xref: /linux/arch/powerpc/kernel/traps.c (revision 96a6de1a541c86e9e67b9c310c14db4099bd1cbc)
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 /*
373  * The interrupt architecture has a quirk in that the HV interrupts excluding
374  * the NMIs (0x100 and 0x200) do not clear MSR[RI] at entry. The first thing
375  * that an interrupt handler must do is save off a GPR into a scratch register,
376  * and all interrupts on POWERNV (HV=1) use the HSPRG1 register as scratch.
377  * Therefore an NMI can clobber an HV interrupt's live HSPRG1 without noticing
378  * that it is non-reentrant, which leads to random data corruption.
379  *
380  * The solution is for NMI interrupts in HV mode to check if they originated
381  * from these critical HV interrupt regions. If so, then mark them not
382  * recoverable.
383  *
384  * An alternative would be for HV NMIs to use SPRG for scratch to avoid the
385  * HSPRG1 clobber, however this would cause guest SPRG to be clobbered. Linux
386  * guests should always have MSR[RI]=0 when its scratch SPRG is in use, so
387  * that would work. However any other guest OS that may have the SPRG live
388  * and MSR[RI]=1 could encounter silent corruption.
389  *
390  * Builds that do not support KVM could take this second option to increase
391  * the recoverability of NMIs.
392  */
393 void hv_nmi_check_nonrecoverable(struct pt_regs *regs)
394 {
395 #ifdef CONFIG_PPC_POWERNV
396 	unsigned long kbase = (unsigned long)_stext;
397 	unsigned long nip = regs->nip;
398 
399 	if (!(regs->msr & MSR_RI))
400 		return;
401 	if (!(regs->msr & MSR_HV))
402 		return;
403 	if (regs->msr & MSR_PR)
404 		return;
405 
406 	/*
407 	 * Now test if the interrupt has hit a range that may be using
408 	 * HSPRG1 without having RI=0 (i.e., an HSRR interrupt). The
409 	 * problem ranges all run un-relocated. Test real and virt modes
410 	 * at the same time by droping the high bit of the nip (virt mode
411 	 * entry points still have the +0x4000 offset).
412 	 */
413 	nip &= ~0xc000000000000000ULL;
414 	if ((nip >= 0x500 && nip < 0x600) || (nip >= 0x4500 && nip < 0x4600))
415 		goto nonrecoverable;
416 	if ((nip >= 0x980 && nip < 0xa00) || (nip >= 0x4980 && nip < 0x4a00))
417 		goto nonrecoverable;
418 	if ((nip >= 0xe00 && nip < 0xec0) || (nip >= 0x4e00 && nip < 0x4ec0))
419 		goto nonrecoverable;
420 	if ((nip >= 0xf80 && nip < 0xfa0) || (nip >= 0x4f80 && nip < 0x4fa0))
421 		goto nonrecoverable;
422 
423 	/* Trampoline code runs un-relocated so subtract kbase. */
424 	if (nip >= (unsigned long)(start_real_trampolines - kbase) &&
425 			nip < (unsigned long)(end_real_trampolines - kbase))
426 		goto nonrecoverable;
427 	if (nip >= (unsigned long)(start_virt_trampolines - kbase) &&
428 			nip < (unsigned long)(end_virt_trampolines - kbase))
429 		goto nonrecoverable;
430 	return;
431 
432 nonrecoverable:
433 	regs->msr &= ~MSR_RI;
434 #endif
435 }
436 
437 void system_reset_exception(struct pt_regs *regs)
438 {
439 	unsigned long hsrr0, hsrr1;
440 	bool nested = in_nmi();
441 	bool saved_hsrrs = false;
442 
443 	/*
444 	 * Avoid crashes in case of nested NMI exceptions. Recoverability
445 	 * is determined by RI and in_nmi
446 	 */
447 	if (!nested)
448 		nmi_enter();
449 
450 	/*
451 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
452 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
453 	 * OPAL), so save them here and restore them before returning.
454 	 *
455 	 * Machine checks don't need to save HSRRs, as the real mode handler
456 	 * is careful to avoid them, and the regular handler is not delivered
457 	 * as an NMI.
458 	 */
459 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
460 		hsrr0 = mfspr(SPRN_HSRR0);
461 		hsrr1 = mfspr(SPRN_HSRR1);
462 		saved_hsrrs = true;
463 	}
464 
465 	hv_nmi_check_nonrecoverable(regs);
466 
467 	__this_cpu_inc(irq_stat.sreset_irqs);
468 
469 	/* See if any machine dependent calls */
470 	if (ppc_md.system_reset_exception) {
471 		if (ppc_md.system_reset_exception(regs))
472 			goto out;
473 	}
474 
475 	if (debugger(regs))
476 		goto out;
477 
478 	/*
479 	 * A system reset is a request to dump, so we always send
480 	 * it through the crashdump code (if fadump or kdump are
481 	 * registered).
482 	 */
483 	crash_fadump(regs, "System Reset");
484 
485 	crash_kexec(regs);
486 
487 	/*
488 	 * We aren't the primary crash CPU. We need to send it
489 	 * to a holding pattern to avoid it ending up in the panic
490 	 * code.
491 	 */
492 	crash_kexec_secondary(regs);
493 
494 	/*
495 	 * No debugger or crash dump registered, print logs then
496 	 * panic.
497 	 */
498 	die("System Reset", regs, SIGABRT);
499 
500 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
501 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
502 	nmi_panic(regs, "System Reset");
503 
504 out:
505 #ifdef CONFIG_PPC_BOOK3S_64
506 	BUG_ON(get_paca()->in_nmi == 0);
507 	if (get_paca()->in_nmi > 1)
508 		nmi_panic(regs, "Unrecoverable nested System Reset");
509 #endif
510 	/* Must die if the interrupt is not recoverable */
511 	if (!(regs->msr & MSR_RI))
512 		nmi_panic(regs, "Unrecoverable System Reset");
513 
514 	if (saved_hsrrs) {
515 		mtspr(SPRN_HSRR0, hsrr0);
516 		mtspr(SPRN_HSRR1, hsrr1);
517 	}
518 
519 	if (!nested)
520 		nmi_exit();
521 
522 	/* What should we do here? We could issue a shutdown or hard reset. */
523 }
524 
525 /*
526  * I/O accesses can cause machine checks on powermacs.
527  * Check if the NIP corresponds to the address of a sync
528  * instruction for which there is an entry in the exception
529  * table.
530  * Note that the 601 only takes a machine check on TEA
531  * (transfer error ack) signal assertion, and does not
532  * set any of the top 16 bits of SRR1.
533  *  -- paulus.
534  */
535 static inline int check_io_access(struct pt_regs *regs)
536 {
537 #ifdef CONFIG_PPC32
538 	unsigned long msr = regs->msr;
539 	const struct exception_table_entry *entry;
540 	unsigned int *nip = (unsigned int *)regs->nip;
541 
542 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
543 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
544 		/*
545 		 * Check that it's a sync instruction, or somewhere
546 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
547 		 * As the address is in the exception table
548 		 * we should be able to read the instr there.
549 		 * For the debug message, we look at the preceding
550 		 * load or store.
551 		 */
552 		if (*nip == PPC_INST_NOP)
553 			nip -= 2;
554 		else if (*nip == PPC_INST_ISYNC)
555 			--nip;
556 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
557 			unsigned int rb;
558 
559 			--nip;
560 			rb = (*nip >> 11) & 0x1f;
561 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
562 			       (*nip & 0x100)? "OUT to": "IN from",
563 			       regs->gpr[rb] - _IO_BASE, nip);
564 			regs->msr |= MSR_RI;
565 			regs->nip = extable_fixup(entry);
566 			return 1;
567 		}
568 	}
569 #endif /* CONFIG_PPC32 */
570 	return 0;
571 }
572 
573 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
574 /* On 4xx, the reason for the machine check or program exception
575    is in the ESR. */
576 #define get_reason(regs)	((regs)->dsisr)
577 #define REASON_FP		ESR_FP
578 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
579 #define REASON_PRIVILEGED	ESR_PPR
580 #define REASON_TRAP		ESR_PTR
581 
582 /* single-step stuff */
583 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
584 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
585 #define clear_br_trace(regs)	do {} while(0)
586 #else
587 /* On non-4xx, the reason for the machine check or program
588    exception is in the MSR. */
589 #define get_reason(regs)	((regs)->msr)
590 #define REASON_TM		SRR1_PROGTM
591 #define REASON_FP		SRR1_PROGFPE
592 #define REASON_ILLEGAL		SRR1_PROGILL
593 #define REASON_PRIVILEGED	SRR1_PROGPRIV
594 #define REASON_TRAP		SRR1_PROGTRAP
595 
596 #define single_stepping(regs)	((regs)->msr & MSR_SE)
597 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
598 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
599 #endif
600 
601 #if defined(CONFIG_E500)
602 int machine_check_e500mc(struct pt_regs *regs)
603 {
604 	unsigned long mcsr = mfspr(SPRN_MCSR);
605 	unsigned long pvr = mfspr(SPRN_PVR);
606 	unsigned long reason = mcsr;
607 	int recoverable = 1;
608 
609 	if (reason & MCSR_LD) {
610 		recoverable = fsl_rio_mcheck_exception(regs);
611 		if (recoverable == 1)
612 			goto silent_out;
613 	}
614 
615 	printk("Machine check in kernel mode.\n");
616 	printk("Caused by (from MCSR=%lx): ", reason);
617 
618 	if (reason & MCSR_MCP)
619 		pr_cont("Machine Check Signal\n");
620 
621 	if (reason & MCSR_ICPERR) {
622 		pr_cont("Instruction Cache Parity Error\n");
623 
624 		/*
625 		 * This is recoverable by invalidating the i-cache.
626 		 */
627 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
628 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
629 			;
630 
631 		/*
632 		 * This will generally be accompanied by an instruction
633 		 * fetch error report -- only treat MCSR_IF as fatal
634 		 * if it wasn't due to an L1 parity error.
635 		 */
636 		reason &= ~MCSR_IF;
637 	}
638 
639 	if (reason & MCSR_DCPERR_MC) {
640 		pr_cont("Data Cache Parity Error\n");
641 
642 		/*
643 		 * In write shadow mode we auto-recover from the error, but it
644 		 * may still get logged and cause a machine check.  We should
645 		 * only treat the non-write shadow case as non-recoverable.
646 		 */
647 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
648 		 * is not implemented but L1 data cache always runs in write
649 		 * shadow mode. Hence on data cache parity errors HW will
650 		 * automatically invalidate the L1 Data Cache.
651 		 */
652 		if (PVR_VER(pvr) != PVR_VER_E6500) {
653 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
654 				recoverable = 0;
655 		}
656 	}
657 
658 	if (reason & MCSR_L2MMU_MHIT) {
659 		pr_cont("Hit on multiple TLB entries\n");
660 		recoverable = 0;
661 	}
662 
663 	if (reason & MCSR_NMI)
664 		pr_cont("Non-maskable interrupt\n");
665 
666 	if (reason & MCSR_IF) {
667 		pr_cont("Instruction Fetch Error Report\n");
668 		recoverable = 0;
669 	}
670 
671 	if (reason & MCSR_LD) {
672 		pr_cont("Load Error Report\n");
673 		recoverable = 0;
674 	}
675 
676 	if (reason & MCSR_ST) {
677 		pr_cont("Store Error Report\n");
678 		recoverable = 0;
679 	}
680 
681 	if (reason & MCSR_LDG) {
682 		pr_cont("Guarded Load Error Report\n");
683 		recoverable = 0;
684 	}
685 
686 	if (reason & MCSR_TLBSYNC)
687 		pr_cont("Simultaneous tlbsync operations\n");
688 
689 	if (reason & MCSR_BSL2_ERR) {
690 		pr_cont("Level 2 Cache Error\n");
691 		recoverable = 0;
692 	}
693 
694 	if (reason & MCSR_MAV) {
695 		u64 addr;
696 
697 		addr = mfspr(SPRN_MCAR);
698 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
699 
700 		pr_cont("Machine Check %s Address: %#llx\n",
701 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
702 	}
703 
704 silent_out:
705 	mtspr(SPRN_MCSR, mcsr);
706 	return mfspr(SPRN_MCSR) == 0 && recoverable;
707 }
708 
709 int machine_check_e500(struct pt_regs *regs)
710 {
711 	unsigned long reason = mfspr(SPRN_MCSR);
712 
713 	if (reason & MCSR_BUS_RBERR) {
714 		if (fsl_rio_mcheck_exception(regs))
715 			return 1;
716 		if (fsl_pci_mcheck_exception(regs))
717 			return 1;
718 	}
719 
720 	printk("Machine check in kernel mode.\n");
721 	printk("Caused by (from MCSR=%lx): ", reason);
722 
723 	if (reason & MCSR_MCP)
724 		pr_cont("Machine Check Signal\n");
725 	if (reason & MCSR_ICPERR)
726 		pr_cont("Instruction Cache Parity Error\n");
727 	if (reason & MCSR_DCP_PERR)
728 		pr_cont("Data Cache Push Parity Error\n");
729 	if (reason & MCSR_DCPERR)
730 		pr_cont("Data Cache Parity Error\n");
731 	if (reason & MCSR_BUS_IAERR)
732 		pr_cont("Bus - Instruction Address Error\n");
733 	if (reason & MCSR_BUS_RAERR)
734 		pr_cont("Bus - Read Address Error\n");
735 	if (reason & MCSR_BUS_WAERR)
736 		pr_cont("Bus - Write Address Error\n");
737 	if (reason & MCSR_BUS_IBERR)
738 		pr_cont("Bus - Instruction Data Error\n");
739 	if (reason & MCSR_BUS_RBERR)
740 		pr_cont("Bus - Read Data Bus Error\n");
741 	if (reason & MCSR_BUS_WBERR)
742 		pr_cont("Bus - Write Data Bus Error\n");
743 	if (reason & MCSR_BUS_IPERR)
744 		pr_cont("Bus - Instruction Parity Error\n");
745 	if (reason & MCSR_BUS_RPERR)
746 		pr_cont("Bus - Read Parity Error\n");
747 
748 	return 0;
749 }
750 
751 int machine_check_generic(struct pt_regs *regs)
752 {
753 	return 0;
754 }
755 #elif defined(CONFIG_E200)
756 int machine_check_e200(struct pt_regs *regs)
757 {
758 	unsigned long reason = mfspr(SPRN_MCSR);
759 
760 	printk("Machine check in kernel mode.\n");
761 	printk("Caused by (from MCSR=%lx): ", reason);
762 
763 	if (reason & MCSR_MCP)
764 		pr_cont("Machine Check Signal\n");
765 	if (reason & MCSR_CP_PERR)
766 		pr_cont("Cache Push Parity Error\n");
767 	if (reason & MCSR_CPERR)
768 		pr_cont("Cache Parity Error\n");
769 	if (reason & MCSR_EXCP_ERR)
770 		pr_cont("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
771 	if (reason & MCSR_BUS_IRERR)
772 		pr_cont("Bus - Read Bus Error on instruction fetch\n");
773 	if (reason & MCSR_BUS_DRERR)
774 		pr_cont("Bus - Read Bus Error on data load\n");
775 	if (reason & MCSR_BUS_WRERR)
776 		pr_cont("Bus - Write Bus Error on buffered store or cache line push\n");
777 
778 	return 0;
779 }
780 #elif defined(CONFIG_PPC32)
781 int machine_check_generic(struct pt_regs *regs)
782 {
783 	unsigned long reason = regs->msr;
784 
785 	printk("Machine check in kernel mode.\n");
786 	printk("Caused by (from SRR1=%lx): ", reason);
787 	switch (reason & 0x601F0000) {
788 	case 0x80000:
789 		pr_cont("Machine check signal\n");
790 		break;
791 	case 0:		/* for 601 */
792 	case 0x40000:
793 	case 0x140000:	/* 7450 MSS error and TEA */
794 		pr_cont("Transfer error ack signal\n");
795 		break;
796 	case 0x20000:
797 		pr_cont("Data parity error signal\n");
798 		break;
799 	case 0x10000:
800 		pr_cont("Address parity error signal\n");
801 		break;
802 	case 0x20000000:
803 		pr_cont("L1 Data Cache error\n");
804 		break;
805 	case 0x40000000:
806 		pr_cont("L1 Instruction Cache error\n");
807 		break;
808 	case 0x00100000:
809 		pr_cont("L2 data cache parity error\n");
810 		break;
811 	default:
812 		pr_cont("Unknown values in msr\n");
813 	}
814 	return 0;
815 }
816 #endif /* everything else */
817 
818 void machine_check_exception(struct pt_regs *regs)
819 {
820 	int recover = 0;
821 	bool nested = in_nmi();
822 	if (!nested)
823 		nmi_enter();
824 
825 	__this_cpu_inc(irq_stat.mce_exceptions);
826 
827 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
828 
829 	/* See if any machine dependent calls. In theory, we would want
830 	 * to call the CPU first, and call the ppc_md. one if the CPU
831 	 * one returns a positive number. However there is existing code
832 	 * that assumes the board gets a first chance, so let's keep it
833 	 * that way for now and fix things later. --BenH.
834 	 */
835 	if (ppc_md.machine_check_exception)
836 		recover = ppc_md.machine_check_exception(regs);
837 	else if (cur_cpu_spec->machine_check)
838 		recover = cur_cpu_spec->machine_check(regs);
839 
840 	if (recover > 0)
841 		goto bail;
842 
843 	if (debugger_fault_handler(regs))
844 		goto bail;
845 
846 	if (check_io_access(regs))
847 		goto bail;
848 
849 	if (!nested)
850 		nmi_exit();
851 
852 	die("Machine check", regs, SIGBUS);
853 
854 	/* Must die if the interrupt is not recoverable */
855 	if (!(regs->msr & MSR_RI))
856 		nmi_panic(regs, "Unrecoverable Machine check");
857 
858 	return;
859 
860 bail:
861 	if (!nested)
862 		nmi_exit();
863 }
864 
865 void SMIException(struct pt_regs *regs)
866 {
867 	die("System Management Interrupt", regs, SIGABRT);
868 }
869 
870 #ifdef CONFIG_VSX
871 static void p9_hmi_special_emu(struct pt_regs *regs)
872 {
873 	unsigned int ra, rb, t, i, sel, instr, rc;
874 	const void __user *addr;
875 	u8 vbuf[16], *vdst;
876 	unsigned long ea, msr, msr_mask;
877 	bool swap;
878 
879 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
880 		return;
881 
882 	/*
883 	 * lxvb16x	opcode: 0x7c0006d8
884 	 * lxvd2x	opcode: 0x7c000698
885 	 * lxvh8x	opcode: 0x7c000658
886 	 * lxvw4x	opcode: 0x7c000618
887 	 */
888 	if ((instr & 0xfc00073e) != 0x7c000618) {
889 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
890 			 " instr=%08x\n",
891 			 smp_processor_id(), current->comm, current->pid,
892 			 regs->nip, instr);
893 		return;
894 	}
895 
896 	/* Grab vector registers into the task struct */
897 	msr = regs->msr; /* Grab msr before we flush the bits */
898 	flush_vsx_to_thread(current);
899 	enable_kernel_altivec();
900 
901 	/*
902 	 * Is userspace running with a different endian (this is rare but
903 	 * not impossible)
904 	 */
905 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
906 
907 	/* Decode the instruction */
908 	ra = (instr >> 16) & 0x1f;
909 	rb = (instr >> 11) & 0x1f;
910 	t = (instr >> 21) & 0x1f;
911 	if (instr & 1)
912 		vdst = (u8 *)&current->thread.vr_state.vr[t];
913 	else
914 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
915 
916 	/* Grab the vector address */
917 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
918 	if (is_32bit_task())
919 		ea &= 0xfffffffful;
920 	addr = (__force const void __user *)ea;
921 
922 	/* Check it */
923 	if (!access_ok(addr, 16)) {
924 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
925 			 " instr=%08x addr=%016lx\n",
926 			 smp_processor_id(), current->comm, current->pid,
927 			 regs->nip, instr, (unsigned long)addr);
928 		return;
929 	}
930 
931 	/* Read the vector */
932 	rc = 0;
933 	if ((unsigned long)addr & 0xfUL)
934 		/* unaligned case */
935 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
936 	else
937 		__get_user_atomic_128_aligned(vbuf, addr, rc);
938 	if (rc) {
939 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
940 			 " instr=%08x addr=%016lx\n",
941 			 smp_processor_id(), current->comm, current->pid,
942 			 regs->nip, instr, (unsigned long)addr);
943 		return;
944 	}
945 
946 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
947 		 " instr=%08x addr=%016lx\n",
948 		 smp_processor_id(), current->comm, current->pid, regs->nip,
949 		 instr, (unsigned long) addr);
950 
951 	/* Grab instruction "selector" */
952 	sel = (instr >> 6) & 3;
953 
954 	/*
955 	 * Check to make sure the facility is actually enabled. This
956 	 * could happen if we get a false positive hit.
957 	 *
958 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
959 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
960 	 */
961 	msr_mask = MSR_VSX;
962 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
963 		msr_mask = MSR_VEC;
964 	if (!(msr & msr_mask)) {
965 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
966 			 " instr=%08x msr:%016lx\n",
967 			 smp_processor_id(), current->comm, current->pid,
968 			 regs->nip, instr, msr);
969 		return;
970 	}
971 
972 	/* Do logging here before we modify sel based on endian */
973 	switch (sel) {
974 	case 0:	/* lxvw4x */
975 		PPC_WARN_EMULATED(lxvw4x, regs);
976 		break;
977 	case 1: /* lxvh8x */
978 		PPC_WARN_EMULATED(lxvh8x, regs);
979 		break;
980 	case 2: /* lxvd2x */
981 		PPC_WARN_EMULATED(lxvd2x, regs);
982 		break;
983 	case 3: /* lxvb16x */
984 		PPC_WARN_EMULATED(lxvb16x, regs);
985 		break;
986 	}
987 
988 #ifdef __LITTLE_ENDIAN__
989 	/*
990 	 * An LE kernel stores the vector in the task struct as an LE
991 	 * byte array (effectively swapping both the components and
992 	 * the content of the components). Those instructions expect
993 	 * the components to remain in ascending address order, so we
994 	 * swap them back.
995 	 *
996 	 * If we are running a BE user space, the expectation is that
997 	 * of a simple memcpy, so forcing the emulation to look like
998 	 * a lxvb16x should do the trick.
999 	 */
1000 	if (swap)
1001 		sel = 3;
1002 
1003 	switch (sel) {
1004 	case 0:	/* lxvw4x */
1005 		for (i = 0; i < 4; i++)
1006 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1007 		break;
1008 	case 1: /* lxvh8x */
1009 		for (i = 0; i < 8; i++)
1010 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1011 		break;
1012 	case 2: /* lxvd2x */
1013 		for (i = 0; i < 2; i++)
1014 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1015 		break;
1016 	case 3: /* lxvb16x */
1017 		for (i = 0; i < 16; i++)
1018 			vdst[i] = vbuf[15-i];
1019 		break;
1020 	}
1021 #else /* __LITTLE_ENDIAN__ */
1022 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1023 	if (!swap)
1024 		sel = 3;
1025 
1026 	/* Otherwise, we need to swap the content of the components */
1027 	switch (sel) {
1028 	case 0:	/* lxvw4x */
1029 		for (i = 0; i < 4; i++)
1030 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1031 		break;
1032 	case 1: /* lxvh8x */
1033 		for (i = 0; i < 8; i++)
1034 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1035 		break;
1036 	case 2: /* lxvd2x */
1037 		for (i = 0; i < 2; i++)
1038 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1039 		break;
1040 	case 3: /* lxvb16x */
1041 		memcpy(vdst, vbuf, 16);
1042 		break;
1043 	}
1044 #endif /* !__LITTLE_ENDIAN__ */
1045 
1046 	/* Go to next instruction */
1047 	regs->nip += 4;
1048 }
1049 #endif /* CONFIG_VSX */
1050 
1051 void handle_hmi_exception(struct pt_regs *regs)
1052 {
1053 	struct pt_regs *old_regs;
1054 
1055 	old_regs = set_irq_regs(regs);
1056 	irq_enter();
1057 
1058 #ifdef CONFIG_VSX
1059 	/* Real mode flagged P9 special emu is needed */
1060 	if (local_paca->hmi_p9_special_emu) {
1061 		local_paca->hmi_p9_special_emu = 0;
1062 
1063 		/*
1064 		 * We don't want to take page faults while doing the
1065 		 * emulation, we just replay the instruction if necessary.
1066 		 */
1067 		pagefault_disable();
1068 		p9_hmi_special_emu(regs);
1069 		pagefault_enable();
1070 	}
1071 #endif /* CONFIG_VSX */
1072 
1073 	if (ppc_md.handle_hmi_exception)
1074 		ppc_md.handle_hmi_exception(regs);
1075 
1076 	irq_exit();
1077 	set_irq_regs(old_regs);
1078 }
1079 
1080 void unknown_exception(struct pt_regs *regs)
1081 {
1082 	enum ctx_state prev_state = exception_enter();
1083 
1084 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1085 	       regs->nip, regs->msr, regs->trap);
1086 
1087 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1088 
1089 	exception_exit(prev_state);
1090 }
1091 
1092 void instruction_breakpoint_exception(struct pt_regs *regs)
1093 {
1094 	enum ctx_state prev_state = exception_enter();
1095 
1096 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1097 					5, SIGTRAP) == NOTIFY_STOP)
1098 		goto bail;
1099 	if (debugger_iabr_match(regs))
1100 		goto bail;
1101 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1102 
1103 bail:
1104 	exception_exit(prev_state);
1105 }
1106 
1107 void RunModeException(struct pt_regs *regs)
1108 {
1109 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1110 }
1111 
1112 void single_step_exception(struct pt_regs *regs)
1113 {
1114 	enum ctx_state prev_state = exception_enter();
1115 
1116 	clear_single_step(regs);
1117 	clear_br_trace(regs);
1118 
1119 	if (kprobe_post_handler(regs))
1120 		return;
1121 
1122 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1123 					5, SIGTRAP) == NOTIFY_STOP)
1124 		goto bail;
1125 	if (debugger_sstep(regs))
1126 		goto bail;
1127 
1128 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1129 
1130 bail:
1131 	exception_exit(prev_state);
1132 }
1133 NOKPROBE_SYMBOL(single_step_exception);
1134 
1135 /*
1136  * After we have successfully emulated an instruction, we have to
1137  * check if the instruction was being single-stepped, and if so,
1138  * pretend we got a single-step exception.  This was pointed out
1139  * by Kumar Gala.  -- paulus
1140  */
1141 static void emulate_single_step(struct pt_regs *regs)
1142 {
1143 	if (single_stepping(regs))
1144 		single_step_exception(regs);
1145 }
1146 
1147 static inline int __parse_fpscr(unsigned long fpscr)
1148 {
1149 	int ret = FPE_FLTUNK;
1150 
1151 	/* Invalid operation */
1152 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1153 		ret = FPE_FLTINV;
1154 
1155 	/* Overflow */
1156 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1157 		ret = FPE_FLTOVF;
1158 
1159 	/* Underflow */
1160 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1161 		ret = FPE_FLTUND;
1162 
1163 	/* Divide by zero */
1164 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1165 		ret = FPE_FLTDIV;
1166 
1167 	/* Inexact result */
1168 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1169 		ret = FPE_FLTRES;
1170 
1171 	return ret;
1172 }
1173 
1174 static void parse_fpe(struct pt_regs *regs)
1175 {
1176 	int code = 0;
1177 
1178 	flush_fp_to_thread(current);
1179 
1180 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1181 
1182 	_exception(SIGFPE, regs, code, regs->nip);
1183 }
1184 
1185 /*
1186  * Illegal instruction emulation support.  Originally written to
1187  * provide the PVR to user applications using the mfspr rd, PVR.
1188  * Return non-zero if we can't emulate, or -EFAULT if the associated
1189  * memory access caused an access fault.  Return zero on success.
1190  *
1191  * There are a couple of ways to do this, either "decode" the instruction
1192  * or directly match lots of bits.  In this case, matching lots of
1193  * bits is faster and easier.
1194  *
1195  */
1196 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1197 {
1198 	u8 rT = (instword >> 21) & 0x1f;
1199 	u8 rA = (instword >> 16) & 0x1f;
1200 	u8 NB_RB = (instword >> 11) & 0x1f;
1201 	u32 num_bytes;
1202 	unsigned long EA;
1203 	int pos = 0;
1204 
1205 	/* Early out if we are an invalid form of lswx */
1206 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1207 		if ((rT == rA) || (rT == NB_RB))
1208 			return -EINVAL;
1209 
1210 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1211 
1212 	switch (instword & PPC_INST_STRING_MASK) {
1213 		case PPC_INST_LSWX:
1214 		case PPC_INST_STSWX:
1215 			EA += NB_RB;
1216 			num_bytes = regs->xer & 0x7f;
1217 			break;
1218 		case PPC_INST_LSWI:
1219 		case PPC_INST_STSWI:
1220 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1221 			break;
1222 		default:
1223 			return -EINVAL;
1224 	}
1225 
1226 	while (num_bytes != 0)
1227 	{
1228 		u8 val;
1229 		u32 shift = 8 * (3 - (pos & 0x3));
1230 
1231 		/* if process is 32-bit, clear upper 32 bits of EA */
1232 		if ((regs->msr & MSR_64BIT) == 0)
1233 			EA &= 0xFFFFFFFF;
1234 
1235 		switch ((instword & PPC_INST_STRING_MASK)) {
1236 			case PPC_INST_LSWX:
1237 			case PPC_INST_LSWI:
1238 				if (get_user(val, (u8 __user *)EA))
1239 					return -EFAULT;
1240 				/* first time updating this reg,
1241 				 * zero it out */
1242 				if (pos == 0)
1243 					regs->gpr[rT] = 0;
1244 				regs->gpr[rT] |= val << shift;
1245 				break;
1246 			case PPC_INST_STSWI:
1247 			case PPC_INST_STSWX:
1248 				val = regs->gpr[rT] >> shift;
1249 				if (put_user(val, (u8 __user *)EA))
1250 					return -EFAULT;
1251 				break;
1252 		}
1253 		/* move EA to next address */
1254 		EA += 1;
1255 		num_bytes--;
1256 
1257 		/* manage our position within the register */
1258 		if (++pos == 4) {
1259 			pos = 0;
1260 			if (++rT == 32)
1261 				rT = 0;
1262 		}
1263 	}
1264 
1265 	return 0;
1266 }
1267 
1268 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1269 {
1270 	u32 ra,rs;
1271 	unsigned long tmp;
1272 
1273 	ra = (instword >> 16) & 0x1f;
1274 	rs = (instword >> 21) & 0x1f;
1275 
1276 	tmp = regs->gpr[rs];
1277 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1278 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1279 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1280 	regs->gpr[ra] = tmp;
1281 
1282 	return 0;
1283 }
1284 
1285 static int emulate_isel(struct pt_regs *regs, u32 instword)
1286 {
1287 	u8 rT = (instword >> 21) & 0x1f;
1288 	u8 rA = (instword >> 16) & 0x1f;
1289 	u8 rB = (instword >> 11) & 0x1f;
1290 	u8 BC = (instword >> 6) & 0x1f;
1291 	u8 bit;
1292 	unsigned long tmp;
1293 
1294 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1295 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1296 
1297 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1298 
1299 	return 0;
1300 }
1301 
1302 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1303 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1304 {
1305         /* If we're emulating a load/store in an active transaction, we cannot
1306          * emulate it as the kernel operates in transaction suspended context.
1307          * We need to abort the transaction.  This creates a persistent TM
1308          * abort so tell the user what caused it with a new code.
1309 	 */
1310 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1311 		tm_enable();
1312 		tm_abort(cause);
1313 		return true;
1314 	}
1315 	return false;
1316 }
1317 #else
1318 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1319 {
1320 	return false;
1321 }
1322 #endif
1323 
1324 static int emulate_instruction(struct pt_regs *regs)
1325 {
1326 	u32 instword;
1327 	u32 rd;
1328 
1329 	if (!user_mode(regs))
1330 		return -EINVAL;
1331 	CHECK_FULL_REGS(regs);
1332 
1333 	if (get_user(instword, (u32 __user *)(regs->nip)))
1334 		return -EFAULT;
1335 
1336 	/* Emulate the mfspr rD, PVR. */
1337 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1338 		PPC_WARN_EMULATED(mfpvr, regs);
1339 		rd = (instword >> 21) & 0x1f;
1340 		regs->gpr[rd] = mfspr(SPRN_PVR);
1341 		return 0;
1342 	}
1343 
1344 	/* Emulating the dcba insn is just a no-op.  */
1345 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1346 		PPC_WARN_EMULATED(dcba, regs);
1347 		return 0;
1348 	}
1349 
1350 	/* Emulate the mcrxr insn.  */
1351 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1352 		int shift = (instword >> 21) & 0x1c;
1353 		unsigned long msk = 0xf0000000UL >> shift;
1354 
1355 		PPC_WARN_EMULATED(mcrxr, regs);
1356 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1357 		regs->xer &= ~0xf0000000UL;
1358 		return 0;
1359 	}
1360 
1361 	/* Emulate load/store string insn. */
1362 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1363 		if (tm_abort_check(regs,
1364 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1365 			return -EINVAL;
1366 		PPC_WARN_EMULATED(string, regs);
1367 		return emulate_string_inst(regs, instword);
1368 	}
1369 
1370 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1371 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1372 		PPC_WARN_EMULATED(popcntb, regs);
1373 		return emulate_popcntb_inst(regs, instword);
1374 	}
1375 
1376 	/* Emulate isel (Integer Select) instruction */
1377 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1378 		PPC_WARN_EMULATED(isel, regs);
1379 		return emulate_isel(regs, instword);
1380 	}
1381 
1382 	/* Emulate sync instruction variants */
1383 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1384 		PPC_WARN_EMULATED(sync, regs);
1385 		asm volatile("sync");
1386 		return 0;
1387 	}
1388 
1389 #ifdef CONFIG_PPC64
1390 	/* Emulate the mfspr rD, DSCR. */
1391 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1392 		PPC_INST_MFSPR_DSCR_USER) ||
1393 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1394 		PPC_INST_MFSPR_DSCR)) &&
1395 			cpu_has_feature(CPU_FTR_DSCR)) {
1396 		PPC_WARN_EMULATED(mfdscr, regs);
1397 		rd = (instword >> 21) & 0x1f;
1398 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1399 		return 0;
1400 	}
1401 	/* Emulate the mtspr DSCR, rD. */
1402 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1403 		PPC_INST_MTSPR_DSCR_USER) ||
1404 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1405 		PPC_INST_MTSPR_DSCR)) &&
1406 			cpu_has_feature(CPU_FTR_DSCR)) {
1407 		PPC_WARN_EMULATED(mtdscr, regs);
1408 		rd = (instword >> 21) & 0x1f;
1409 		current->thread.dscr = regs->gpr[rd];
1410 		current->thread.dscr_inherit = 1;
1411 		mtspr(SPRN_DSCR, current->thread.dscr);
1412 		return 0;
1413 	}
1414 #endif
1415 
1416 	return -EINVAL;
1417 }
1418 
1419 int is_valid_bugaddr(unsigned long addr)
1420 {
1421 	return is_kernel_addr(addr);
1422 }
1423 
1424 #ifdef CONFIG_MATH_EMULATION
1425 static int emulate_math(struct pt_regs *regs)
1426 {
1427 	int ret;
1428 	extern int do_mathemu(struct pt_regs *regs);
1429 
1430 	ret = do_mathemu(regs);
1431 	if (ret >= 0)
1432 		PPC_WARN_EMULATED(math, regs);
1433 
1434 	switch (ret) {
1435 	case 0:
1436 		emulate_single_step(regs);
1437 		return 0;
1438 	case 1: {
1439 			int code = 0;
1440 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1441 			_exception(SIGFPE, regs, code, regs->nip);
1442 			return 0;
1443 		}
1444 	case -EFAULT:
1445 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1446 		return 0;
1447 	}
1448 
1449 	return -1;
1450 }
1451 #else
1452 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1453 #endif
1454 
1455 void program_check_exception(struct pt_regs *regs)
1456 {
1457 	enum ctx_state prev_state = exception_enter();
1458 	unsigned int reason = get_reason(regs);
1459 
1460 	/* We can now get here via a FP Unavailable exception if the core
1461 	 * has no FPU, in that case the reason flags will be 0 */
1462 
1463 	if (reason & REASON_FP) {
1464 		/* IEEE FP exception */
1465 		parse_fpe(regs);
1466 		goto bail;
1467 	}
1468 	if (reason & REASON_TRAP) {
1469 		unsigned long bugaddr;
1470 		/* Debugger is first in line to stop recursive faults in
1471 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1472 		if (debugger_bpt(regs))
1473 			goto bail;
1474 
1475 		if (kprobe_handler(regs))
1476 			goto bail;
1477 
1478 		/* trap exception */
1479 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1480 				== NOTIFY_STOP)
1481 			goto bail;
1482 
1483 		bugaddr = regs->nip;
1484 		/*
1485 		 * Fixup bugaddr for BUG_ON() in real mode
1486 		 */
1487 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1488 			bugaddr += PAGE_OFFSET;
1489 
1490 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1491 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1492 			regs->nip += 4;
1493 			goto bail;
1494 		}
1495 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1496 		goto bail;
1497 	}
1498 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1499 	if (reason & REASON_TM) {
1500 		/* This is a TM "Bad Thing Exception" program check.
1501 		 * This occurs when:
1502 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1503 		 *    transition in TM states.
1504 		 * -  A trechkpt is attempted when transactional.
1505 		 * -  A treclaim is attempted when non transactional.
1506 		 * -  A tend is illegally attempted.
1507 		 * -  writing a TM SPR when transactional.
1508 		 *
1509 		 * If usermode caused this, it's done something illegal and
1510 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1511 		 * operand to distinguish from the instruction just being bad
1512 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1513 		 * illegal /placement/ of a valid instruction.
1514 		 */
1515 		if (user_mode(regs)) {
1516 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1517 			goto bail;
1518 		} else {
1519 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1520 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1521 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1522 			die("Unrecoverable exception", regs, SIGABRT);
1523 		}
1524 	}
1525 #endif
1526 
1527 	/*
1528 	 * If we took the program check in the kernel skip down to sending a
1529 	 * SIGILL. The subsequent cases all relate to emulating instructions
1530 	 * which we should only do for userspace. We also do not want to enable
1531 	 * interrupts for kernel faults because that might lead to further
1532 	 * faults, and loose the context of the original exception.
1533 	 */
1534 	if (!user_mode(regs))
1535 		goto sigill;
1536 
1537 	/* We restore the interrupt state now */
1538 	if (!arch_irq_disabled_regs(regs))
1539 		local_irq_enable();
1540 
1541 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1542 	 * but there seems to be a hardware bug on the 405GP (RevD)
1543 	 * that means ESR is sometimes set incorrectly - either to
1544 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1545 	 * hardware people - not sure if it can happen on any illegal
1546 	 * instruction or only on FP instructions, whether there is a
1547 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1548 	 */
1549 	if (!emulate_math(regs))
1550 		goto bail;
1551 
1552 	/* Try to emulate it if we should. */
1553 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1554 		switch (emulate_instruction(regs)) {
1555 		case 0:
1556 			regs->nip += 4;
1557 			emulate_single_step(regs);
1558 			goto bail;
1559 		case -EFAULT:
1560 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1561 			goto bail;
1562 		}
1563 	}
1564 
1565 sigill:
1566 	if (reason & REASON_PRIVILEGED)
1567 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1568 	else
1569 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1570 
1571 bail:
1572 	exception_exit(prev_state);
1573 }
1574 NOKPROBE_SYMBOL(program_check_exception);
1575 
1576 /*
1577  * This occurs when running in hypervisor mode on POWER6 or later
1578  * and an illegal instruction is encountered.
1579  */
1580 void emulation_assist_interrupt(struct pt_regs *regs)
1581 {
1582 	regs->msr |= REASON_ILLEGAL;
1583 	program_check_exception(regs);
1584 }
1585 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1586 
1587 void alignment_exception(struct pt_regs *regs)
1588 {
1589 	enum ctx_state prev_state = exception_enter();
1590 	int sig, code, fixed = 0;
1591 
1592 	/* We restore the interrupt state now */
1593 	if (!arch_irq_disabled_regs(regs))
1594 		local_irq_enable();
1595 
1596 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1597 		goto bail;
1598 
1599 	/* we don't implement logging of alignment exceptions */
1600 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1601 		fixed = fix_alignment(regs);
1602 
1603 	if (fixed == 1) {
1604 		regs->nip += 4;	/* skip over emulated instruction */
1605 		emulate_single_step(regs);
1606 		goto bail;
1607 	}
1608 
1609 	/* Operand address was bad */
1610 	if (fixed == -EFAULT) {
1611 		sig = SIGSEGV;
1612 		code = SEGV_ACCERR;
1613 	} else {
1614 		sig = SIGBUS;
1615 		code = BUS_ADRALN;
1616 	}
1617 	if (user_mode(regs))
1618 		_exception(sig, regs, code, regs->dar);
1619 	else
1620 		bad_page_fault(regs, regs->dar, sig);
1621 
1622 bail:
1623 	exception_exit(prev_state);
1624 }
1625 
1626 void StackOverflow(struct pt_regs *regs)
1627 {
1628 	pr_crit("Kernel stack overflow in process %s[%d], r1=%lx\n",
1629 		current->comm, task_pid_nr(current), regs->gpr[1]);
1630 	debugger(regs);
1631 	show_regs(regs);
1632 	panic("kernel stack overflow");
1633 }
1634 
1635 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1636 {
1637 	enum ctx_state prev_state = exception_enter();
1638 
1639 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1640 			  "%lx at %lx\n", regs->trap, regs->nip);
1641 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1642 
1643 	exception_exit(prev_state);
1644 }
1645 
1646 void altivec_unavailable_exception(struct pt_regs *regs)
1647 {
1648 	enum ctx_state prev_state = exception_enter();
1649 
1650 	if (user_mode(regs)) {
1651 		/* A user program has executed an altivec instruction,
1652 		   but this kernel doesn't support altivec. */
1653 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1654 		goto bail;
1655 	}
1656 
1657 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1658 			"%lx at %lx\n", regs->trap, regs->nip);
1659 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1660 
1661 bail:
1662 	exception_exit(prev_state);
1663 }
1664 
1665 void vsx_unavailable_exception(struct pt_regs *regs)
1666 {
1667 	if (user_mode(regs)) {
1668 		/* A user program has executed an vsx instruction,
1669 		   but this kernel doesn't support vsx. */
1670 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1671 		return;
1672 	}
1673 
1674 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1675 			"%lx at %lx\n", regs->trap, regs->nip);
1676 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1677 }
1678 
1679 #ifdef CONFIG_PPC64
1680 static void tm_unavailable(struct pt_regs *regs)
1681 {
1682 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1683 	if (user_mode(regs)) {
1684 		current->thread.load_tm++;
1685 		regs->msr |= MSR_TM;
1686 		tm_enable();
1687 		tm_restore_sprs(&current->thread);
1688 		return;
1689 	}
1690 #endif
1691 	pr_emerg("Unrecoverable TM Unavailable Exception "
1692 			"%lx at %lx\n", regs->trap, regs->nip);
1693 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1694 }
1695 
1696 void facility_unavailable_exception(struct pt_regs *regs)
1697 {
1698 	static char *facility_strings[] = {
1699 		[FSCR_FP_LG] = "FPU",
1700 		[FSCR_VECVSX_LG] = "VMX/VSX",
1701 		[FSCR_DSCR_LG] = "DSCR",
1702 		[FSCR_PM_LG] = "PMU SPRs",
1703 		[FSCR_BHRB_LG] = "BHRB",
1704 		[FSCR_TM_LG] = "TM",
1705 		[FSCR_EBB_LG] = "EBB",
1706 		[FSCR_TAR_LG] = "TAR",
1707 		[FSCR_MSGP_LG] = "MSGP",
1708 		[FSCR_SCV_LG] = "SCV",
1709 	};
1710 	char *facility = "unknown";
1711 	u64 value;
1712 	u32 instword, rd;
1713 	u8 status;
1714 	bool hv;
1715 
1716 	hv = (TRAP(regs) == 0xf80);
1717 	if (hv)
1718 		value = mfspr(SPRN_HFSCR);
1719 	else
1720 		value = mfspr(SPRN_FSCR);
1721 
1722 	status = value >> 56;
1723 	if ((hv || status >= 2) &&
1724 	    (status < ARRAY_SIZE(facility_strings)) &&
1725 	    facility_strings[status])
1726 		facility = facility_strings[status];
1727 
1728 	/* We should not have taken this interrupt in kernel */
1729 	if (!user_mode(regs)) {
1730 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1731 			 facility, status, regs->nip);
1732 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1733 	}
1734 
1735 	/* We restore the interrupt state now */
1736 	if (!arch_irq_disabled_regs(regs))
1737 		local_irq_enable();
1738 
1739 	if (status == FSCR_DSCR_LG) {
1740 		/*
1741 		 * User is accessing the DSCR register using the problem
1742 		 * state only SPR number (0x03) either through a mfspr or
1743 		 * a mtspr instruction. If it is a write attempt through
1744 		 * a mtspr, then we set the inherit bit. This also allows
1745 		 * the user to write or read the register directly in the
1746 		 * future by setting via the FSCR DSCR bit. But in case it
1747 		 * is a read DSCR attempt through a mfspr instruction, we
1748 		 * just emulate the instruction instead. This code path will
1749 		 * always emulate all the mfspr instructions till the user
1750 		 * has attempted at least one mtspr instruction. This way it
1751 		 * preserves the same behaviour when the user is accessing
1752 		 * the DSCR through privilege level only SPR number (0x11)
1753 		 * which is emulated through illegal instruction exception.
1754 		 * We always leave HFSCR DSCR set.
1755 		 */
1756 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1757 			pr_err("Failed to fetch the user instruction\n");
1758 			return;
1759 		}
1760 
1761 		/* Write into DSCR (mtspr 0x03, RS) */
1762 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1763 				== PPC_INST_MTSPR_DSCR_USER) {
1764 			rd = (instword >> 21) & 0x1f;
1765 			current->thread.dscr = regs->gpr[rd];
1766 			current->thread.dscr_inherit = 1;
1767 			current->thread.fscr |= FSCR_DSCR;
1768 			mtspr(SPRN_FSCR, current->thread.fscr);
1769 		}
1770 
1771 		/* Read from DSCR (mfspr RT, 0x03) */
1772 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1773 				== PPC_INST_MFSPR_DSCR_USER) {
1774 			if (emulate_instruction(regs)) {
1775 				pr_err("DSCR based mfspr emulation failed\n");
1776 				return;
1777 			}
1778 			regs->nip += 4;
1779 			emulate_single_step(regs);
1780 		}
1781 		return;
1782 	}
1783 
1784 	if (status == FSCR_TM_LG) {
1785 		/*
1786 		 * If we're here then the hardware is TM aware because it
1787 		 * generated an exception with FSRM_TM set.
1788 		 *
1789 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1790 		 * told us not to do TM, or the kernel is not built with TM
1791 		 * support.
1792 		 *
1793 		 * If both of those things are true, then userspace can spam the
1794 		 * console by triggering the printk() below just by continually
1795 		 * doing tbegin (or any TM instruction). So in that case just
1796 		 * send the process a SIGILL immediately.
1797 		 */
1798 		if (!cpu_has_feature(CPU_FTR_TM))
1799 			goto out;
1800 
1801 		tm_unavailable(regs);
1802 		return;
1803 	}
1804 
1805 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1806 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1807 
1808 out:
1809 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1810 }
1811 #endif
1812 
1813 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1814 
1815 void fp_unavailable_tm(struct pt_regs *regs)
1816 {
1817 	/* Note:  This does not handle any kind of FP laziness. */
1818 
1819 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1820 		 regs->nip, regs->msr);
1821 
1822         /* We can only have got here if the task started using FP after
1823          * beginning the transaction.  So, the transactional regs are just a
1824          * copy of the checkpointed ones.  But, we still need to recheckpoint
1825          * as we're enabling FP for the process; it will return, abort the
1826          * transaction, and probably retry but now with FP enabled.  So the
1827          * checkpointed FP registers need to be loaded.
1828 	 */
1829 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1830 
1831 	/*
1832 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1833 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1834 	 *
1835 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1836 	 * recheckpoint.
1837 	 */
1838 
1839 	/* Enable FP for the task: */
1840 	current->thread.load_fp = 1;
1841 
1842 	/*
1843 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1844 	 */
1845 	tm_recheckpoint(&current->thread);
1846 }
1847 
1848 void altivec_unavailable_tm(struct pt_regs *regs)
1849 {
1850 	/* See the comments in fp_unavailable_tm().  This function operates
1851 	 * the same way.
1852 	 */
1853 
1854 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1855 		 "MSR=%lx\n",
1856 		 regs->nip, regs->msr);
1857 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1858 	current->thread.load_vec = 1;
1859 	tm_recheckpoint(&current->thread);
1860 	current->thread.used_vr = 1;
1861 }
1862 
1863 void vsx_unavailable_tm(struct pt_regs *regs)
1864 {
1865 	/* See the comments in fp_unavailable_tm().  This works similarly,
1866 	 * though we're loading both FP and VEC registers in here.
1867 	 *
1868 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1869 	 * regs.  Either way, set MSR_VSX.
1870 	 */
1871 
1872 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1873 		 "MSR=%lx\n",
1874 		 regs->nip, regs->msr);
1875 
1876 	current->thread.used_vsr = 1;
1877 
1878 	/* This reclaims FP and/or VR regs if they're already enabled */
1879 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1880 
1881 	current->thread.load_vec = 1;
1882 	current->thread.load_fp = 1;
1883 
1884 	tm_recheckpoint(&current->thread);
1885 }
1886 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1887 
1888 void performance_monitor_exception(struct pt_regs *regs)
1889 {
1890 	__this_cpu_inc(irq_stat.pmu_irqs);
1891 
1892 	perf_irq(regs);
1893 }
1894 
1895 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1896 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1897 {
1898 	int changed = 0;
1899 	/*
1900 	 * Determine the cause of the debug event, clear the
1901 	 * event flags and send a trap to the handler. Torez
1902 	 */
1903 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1904 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1905 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1906 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1907 #endif
1908 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1909 			     5);
1910 		changed |= 0x01;
1911 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1912 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1913 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1914 			     6);
1915 		changed |= 0x01;
1916 	}  else if (debug_status & DBSR_IAC1) {
1917 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1918 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1919 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1920 			     1);
1921 		changed |= 0x01;
1922 	}  else if (debug_status & DBSR_IAC2) {
1923 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1924 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1925 			     2);
1926 		changed |= 0x01;
1927 	}  else if (debug_status & DBSR_IAC3) {
1928 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1929 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1930 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1931 			     3);
1932 		changed |= 0x01;
1933 	}  else if (debug_status & DBSR_IAC4) {
1934 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1935 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1936 			     4);
1937 		changed |= 0x01;
1938 	}
1939 	/*
1940 	 * At the point this routine was called, the MSR(DE) was turned off.
1941 	 * Check all other debug flags and see if that bit needs to be turned
1942 	 * back on or not.
1943 	 */
1944 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1945 			       current->thread.debug.dbcr1))
1946 		regs->msr |= MSR_DE;
1947 	else
1948 		/* Make sure the IDM flag is off */
1949 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1950 
1951 	if (changed & 0x01)
1952 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1953 }
1954 
1955 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1956 {
1957 	current->thread.debug.dbsr = debug_status;
1958 
1959 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1960 	 * on server, it stops on the target of the branch. In order to simulate
1961 	 * the server behaviour, we thus restart right away with a single step
1962 	 * instead of stopping here when hitting a BT
1963 	 */
1964 	if (debug_status & DBSR_BT) {
1965 		regs->msr &= ~MSR_DE;
1966 
1967 		/* Disable BT */
1968 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1969 		/* Clear the BT event */
1970 		mtspr(SPRN_DBSR, DBSR_BT);
1971 
1972 		/* Do the single step trick only when coming from userspace */
1973 		if (user_mode(regs)) {
1974 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1975 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1976 			regs->msr |= MSR_DE;
1977 			return;
1978 		}
1979 
1980 		if (kprobe_post_handler(regs))
1981 			return;
1982 
1983 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1984 			       5, SIGTRAP) == NOTIFY_STOP) {
1985 			return;
1986 		}
1987 		if (debugger_sstep(regs))
1988 			return;
1989 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1990 		regs->msr &= ~MSR_DE;
1991 
1992 		/* Disable instruction completion */
1993 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
1994 		/* Clear the instruction completion event */
1995 		mtspr(SPRN_DBSR, DBSR_IC);
1996 
1997 		if (kprobe_post_handler(regs))
1998 			return;
1999 
2000 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2001 			       5, SIGTRAP) == NOTIFY_STOP) {
2002 			return;
2003 		}
2004 
2005 		if (debugger_sstep(regs))
2006 			return;
2007 
2008 		if (user_mode(regs)) {
2009 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2010 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2011 					       current->thread.debug.dbcr1))
2012 				regs->msr |= MSR_DE;
2013 			else
2014 				/* Make sure the IDM bit is off */
2015 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2016 		}
2017 
2018 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2019 	} else
2020 		handle_debug(regs, debug_status);
2021 }
2022 NOKPROBE_SYMBOL(DebugException);
2023 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2024 
2025 #if !defined(CONFIG_TAU_INT)
2026 void TAUException(struct pt_regs *regs)
2027 {
2028 	printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
2029 	       regs->nip, regs->msr, regs->trap, print_tainted());
2030 }
2031 #endif /* CONFIG_INT_TAU */
2032 
2033 #ifdef CONFIG_ALTIVEC
2034 void altivec_assist_exception(struct pt_regs *regs)
2035 {
2036 	int err;
2037 
2038 	if (!user_mode(regs)) {
2039 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2040 		       " at %lx\n", regs->nip);
2041 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2042 	}
2043 
2044 	flush_altivec_to_thread(current);
2045 
2046 	PPC_WARN_EMULATED(altivec, regs);
2047 	err = emulate_altivec(regs);
2048 	if (err == 0) {
2049 		regs->nip += 4;		/* skip emulated instruction */
2050 		emulate_single_step(regs);
2051 		return;
2052 	}
2053 
2054 	if (err == -EFAULT) {
2055 		/* got an error reading the instruction */
2056 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2057 	} else {
2058 		/* didn't recognize the instruction */
2059 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2060 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2061 				   "in %s at %lx\n", current->comm, regs->nip);
2062 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2063 	}
2064 }
2065 #endif /* CONFIG_ALTIVEC */
2066 
2067 #ifdef CONFIG_FSL_BOOKE
2068 void CacheLockingException(struct pt_regs *regs, unsigned long address,
2069 			   unsigned long error_code)
2070 {
2071 	/* We treat cache locking instructions from the user
2072 	 * as priv ops, in the future we could try to do
2073 	 * something smarter
2074 	 */
2075 	if (error_code & (ESR_DLK|ESR_ILK))
2076 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2077 	return;
2078 }
2079 #endif /* CONFIG_FSL_BOOKE */
2080 
2081 #ifdef CONFIG_SPE
2082 void SPEFloatingPointException(struct pt_regs *regs)
2083 {
2084 	extern int do_spe_mathemu(struct pt_regs *regs);
2085 	unsigned long spefscr;
2086 	int fpexc_mode;
2087 	int code = FPE_FLTUNK;
2088 	int err;
2089 
2090 	flush_spe_to_thread(current);
2091 
2092 	spefscr = current->thread.spefscr;
2093 	fpexc_mode = current->thread.fpexc_mode;
2094 
2095 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2096 		code = FPE_FLTOVF;
2097 	}
2098 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2099 		code = FPE_FLTUND;
2100 	}
2101 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2102 		code = FPE_FLTDIV;
2103 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2104 		code = FPE_FLTINV;
2105 	}
2106 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2107 		code = FPE_FLTRES;
2108 
2109 	err = do_spe_mathemu(regs);
2110 	if (err == 0) {
2111 		regs->nip += 4;		/* skip emulated instruction */
2112 		emulate_single_step(regs);
2113 		return;
2114 	}
2115 
2116 	if (err == -EFAULT) {
2117 		/* got an error reading the instruction */
2118 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2119 	} else if (err == -EINVAL) {
2120 		/* didn't recognize the instruction */
2121 		printk(KERN_ERR "unrecognized spe instruction "
2122 		       "in %s at %lx\n", current->comm, regs->nip);
2123 	} else {
2124 		_exception(SIGFPE, regs, code, regs->nip);
2125 	}
2126 
2127 	return;
2128 }
2129 
2130 void SPEFloatingPointRoundException(struct pt_regs *regs)
2131 {
2132 	extern int speround_handler(struct pt_regs *regs);
2133 	int err;
2134 
2135 	preempt_disable();
2136 	if (regs->msr & MSR_SPE)
2137 		giveup_spe(current);
2138 	preempt_enable();
2139 
2140 	regs->nip -= 4;
2141 	err = speround_handler(regs);
2142 	if (err == 0) {
2143 		regs->nip += 4;		/* skip emulated instruction */
2144 		emulate_single_step(regs);
2145 		return;
2146 	}
2147 
2148 	if (err == -EFAULT) {
2149 		/* got an error reading the instruction */
2150 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2151 	} else if (err == -EINVAL) {
2152 		/* didn't recognize the instruction */
2153 		printk(KERN_ERR "unrecognized spe instruction "
2154 		       "in %s at %lx\n", current->comm, regs->nip);
2155 	} else {
2156 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2157 		return;
2158 	}
2159 }
2160 #endif
2161 
2162 /*
2163  * We enter here if we get an unrecoverable exception, that is, one
2164  * that happened at a point where the RI (recoverable interrupt) bit
2165  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2166  * we therefore lost state by taking this exception.
2167  */
2168 void unrecoverable_exception(struct pt_regs *regs)
2169 {
2170 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2171 		 regs->trap, regs->nip, regs->msr);
2172 	die("Unrecoverable exception", regs, SIGABRT);
2173 }
2174 NOKPROBE_SYMBOL(unrecoverable_exception);
2175 
2176 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2177 /*
2178  * Default handler for a Watchdog exception,
2179  * spins until a reboot occurs
2180  */
2181 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2182 {
2183 	/* Generic WatchdogHandler, implement your own */
2184 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2185 	return;
2186 }
2187 
2188 void WatchdogException(struct pt_regs *regs)
2189 {
2190 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2191 	WatchdogHandler(regs);
2192 }
2193 #endif
2194 
2195 /*
2196  * We enter here if we discover during exception entry that we are
2197  * running in supervisor mode with a userspace value in the stack pointer.
2198  */
2199 void kernel_bad_stack(struct pt_regs *regs)
2200 {
2201 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2202 	       regs->gpr[1], regs->nip);
2203 	die("Bad kernel stack pointer", regs, SIGABRT);
2204 }
2205 NOKPROBE_SYMBOL(kernel_bad_stack);
2206 
2207 void __init trap_init(void)
2208 {
2209 }
2210 
2211 
2212 #ifdef CONFIG_PPC_EMULATED_STATS
2213 
2214 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2215 
2216 struct ppc_emulated ppc_emulated = {
2217 #ifdef CONFIG_ALTIVEC
2218 	WARN_EMULATED_SETUP(altivec),
2219 #endif
2220 	WARN_EMULATED_SETUP(dcba),
2221 	WARN_EMULATED_SETUP(dcbz),
2222 	WARN_EMULATED_SETUP(fp_pair),
2223 	WARN_EMULATED_SETUP(isel),
2224 	WARN_EMULATED_SETUP(mcrxr),
2225 	WARN_EMULATED_SETUP(mfpvr),
2226 	WARN_EMULATED_SETUP(multiple),
2227 	WARN_EMULATED_SETUP(popcntb),
2228 	WARN_EMULATED_SETUP(spe),
2229 	WARN_EMULATED_SETUP(string),
2230 	WARN_EMULATED_SETUP(sync),
2231 	WARN_EMULATED_SETUP(unaligned),
2232 #ifdef CONFIG_MATH_EMULATION
2233 	WARN_EMULATED_SETUP(math),
2234 #endif
2235 #ifdef CONFIG_VSX
2236 	WARN_EMULATED_SETUP(vsx),
2237 #endif
2238 #ifdef CONFIG_PPC64
2239 	WARN_EMULATED_SETUP(mfdscr),
2240 	WARN_EMULATED_SETUP(mtdscr),
2241 	WARN_EMULATED_SETUP(lq_stq),
2242 	WARN_EMULATED_SETUP(lxvw4x),
2243 	WARN_EMULATED_SETUP(lxvh8x),
2244 	WARN_EMULATED_SETUP(lxvd2x),
2245 	WARN_EMULATED_SETUP(lxvb16x),
2246 #endif
2247 };
2248 
2249 u32 ppc_warn_emulated;
2250 
2251 void ppc_warn_emulated_print(const char *type)
2252 {
2253 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2254 			    type);
2255 }
2256 
2257 static int __init ppc_warn_emulated_init(void)
2258 {
2259 	struct dentry *dir, *d;
2260 	unsigned int i;
2261 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2262 
2263 	if (!powerpc_debugfs_root)
2264 		return -ENODEV;
2265 
2266 	dir = debugfs_create_dir("emulated_instructions",
2267 				 powerpc_debugfs_root);
2268 	if (!dir)
2269 		return -ENOMEM;
2270 
2271 	d = debugfs_create_u32("do_warn", 0644, dir,
2272 			       &ppc_warn_emulated);
2273 	if (!d)
2274 		goto fail;
2275 
2276 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) {
2277 		d = debugfs_create_u32(entries[i].name, 0644, dir,
2278 				       (u32 *)&entries[i].val.counter);
2279 		if (!d)
2280 			goto fail;
2281 	}
2282 
2283 	return 0;
2284 
2285 fail:
2286 	debugfs_remove_recursive(dir);
2287 	return -ENOMEM;
2288 }
2289 
2290 device_initcall(ppc_warn_emulated_init);
2291 
2292 #endif /* CONFIG_PPC_EMULATED_STATS */
2293