -EINVAL) {
2040 		/* didn't recognize the instruction */
2041 		printk(KERN_ERR "unrecognized spe instruction "
2042 		       "in %s at %lx\n", current->comm, regs->nip);
2043 	} else {
2044 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2045 		return;
2046 	}
2047 }
2048 #endif
2049 
2050 /*
2051  * We enter here if we get an unrecoverable exception, that is, one
2052  * that happened at a point where the RI (recoverable interrupt) bit
2053  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2054  * we therefore lost state by taking this exception.
2055  */
2056 void unrecoverable_exception(struct pt_regs *regs)
2057 {
2058 	printk(KERN_EMERG "Unrecoverable exception %lx at %lx\n",
2059 	       regs->trap, regs->nip);
2060 	die("Unrecoverable exception", regs, SIGABRT);
2061 }
2062 NOKPROBE_SYMBOL(unrecoverable_exception);
2063 
2064 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2065 /*
2066  * Default handler for a Watchdog exception,
2067  * spins until a reboot occurs
2068  */
2069 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2070 {
2071 	/* Generic WatchdogHandler, implement your own */
2072 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2073 	return;
2074 }
2075 
2076 void WatchdogException(struct pt_regs *regs)
2077 {
2078 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2079 	WatchdogHandler(regs);
2080 }
2081 #endif
2082 
2083 /*
2084  * We enter here if we discover during exception entry that we are
2085  * running in supervisor mode with a userspace value in the stack pointer.
2086  */
2087 void kernel_bad_stack(struct pt_regs *regs)
2088 {
2089 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2090 	       regs->gpr[1], regs->nip);
2091 	die("Bad kernel stack pointer", regs, SIGABRT);
2092 }
2093 NOKPROBE_SYMBOL(kernel_bad_stack);
2094 
2095 void __init trap_init(void)
2096 {
2097 }
2098 
2099 
2100 #ifdef CONFIG_PPC_EMULATED_STATS
2101 
2102 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2103 
2104 struct ppc_emulated ppc_emulated = {
2105 #ifdef CONFIG_ALTIVEC
2106 	WARN_EMULATED_SETUP(altivec),
2107 #endif
2108 	WARN_EMULATED_SETUP(dcba),
2109 	WARN_EMULATED_SETUP(dcbz),
2110 	WARN_EMULATED_SETUP(fp_pair),
2111 	WARN_EMULATED_SETUP(isel),
2112 	WARN_EMULATED_SETUP(mcrxr),
2113 	WARN_EMULATED_SETUP(mfpvr),
2114 	WARN_EMULATED_SETUP(multiple),
2115 	WARN_EMULATED_SETUP(popcntb),
2116 	WARN_EMULATED_SETUP(spe),
2117 	WARN_EMULATED_SETUP(string),
2118 	WARN_EMULATED_SETUP(sync),
2119 	WARN_EMULATED_SETUP(unaligned),
2120 #ifdef CONFIG_MATH_EMULATION
2121 	WARN_EMULATED_SETUP(math),
2122 #endif
2123 #ifdef CONFIG_VSX
2124 	WARN_EMULATED_SETUP(vsx),
2125 #endif
2126 #ifdef CONFIG_PPC64
2127 	WARN_EMULATED_SETUP(mfdscr),
2128 	WARN_EMULATED_SETUP(mtdscr),
2129 	WARN_EMULATED_SETUP(lq_stq),
2130 	WARN_EMULATED_SETUP(lxvw4x),
2131 	WARN_EMULATED_SETUP(lxvh8x),
2132 	WARN_EMULATED_SETUP(lxvd2x),
2133 	WARN_EMULATED_SETUP(lxvb16x),
2134 #endif
2135 };
2136 
2137 u32 ppc_warn_emulated;
2138 
2139 void ppc_warn_emulated_print(const char *type)
2140 {
2141 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2142 			    type);
2143 }
2144 
2145 static int __init ppc_warn_emulated_init(void)
2146 {
2147 	struct dentry *dir, *d;
2148 	unsigned int i;
2149 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2150 
2151 	if (!powerpc_debugfs_root)
2152 		return -ENODEV;
2153 
2154 	dir = debugfs_create_dir("emulated_instructions",
2155 				 powerpc_debugfs_root);
2156 	if (!dir)
2157 		return -ENOMEM;
2158 
2159 	d = debugfs_create_u32("do_warn", 0644, dir,
2160 			       &ppc_warn_emulated);
2161 	if (!d)
2162 		goto fail;
2163 
2164 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) {
2165 		d = debugfs_create_u32(entries[i].name, 0644, dir,
2166 				       (u32 *)&entries[i].val.counter);
2167 		if (!d)
2168 			goto fail;
2169 	}
2170 
2171 	return 0;
2172 
2173 fail:
2174 	debugfs_remove_recursive(dir);
2175 	return -ENOMEM;
2176 }
2177 
2178 device_initcall(ppc_warn_emulated_init);
2179 
2180 #endif /* CONFIG_PPC_EMULATED_STATS */
2181