xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision 28dce2c4)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Copyright (C) 1995-1996  Gary Thomas (gdt@linuxppc.org)
4  *  Copyright 2007-2010 Freescale Semiconductor, Inc.
5  *
6  *  Modified by Cort Dougan (cort@cs.nmt.edu)
7  *  and Paul Mackerras (paulus@samba.org)
8  */
9 
10 /*
11  * This file handles the architecture-dependent parts of hardware exceptions
12  */
13 
14 #include <linux/errno.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/pkeys.h>
20 #include <linux/stddef.h>
21 #include <linux/unistd.h>
22 #include <linux/ptrace.h>
23 #include <linux/user.h>
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/extable.h>
27 #include <linux/module.h>	/* print_modules */
28 #include <linux/prctl.h>
29 #include <linux/delay.h>
30 #include <linux/kprobes.h>
31 #include <linux/kexec.h>
32 #include <linux/backlight.h>
33 #include <linux/bug.h>
34 #include <linux/kdebug.h>
35 #include <linux/ratelimit.h>
36 #include <linux/context_tracking.h>
37 #include <linux/smp.h>
38 #include <linux/console.h>
39 #include <linux/kmsg_dump.h>
40 
41 #include <asm/emulated_ops.h>
42 #include <linux/uaccess.h>
43 #include <asm/debugfs.h>
44 #include <asm/interrupt.h>
45 #include <asm/io.h>
46 #include <asm/machdep.h>
47 #include <asm/rtas.h>
48 #include <asm/pmc.h>
49 #include <asm/reg.h>
50 #ifdef CONFIG_PMAC_BACKLIGHT
51 #include <asm/backlight.h>
52 #endif
53 #ifdef CONFIG_PPC64
54 #include <asm/firmware.h>
55 #include <asm/processor.h>
56 #endif
57 #include <asm/kexec.h>
58 #include <asm/ppc-opcode.h>
59 #include <asm/rio.h>
60 #include <asm/fadump.h>
61 #include <asm/switch_to.h>
62 #include <asm/tm.h>
63 #include <asm/debug.h>
64 #include <asm/asm-prototypes.h>
65 #include <asm/hmi.h>
66 #include <sysdev/fsl_pci.h>
67 #include <asm/kprobes.h>
68 #include <asm/stacktrace.h>
69 #include <asm/nmi.h>
70 
71 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
72 int (*__debugger)(struct pt_regs *regs) __read_mostly;
73 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
74 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
75 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
79 
80 EXPORT_SYMBOL(__debugger);
81 EXPORT_SYMBOL(__debugger_ipi);
82 EXPORT_SYMBOL(__debugger_bpt);
83 EXPORT_SYMBOL(__debugger_sstep);
84 EXPORT_SYMBOL(__debugger_iabr_match);
85 EXPORT_SYMBOL(__debugger_break_match);
86 EXPORT_SYMBOL(__debugger_fault_handler);
87 #endif
88 
89 /* Transactional Memory trap debug */
90 #ifdef TM_DEBUG_SW
91 #define TM_DEBUG(x...) printk(KERN_INFO x)
92 #else
93 #define TM_DEBUG(x...) do { } while(0)
94 #endif
95 
96 static const char *signame(int signr)
97 {
98 	switch (signr) {
99 	case SIGBUS:	return "bus error";
100 	case SIGFPE:	return "floating point exception";
101 	case SIGILL:	return "illegal instruction";
102 	case SIGSEGV:	return "segfault";
103 	case SIGTRAP:	return "unhandled trap";
104 	}
105 
106 	return "unknown signal";
107 }
108 
109 /*
110  * Trap & Exception support
111  */
112 
113 #ifdef CONFIG_PMAC_BACKLIGHT
114 static void pmac_backlight_unblank(void)
115 {
116 	mutex_lock(&pmac_backlight_mutex);
117 	if (pmac_backlight) {
118 		struct backlight_properties *props;
119 
120 		props = &pmac_backlight->props;
121 		props->brightness = props->max_brightness;
122 		props->power = FB_BLANK_UNBLANK;
123 		backlight_update_status(pmac_backlight);
124 	}
125 	mutex_unlock(&pmac_backlight_mutex);
126 }
127 #else
128 static inline void pmac_backlight_unblank(void) { }
129 #endif
130 
131 /*
132  * If oops/die is expected to crash the machine, return true here.
133  *
134  * This should not be expected to be 100% accurate, there may be
135  * notifiers registered or other unexpected conditions that may bring
136  * down the kernel. Or if the current process in the kernel is holding
137  * locks or has other critical state, the kernel may become effectively
138  * unusable anyway.
139  */
140 bool die_will_crash(void)
141 {
142 	if (should_fadump_crash())
143 		return true;
144 	if (kexec_should_crash(current))
145 		return true;
146 	if (in_interrupt() || panic_on_oops ||
147 			!current->pid || is_global_init(current))
148 		return true;
149 
150 	return false;
151 }
152 
153 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
154 static int die_owner = -1;
155 static unsigned int die_nest_count;
156 static int die_counter;
157 
158 extern void panic_flush_kmsg_start(void)
159 {
160 	/*
161 	 * These are mostly taken from kernel/panic.c, but tries to do
162 	 * relatively minimal work. Don't use delay functions (TB may
163 	 * be broken), don't crash dump (need to set a firmware log),
164 	 * don't run notifiers. We do want to get some information to
165 	 * Linux console.
166 	 */
167 	console_verbose();
168 	bust_spinlocks(1);
169 }
170 
171 extern void panic_flush_kmsg_end(void)
172 {
173 	printk_safe_flush_on_panic();
174 	kmsg_dump(KMSG_DUMP_PANIC);
175 	bust_spinlocks(0);
176 	debug_locks_off();
177 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
178 }
179 
180 static unsigned long oops_begin(struct pt_regs *regs)
181 {
182 	int cpu;
183 	unsigned long flags;
184 
185 	oops_enter();
186 
187 	/* racy, but better than risking deadlock. */
188 	raw_local_irq_save(flags);
189 	cpu = smp_processor_id();
190 	if (!arch_spin_trylock(&die_lock)) {
191 		if (cpu == die_owner)
192 			/* nested oops. should stop eventually */;
193 		else
194 			arch_spin_lock(&die_lock);
195 	}
196 	die_nest_count++;
197 	die_owner = cpu;
198 	console_verbose();
199 	bust_spinlocks(1);
200 	if (machine_is(powermac))
201 		pmac_backlight_unblank();
202 	return flags;
203 }
204 NOKPROBE_SYMBOL(oops_begin);
205 
206 static void oops_end(unsigned long flags, struct pt_regs *regs,
207 			       int signr)
208 {
209 	bust_spinlocks(0);
210 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
211 	die_nest_count--;
212 	oops_exit();
213 	printk("\n");
214 	if (!die_nest_count) {
215 		/* Nest count reaches zero, release the lock. */
216 		die_owner = -1;
217 		arch_spin_unlock(&die_lock);
218 	}
219 	raw_local_irq_restore(flags);
220 
221 	/*
222 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
223 	 */
224 	if (TRAP(regs) == INTERRUPT_SYSTEM_RESET)
225 		return;
226 
227 	crash_fadump(regs, "die oops");
228 
229 	if (kexec_should_crash(current))
230 		crash_kexec(regs);
231 
232 	if (!signr)
233 		return;
234 
235 	/*
236 	 * While our oops output is serialised by a spinlock, output
237 	 * from panic() called below can race and corrupt it. If we
238 	 * know we are going to panic, delay for 1 second so we have a
239 	 * chance to get clean backtraces from all CPUs that are oopsing.
240 	 */
241 	if (in_interrupt() || panic_on_oops || !current->pid ||
242 	    is_global_init(current)) {
243 		mdelay(MSEC_PER_SEC);
244 	}
245 
246 	if (panic_on_oops)
247 		panic("Fatal exception");
248 	do_exit(signr);
249 }
250 NOKPROBE_SYMBOL(oops_end);
251 
252 static char *get_mmu_str(void)
253 {
254 	if (early_radix_enabled())
255 		return " MMU=Radix";
256 	if (early_mmu_has_feature(MMU_FTR_HPTE_TABLE))
257 		return " MMU=Hash";
258 	return "";
259 }
260 
261 static int __die(const char *str, struct pt_regs *regs, long err)
262 {
263 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
264 
265 	printk("%s PAGE_SIZE=%luK%s%s%s%s%s%s %s\n",
266 	       IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE",
267 	       PAGE_SIZE / 1024, get_mmu_str(),
268 	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
269 	       IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
270 	       IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "",
271 	       debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
272 	       IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "",
273 	       ppc_md.name ? ppc_md.name : "");
274 
275 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
276 		return 1;
277 
278 	print_modules();
279 	show_regs(regs);
280 
281 	return 0;
282 }
283 NOKPROBE_SYMBOL(__die);
284 
285 void die(const char *str, struct pt_regs *regs, long err)
286 {
287 	unsigned long flags;
288 
289 	/*
290 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
291 	 */
292 	if (TRAP(regs) != INTERRUPT_SYSTEM_RESET) {
293 		if (debugger(regs))
294 			return;
295 	}
296 
297 	flags = oops_begin(regs);
298 	if (__die(str, regs, err))
299 		err = 0;
300 	oops_end(flags, regs, err);
301 }
302 NOKPROBE_SYMBOL(die);
303 
304 void user_single_step_report(struct pt_regs *regs)
305 {
306 	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip);
307 }
308 
309 static void show_signal_msg(int signr, struct pt_regs *regs, int code,
310 			    unsigned long addr)
311 {
312 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
313 				      DEFAULT_RATELIMIT_BURST);
314 
315 	if (!show_unhandled_signals)
316 		return;
317 
318 	if (!unhandled_signal(current, signr))
319 		return;
320 
321 	if (!__ratelimit(&rs))
322 		return;
323 
324 	pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x",
325 		current->comm, current->pid, signame(signr), signr,
326 		addr, regs->nip, regs->link, code);
327 
328 	print_vma_addr(KERN_CONT " in ", regs->nip);
329 
330 	pr_cont("\n");
331 
332 	show_user_instructions(regs);
333 }
334 
335 static bool exception_common(int signr, struct pt_regs *regs, int code,
336 			      unsigned long addr)
337 {
338 	if (!user_mode(regs)) {
339 		die("Exception in kernel mode", regs, signr);
340 		return false;
341 	}
342 
343 	show_signal_msg(signr, regs, code, addr);
344 
345 	if (arch_irqs_disabled())
346 		interrupt_cond_local_irq_enable(regs);
347 
348 	current->thread.trap_nr = code;
349 
350 	return true;
351 }
352 
353 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key)
354 {
355 	if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr))
356 		return;
357 
358 	force_sig_pkuerr((void __user *) addr, key);
359 }
360 
361 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
362 {
363 	if (!exception_common(signr, regs, code, addr))
364 		return;
365 
366 	force_sig_fault(signr, code, (void __user *)addr);
367 }
368 
369 /*
370  * The interrupt architecture has a quirk in that the HV interrupts excluding
371  * the NMIs (0x100 and 0x200) do not clear MSR[RI] at entry. The first thing
372  * that an interrupt handler must do is save off a GPR into a scratch register,
373  * and all interrupts on POWERNV (HV=1) use the HSPRG1 register as scratch.
374  * Therefore an NMI can clobber an HV interrupt's live HSPRG1 without noticing
375  * that it is non-reentrant, which leads to random data corruption.
376  *
377  * The solution is for NMI interrupts in HV mode to check if they originated
378  * from these critical HV interrupt regions. If so, then mark them not
379  * recoverable.
380  *
381  * An alternative would be for HV NMIs to use SPRG for scratch to avoid the
382  * HSPRG1 clobber, however this would cause guest SPRG to be clobbered. Linux
383  * guests should always have MSR[RI]=0 when its scratch SPRG is in use, so
384  * that would work. However any other guest OS that may have the SPRG live
385  * and MSR[RI]=1 could encounter silent corruption.
386  *
387  * Builds that do not support KVM could take this second option to increase
388  * the recoverability of NMIs.
389  */
390 void hv_nmi_check_nonrecoverable(struct pt_regs *regs)
391 {
392 #ifdef CONFIG_PPC_POWERNV
393 	unsigned long kbase = (unsigned long)_stext;
394 	unsigned long nip = regs->nip;
395 
396 	if (!(regs->msr & MSR_RI))
397 		return;
398 	if (!(regs->msr & MSR_HV))
399 		return;
400 	if (regs->msr & MSR_PR)
401 		return;
402 
403 	/*
404 	 * Now test if the interrupt has hit a range that may be using
405 	 * HSPRG1 without having RI=0 (i.e., an HSRR interrupt). The
406 	 * problem ranges all run un-relocated. Test real and virt modes
407 	 * at the same time by dropping the high bit of the nip (virt mode
408 	 * entry points still have the +0x4000 offset).
409 	 */
410 	nip &= ~0xc000000000000000ULL;
411 	if ((nip >= 0x500 && nip < 0x600) || (nip >= 0x4500 && nip < 0x4600))
412 		goto nonrecoverable;
413 	if ((nip >= 0x980 && nip < 0xa00) || (nip >= 0x4980 && nip < 0x4a00))
414 		goto nonrecoverable;
415 	if ((nip >= 0xe00 && nip < 0xec0) || (nip >= 0x4e00 && nip < 0x4ec0))
416 		goto nonrecoverable;
417 	if ((nip >= 0xf80 && nip < 0xfa0) || (nip >= 0x4f80 && nip < 0x4fa0))
418 		goto nonrecoverable;
419 
420 	/* Trampoline code runs un-relocated so subtract kbase. */
421 	if (nip >= (unsigned long)(start_real_trampolines - kbase) &&
422 			nip < (unsigned long)(end_real_trampolines - kbase))
423 		goto nonrecoverable;
424 	if (nip >= (unsigned long)(start_virt_trampolines - kbase) &&
425 			nip < (unsigned long)(end_virt_trampolines - kbase))
426 		goto nonrecoverable;
427 	return;
428 
429 nonrecoverable:
430 	regs->msr &= ~MSR_RI;
431 #endif
432 }
433 DEFINE_INTERRUPT_HANDLER_NMI(system_reset_exception)
434 {
435 	unsigned long hsrr0, hsrr1;
436 	bool saved_hsrrs = false;
437 
438 	/*
439 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
440 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
441 	 * OPAL), so save them here and restore them before returning.
442 	 *
443 	 * Machine checks don't need to save HSRRs, as the real mode handler
444 	 * is careful to avoid them, and the regular handler is not delivered
445 	 * as an NMI.
446 	 */
447 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
448 		hsrr0 = mfspr(SPRN_HSRR0);
449 		hsrr1 = mfspr(SPRN_HSRR1);
450 		saved_hsrrs = true;
451 	}
452 
453 	hv_nmi_check_nonrecoverable(regs);
454 
455 	__this_cpu_inc(irq_stat.sreset_irqs);
456 
457 	/* See if any machine dependent calls */
458 	if (ppc_md.system_reset_exception) {
459 		if (ppc_md.system_reset_exception(regs))
460 			goto out;
461 	}
462 
463 	if (debugger(regs))
464 		goto out;
465 
466 	kmsg_dump(KMSG_DUMP_OOPS);
467 	/*
468 	 * A system reset is a request to dump, so we always send
469 	 * it through the crashdump code (if fadump or kdump are
470 	 * registered).
471 	 */
472 	crash_fadump(regs, "System Reset");
473 
474 	crash_kexec(regs);
475 
476 	/*
477 	 * We aren't the primary crash CPU. We need to send it
478 	 * to a holding pattern to avoid it ending up in the panic
479 	 * code.
480 	 */
481 	crash_kexec_secondary(regs);
482 
483 	/*
484 	 * No debugger or crash dump registered, print logs then
485 	 * panic.
486 	 */
487 	die("System Reset", regs, SIGABRT);
488 
489 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
490 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
491 	nmi_panic(regs, "System Reset");
492 
493 out:
494 #ifdef CONFIG_PPC_BOOK3S_64
495 	BUG_ON(get_paca()->in_nmi == 0);
496 	if (get_paca()->in_nmi > 1)
497 		die("Unrecoverable nested System Reset", regs, SIGABRT);
498 #endif
499 	/* Must die if the interrupt is not recoverable */
500 	if (!(regs->msr & MSR_RI)) {
501 		/* For the reason explained in die_mce, nmi_exit before die */
502 		nmi_exit();
503 		die("Unrecoverable System Reset", regs, SIGABRT);
504 	}
505 
506 	if (saved_hsrrs) {
507 		mtspr(SPRN_HSRR0, hsrr0);
508 		mtspr(SPRN_HSRR1, hsrr1);
509 	}
510 
511 	/* What should we do here? We could issue a shutdown or hard reset. */
512 
513 	return 0;
514 }
515 
516 /*
517  * I/O accesses can cause machine checks on powermacs.
518  * Check if the NIP corresponds to the address of a sync
519  * instruction for which there is an entry in the exception
520  * table.
521  *  -- paulus.
522  */
523 static inline int check_io_access(struct pt_regs *regs)
524 {
525 #ifdef CONFIG_PPC32
526 	unsigned long msr = regs->msr;
527 	const struct exception_table_entry *entry;
528 	unsigned int *nip = (unsigned int *)regs->nip;
529 
530 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
531 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
532 		/*
533 		 * Check that it's a sync instruction, or somewhere
534 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
535 		 * As the address is in the exception table
536 		 * we should be able to read the instr there.
537 		 * For the debug message, we look at the preceding
538 		 * load or store.
539 		 */
540 		if (*nip == PPC_INST_NOP)
541 			nip -= 2;
542 		else if (*nip == PPC_INST_ISYNC)
543 			--nip;
544 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
545 			unsigned int rb;
546 
547 			--nip;
548 			rb = (*nip >> 11) & 0x1f;
549 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
550 			       (*nip & 0x100)? "OUT to": "IN from",
551 			       regs->gpr[rb] - _IO_BASE, nip);
552 			regs->msr |= MSR_RI;
553 			regs->nip = extable_fixup(entry);
554 			return 1;
555 		}
556 	}
557 #endif /* CONFIG_PPC32 */
558 	return 0;
559 }
560 
561 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
562 /* On 4xx, the reason for the machine check or program exception
563    is in the ESR. */
564 #define get_reason(regs)	((regs)->dsisr)
565 #define REASON_FP		ESR_FP
566 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
567 #define REASON_PRIVILEGED	ESR_PPR
568 #define REASON_TRAP		ESR_PTR
569 #define REASON_PREFIXED		0
570 #define REASON_BOUNDARY		0
571 
572 /* single-step stuff */
573 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
574 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
575 #define clear_br_trace(regs)	do {} while(0)
576 #else
577 /* On non-4xx, the reason for the machine check or program
578    exception is in the MSR. */
579 #define get_reason(regs)	((regs)->msr)
580 #define REASON_TM		SRR1_PROGTM
581 #define REASON_FP		SRR1_PROGFPE
582 #define REASON_ILLEGAL		SRR1_PROGILL
583 #define REASON_PRIVILEGED	SRR1_PROGPRIV
584 #define REASON_TRAP		SRR1_PROGTRAP
585 #define REASON_PREFIXED		SRR1_PREFIXED
586 #define REASON_BOUNDARY		SRR1_BOUNDARY
587 
588 #define single_stepping(regs)	((regs)->msr & MSR_SE)
589 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
590 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
591 #endif
592 
593 #define inst_length(reason)	(((reason) & REASON_PREFIXED) ? 8 : 4)
594 
595 #if defined(CONFIG_E500)
596 int machine_check_e500mc(struct pt_regs *regs)
597 {
598 	unsigned long mcsr = mfspr(SPRN_MCSR);
599 	unsigned long pvr = mfspr(SPRN_PVR);
600 	unsigned long reason = mcsr;
601 	int recoverable = 1;
602 
603 	if (reason & MCSR_LD) {
604 		recoverable = fsl_rio_mcheck_exception(regs);
605 		if (recoverable == 1)
606 			goto silent_out;
607 	}
608 
609 	printk("Machine check in kernel mode.\n");
610 	printk("Caused by (from MCSR=%lx): ", reason);
611 
612 	if (reason & MCSR_MCP)
613 		pr_cont("Machine Check Signal\n");
614 
615 	if (reason & MCSR_ICPERR) {
616 		pr_cont("Instruction Cache Parity Error\n");
617 
618 		/*
619 		 * This is recoverable by invalidating the i-cache.
620 		 */
621 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
622 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
623 			;
624 
625 		/*
626 		 * This will generally be accompanied by an instruction
627 		 * fetch error report -- only treat MCSR_IF as fatal
628 		 * if it wasn't due to an L1 parity error.
629 		 */
630 		reason &= ~MCSR_IF;
631 	}
632 
633 	if (reason & MCSR_DCPERR_MC) {
634 		pr_cont("Data Cache Parity Error\n");
635 
636 		/*
637 		 * In write shadow mode we auto-recover from the error, but it
638 		 * may still get logged and cause a machine check.  We should
639 		 * only treat the non-write shadow case as non-recoverable.
640 		 */
641 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
642 		 * is not implemented but L1 data cache always runs in write
643 		 * shadow mode. Hence on data cache parity errors HW will
644 		 * automatically invalidate the L1 Data Cache.
645 		 */
646 		if (PVR_VER(pvr) != PVR_VER_E6500) {
647 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
648 				recoverable = 0;
649 		}
650 	}
651 
652 	if (reason & MCSR_L2MMU_MHIT) {
653 		pr_cont("Hit on multiple TLB entries\n");
654 		recoverable = 0;
655 	}
656 
657 	if (reason & MCSR_NMI)
658 		pr_cont("Non-maskable interrupt\n");
659 
660 	if (reason & MCSR_IF) {
661 		pr_cont("Instruction Fetch Error Report\n");
662 		recoverable = 0;
663 	}
664 
665 	if (reason & MCSR_LD) {
666 		pr_cont("Load Error Report\n");
667 		recoverable = 0;
668 	}
669 
670 	if (reason & MCSR_ST) {
671 		pr_cont("Store Error Report\n");
672 		recoverable = 0;
673 	}
674 
675 	if (reason & MCSR_LDG) {
676 		pr_cont("Guarded Load Error Report\n");
677 		recoverable = 0;
678 	}
679 
680 	if (reason & MCSR_TLBSYNC)
681 		pr_cont("Simultaneous tlbsync operations\n");
682 
683 	if (reason & MCSR_BSL2_ERR) {
684 		pr_cont("Level 2 Cache Error\n");
685 		recoverable = 0;
686 	}
687 
688 	if (reason & MCSR_MAV) {
689 		u64 addr;
690 
691 		addr = mfspr(SPRN_MCAR);
692 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
693 
694 		pr_cont("Machine Check %s Address: %#llx\n",
695 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
696 	}
697 
698 silent_out:
699 	mtspr(SPRN_MCSR, mcsr);
700 	return mfspr(SPRN_MCSR) == 0 && recoverable;
701 }
702 
703 int machine_check_e500(struct pt_regs *regs)
704 {
705 	unsigned long reason = mfspr(SPRN_MCSR);
706 
707 	if (reason & MCSR_BUS_RBERR) {
708 		if (fsl_rio_mcheck_exception(regs))
709 			return 1;
710 		if (fsl_pci_mcheck_exception(regs))
711 			return 1;
712 	}
713 
714 	printk("Machine check in kernel mode.\n");
715 	printk("Caused by (from MCSR=%lx): ", reason);
716 
717 	if (reason & MCSR_MCP)
718 		pr_cont("Machine Check Signal\n");
719 	if (reason & MCSR_ICPERR)
720 		pr_cont("Instruction Cache Parity Error\n");
721 	if (reason & MCSR_DCP_PERR)
722 		pr_cont("Data Cache Push Parity Error\n");
723 	if (reason & MCSR_DCPERR)
724 		pr_cont("Data Cache Parity Error\n");
725 	if (reason & MCSR_BUS_IAERR)
726 		pr_cont("Bus - Instruction Address Error\n");
727 	if (reason & MCSR_BUS_RAERR)
728 		pr_cont("Bus - Read Address Error\n");
729 	if (reason & MCSR_BUS_WAERR)
730 		pr_cont("Bus - Write Address Error\n");
731 	if (reason & MCSR_BUS_IBERR)
732 		pr_cont("Bus - Instruction Data Error\n");
733 	if (reason & MCSR_BUS_RBERR)
734 		pr_cont("Bus - Read Data Bus Error\n");
735 	if (reason & MCSR_BUS_WBERR)
736 		pr_cont("Bus - Write Data Bus Error\n");
737 	if (reason & MCSR_BUS_IPERR)
738 		pr_cont("Bus - Instruction Parity Error\n");
739 	if (reason & MCSR_BUS_RPERR)
740 		pr_cont("Bus - Read Parity Error\n");
741 
742 	return 0;
743 }
744 
745 int machine_check_generic(struct pt_regs *regs)
746 {
747 	return 0;
748 }
749 #elif defined(CONFIG_PPC32)
750 int machine_check_generic(struct pt_regs *regs)
751 {
752 	unsigned long reason = regs->msr;
753 
754 	printk("Machine check in kernel mode.\n");
755 	printk("Caused by (from SRR1=%lx): ", reason);
756 	switch (reason & 0x601F0000) {
757 	case 0x80000:
758 		pr_cont("Machine check signal\n");
759 		break;
760 	case 0x40000:
761 	case 0x140000:	/* 7450 MSS error and TEA */
762 		pr_cont("Transfer error ack signal\n");
763 		break;
764 	case 0x20000:
765 		pr_cont("Data parity error signal\n");
766 		break;
767 	case 0x10000:
768 		pr_cont("Address parity error signal\n");
769 		break;
770 	case 0x20000000:
771 		pr_cont("L1 Data Cache error\n");
772 		break;
773 	case 0x40000000:
774 		pr_cont("L1 Instruction Cache error\n");
775 		break;
776 	case 0x00100000:
777 		pr_cont("L2 data cache parity error\n");
778 		break;
779 	default:
780 		pr_cont("Unknown values in msr\n");
781 	}
782 	return 0;
783 }
784 #endif /* everything else */
785 
786 void die_mce(const char *str, struct pt_regs *regs, long err)
787 {
788 	/*
789 	 * The machine check wants to kill the interrupted context, but
790 	 * do_exit() checks for in_interrupt() and panics in that case, so
791 	 * exit the irq/nmi before calling die.
792 	 */
793 	if (IS_ENABLED(CONFIG_PPC_BOOK3S_64))
794 		irq_exit();
795 	else
796 		nmi_exit();
797 	die(str, regs, err);
798 }
799 
800 /*
801  * BOOK3S_64 does not call this handler as a non-maskable interrupt
802  * (it uses its own early real-mode handler to handle the MCE proper
803  * and then raises irq_work to call this handler when interrupts are
804  * enabled).
805  */
806 #ifdef CONFIG_PPC_BOOK3S_64
807 DEFINE_INTERRUPT_HANDLER_ASYNC(machine_check_exception)
808 #else
809 DEFINE_INTERRUPT_HANDLER_NMI(machine_check_exception)
810 #endif
811 {
812 	int recover = 0;
813 
814 	__this_cpu_inc(irq_stat.mce_exceptions);
815 
816 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
817 
818 	/* See if any machine dependent calls. In theory, we would want
819 	 * to call the CPU first, and call the ppc_md. one if the CPU
820 	 * one returns a positive number. However there is existing code
821 	 * that assumes the board gets a first chance, so let's keep it
822 	 * that way for now and fix things later. --BenH.
823 	 */
824 	if (ppc_md.machine_check_exception)
825 		recover = ppc_md.machine_check_exception(regs);
826 	else if (cur_cpu_spec->machine_check)
827 		recover = cur_cpu_spec->machine_check(regs);
828 
829 	if (recover > 0)
830 		goto bail;
831 
832 	if (debugger_fault_handler(regs))
833 		goto bail;
834 
835 	if (check_io_access(regs))
836 		goto bail;
837 
838 	die_mce("Machine check", regs, SIGBUS);
839 
840 bail:
841 	/* Must die if the interrupt is not recoverable */
842 	if (!(regs->msr & MSR_RI))
843 		die_mce("Unrecoverable Machine check", regs, SIGBUS);
844 
845 #ifdef CONFIG_PPC_BOOK3S_64
846 	return;
847 #else
848 	return 0;
849 #endif
850 }
851 
852 DEFINE_INTERRUPT_HANDLER(SMIException) /* async? */
853 {
854 	die("System Management Interrupt", regs, SIGABRT);
855 }
856 
857 #ifdef CONFIG_VSX
858 static void p9_hmi_special_emu(struct pt_regs *regs)
859 {
860 	unsigned int ra, rb, t, i, sel, instr, rc;
861 	const void __user *addr;
862 	u8 vbuf[16] __aligned(16), *vdst;
863 	unsigned long ea, msr, msr_mask;
864 	bool swap;
865 
866 	if (__get_user(instr, (unsigned int __user *)regs->nip))
867 		return;
868 
869 	/*
870 	 * lxvb16x	opcode: 0x7c0006d8
871 	 * lxvd2x	opcode: 0x7c000698
872 	 * lxvh8x	opcode: 0x7c000658
873 	 * lxvw4x	opcode: 0x7c000618
874 	 */
875 	if ((instr & 0xfc00073e) != 0x7c000618) {
876 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
877 			 " instr=%08x\n",
878 			 smp_processor_id(), current->comm, current->pid,
879 			 regs->nip, instr);
880 		return;
881 	}
882 
883 	/* Grab vector registers into the task struct */
884 	msr = regs->msr; /* Grab msr before we flush the bits */
885 	flush_vsx_to_thread(current);
886 	enable_kernel_altivec();
887 
888 	/*
889 	 * Is userspace running with a different endian (this is rare but
890 	 * not impossible)
891 	 */
892 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
893 
894 	/* Decode the instruction */
895 	ra = (instr >> 16) & 0x1f;
896 	rb = (instr >> 11) & 0x1f;
897 	t = (instr >> 21) & 0x1f;
898 	if (instr & 1)
899 		vdst = (u8 *)&current->thread.vr_state.vr[t];
900 	else
901 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
902 
903 	/* Grab the vector address */
904 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
905 	if (is_32bit_task())
906 		ea &= 0xfffffffful;
907 	addr = (__force const void __user *)ea;
908 
909 	/* Check it */
910 	if (!access_ok(addr, 16)) {
911 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
912 			 " instr=%08x addr=%016lx\n",
913 			 smp_processor_id(), current->comm, current->pid,
914 			 regs->nip, instr, (unsigned long)addr);
915 		return;
916 	}
917 
918 	/* Read the vector */
919 	rc = 0;
920 	if ((unsigned long)addr & 0xfUL)
921 		/* unaligned case */
922 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
923 	else
924 		__get_user_atomic_128_aligned(vbuf, addr, rc);
925 	if (rc) {
926 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
927 			 " instr=%08x addr=%016lx\n",
928 			 smp_processor_id(), current->comm, current->pid,
929 			 regs->nip, instr, (unsigned long)addr);
930 		return;
931 	}
932 
933 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
934 		 " instr=%08x addr=%016lx\n",
935 		 smp_processor_id(), current->comm, current->pid, regs->nip,
936 		 instr, (unsigned long) addr);
937 
938 	/* Grab instruction "selector" */
939 	sel = (instr >> 6) & 3;
940 
941 	/*
942 	 * Check to make sure the facility is actually enabled. This
943 	 * could happen if we get a false positive hit.
944 	 *
945 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
946 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
947 	 */
948 	msr_mask = MSR_VSX;
949 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
950 		msr_mask = MSR_VEC;
951 	if (!(msr & msr_mask)) {
952 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
953 			 " instr=%08x msr:%016lx\n",
954 			 smp_processor_id(), current->comm, current->pid,
955 			 regs->nip, instr, msr);
956 		return;
957 	}
958 
959 	/* Do logging here before we modify sel based on endian */
960 	switch (sel) {
961 	case 0:	/* lxvw4x */
962 		PPC_WARN_EMULATED(lxvw4x, regs);
963 		break;
964 	case 1: /* lxvh8x */
965 		PPC_WARN_EMULATED(lxvh8x, regs);
966 		break;
967 	case 2: /* lxvd2x */
968 		PPC_WARN_EMULATED(lxvd2x, regs);
969 		break;
970 	case 3: /* lxvb16x */
971 		PPC_WARN_EMULATED(lxvb16x, regs);
972 		break;
973 	}
974 
975 #ifdef __LITTLE_ENDIAN__
976 	/*
977 	 * An LE kernel stores the vector in the task struct as an LE
978 	 * byte array (effectively swapping both the components and
979 	 * the content of the components). Those instructions expect
980 	 * the components to remain in ascending address order, so we
981 	 * swap them back.
982 	 *
983 	 * If we are running a BE user space, the expectation is that
984 	 * of a simple memcpy, so forcing the emulation to look like
985 	 * a lxvb16x should do the trick.
986 	 */
987 	if (swap)
988 		sel = 3;
989 
990 	switch (sel) {
991 	case 0:	/* lxvw4x */
992 		for (i = 0; i < 4; i++)
993 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
994 		break;
995 	case 1: /* lxvh8x */
996 		for (i = 0; i < 8; i++)
997 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
998 		break;
999 	case 2: /* lxvd2x */
1000 		for (i = 0; i < 2; i++)
1001 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1002 		break;
1003 	case 3: /* lxvb16x */
1004 		for (i = 0; i < 16; i++)
1005 			vdst[i] = vbuf[15-i];
1006 		break;
1007 	}
1008 #else /* __LITTLE_ENDIAN__ */
1009 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1010 	if (!swap)
1011 		sel = 3;
1012 
1013 	/* Otherwise, we need to swap the content of the components */
1014 	switch (sel) {
1015 	case 0:	/* lxvw4x */
1016 		for (i = 0; i < 4; i++)
1017 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1018 		break;
1019 	case 1: /* lxvh8x */
1020 		for (i = 0; i < 8; i++)
1021 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1022 		break;
1023 	case 2: /* lxvd2x */
1024 		for (i = 0; i < 2; i++)
1025 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1026 		break;
1027 	case 3: /* lxvb16x */
1028 		memcpy(vdst, vbuf, 16);
1029 		break;
1030 	}
1031 #endif /* !__LITTLE_ENDIAN__ */
1032 
1033 	/* Go to next instruction */
1034 	regs->nip += 4;
1035 }
1036 #endif /* CONFIG_VSX */
1037 
1038 DEFINE_INTERRUPT_HANDLER_ASYNC(handle_hmi_exception)
1039 {
1040 	struct pt_regs *old_regs;
1041 
1042 	old_regs = set_irq_regs(regs);
1043 
1044 #ifdef CONFIG_VSX
1045 	/* Real mode flagged P9 special emu is needed */
1046 	if (local_paca->hmi_p9_special_emu) {
1047 		local_paca->hmi_p9_special_emu = 0;
1048 
1049 		/*
1050 		 * We don't want to take page faults while doing the
1051 		 * emulation, we just replay the instruction if necessary.
1052 		 */
1053 		pagefault_disable();
1054 		p9_hmi_special_emu(regs);
1055 		pagefault_enable();
1056 	}
1057 #endif /* CONFIG_VSX */
1058 
1059 	if (ppc_md.handle_hmi_exception)
1060 		ppc_md.handle_hmi_exception(regs);
1061 
1062 	set_irq_regs(old_regs);
1063 }
1064 
1065 DEFINE_INTERRUPT_HANDLER(unknown_exception)
1066 {
1067 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1068 	       regs->nip, regs->msr, regs->trap);
1069 
1070 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1071 }
1072 
1073 DEFINE_INTERRUPT_HANDLER_ASYNC(unknown_async_exception)
1074 {
1075 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1076 	       regs->nip, regs->msr, regs->trap);
1077 
1078 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1079 }
1080 
1081 DEFINE_INTERRUPT_HANDLER_NMI(unknown_nmi_exception)
1082 {
1083 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1084 	       regs->nip, regs->msr, regs->trap);
1085 
1086 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1087 
1088 	return 0;
1089 }
1090 
1091 DEFINE_INTERRUPT_HANDLER(instruction_breakpoint_exception)
1092 {
1093 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1094 					5, SIGTRAP) == NOTIFY_STOP)
1095 		return;
1096 	if (debugger_iabr_match(regs))
1097 		return;
1098 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1099 }
1100 
1101 DEFINE_INTERRUPT_HANDLER(RunModeException)
1102 {
1103 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1104 }
1105 
1106 DEFINE_INTERRUPT_HANDLER(single_step_exception)
1107 {
1108 	clear_single_step(regs);
1109 	clear_br_trace(regs);
1110 
1111 	if (kprobe_post_handler(regs))
1112 		return;
1113 
1114 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1115 					5, SIGTRAP) == NOTIFY_STOP)
1116 		return;
1117 	if (debugger_sstep(regs))
1118 		return;
1119 
1120 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1121 }
1122 
1123 /*
1124  * After we have successfully emulated an instruction, we have to
1125  * check if the instruction was being single-stepped, and if so,
1126  * pretend we got a single-step exception.  This was pointed out
1127  * by Kumar Gala.  -- paulus
1128  */
1129 static void emulate_single_step(struct pt_regs *regs)
1130 {
1131 	if (single_stepping(regs))
1132 		single_step_exception(regs);
1133 }
1134 
1135 static inline int __parse_fpscr(unsigned long fpscr)
1136 {
1137 	int ret = FPE_FLTUNK;
1138 
1139 	/* Invalid operation */
1140 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1141 		ret = FPE_FLTINV;
1142 
1143 	/* Overflow */
1144 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1145 		ret = FPE_FLTOVF;
1146 
1147 	/* Underflow */
1148 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1149 		ret = FPE_FLTUND;
1150 
1151 	/* Divide by zero */
1152 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1153 		ret = FPE_FLTDIV;
1154 
1155 	/* Inexact result */
1156 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1157 		ret = FPE_FLTRES;
1158 
1159 	return ret;
1160 }
1161 
1162 static void parse_fpe(struct pt_regs *regs)
1163 {
1164 	int code = 0;
1165 
1166 	flush_fp_to_thread(current);
1167 
1168 #ifdef CONFIG_PPC_FPU_REGS
1169 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1170 #endif
1171 
1172 	_exception(SIGFPE, regs, code, regs->nip);
1173 }
1174 
1175 /*
1176  * Illegal instruction emulation support.  Originally written to
1177  * provide the PVR to user applications using the mfspr rd, PVR.
1178  * Return non-zero if we can't emulate, or -EFAULT if the associated
1179  * memory access caused an access fault.  Return zero on success.
1180  *
1181  * There are a couple of ways to do this, either "decode" the instruction
1182  * or directly match lots of bits.  In this case, matching lots of
1183  * bits is faster and easier.
1184  *
1185  */
1186 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1187 {
1188 	u8 rT = (instword >> 21) & 0x1f;
1189 	u8 rA = (instword >> 16) & 0x1f;
1190 	u8 NB_RB = (instword >> 11) & 0x1f;
1191 	u32 num_bytes;
1192 	unsigned long EA;
1193 	int pos = 0;
1194 
1195 	/* Early out if we are an invalid form of lswx */
1196 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1197 		if ((rT == rA) || (rT == NB_RB))
1198 			return -EINVAL;
1199 
1200 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1201 
1202 	switch (instword & PPC_INST_STRING_MASK) {
1203 		case PPC_INST_LSWX:
1204 		case PPC_INST_STSWX:
1205 			EA += NB_RB;
1206 			num_bytes = regs->xer & 0x7f;
1207 			break;
1208 		case PPC_INST_LSWI:
1209 		case PPC_INST_STSWI:
1210 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1211 			break;
1212 		default:
1213 			return -EINVAL;
1214 	}
1215 
1216 	while (num_bytes != 0)
1217 	{
1218 		u8 val;
1219 		u32 shift = 8 * (3 - (pos & 0x3));
1220 
1221 		/* if process is 32-bit, clear upper 32 bits of EA */
1222 		if ((regs->msr & MSR_64BIT) == 0)
1223 			EA &= 0xFFFFFFFF;
1224 
1225 		switch ((instword & PPC_INST_STRING_MASK)) {
1226 			case PPC_INST_LSWX:
1227 			case PPC_INST_LSWI:
1228 				if (get_user(val, (u8 __user *)EA))
1229 					return -EFAULT;
1230 				/* first time updating this reg,
1231 				 * zero it out */
1232 				if (pos == 0)
1233 					regs->gpr[rT] = 0;
1234 				regs->gpr[rT] |= val << shift;
1235 				break;
1236 			case PPC_INST_STSWI:
1237 			case PPC_INST_STSWX:
1238 				val = regs->gpr[rT] >> shift;
1239 				if (put_user(val, (u8 __user *)EA))
1240 					return -EFAULT;
1241 				break;
1242 		}
1243 		/* move EA to next address */
1244 		EA += 1;
1245 		num_bytes--;
1246 
1247 		/* manage our position within the register */
1248 		if (++pos == 4) {
1249 			pos = 0;
1250 			if (++rT == 32)
1251 				rT = 0;
1252 		}
1253 	}
1254 
1255 	return 0;
1256 }
1257 
1258 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1259 {
1260 	u32 ra,rs;
1261 	unsigned long tmp;
1262 
1263 	ra = (instword >> 16) & 0x1f;
1264 	rs = (instword >> 21) & 0x1f;
1265 
1266 	tmp = regs->gpr[rs];
1267 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1268 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1269 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1270 	regs->gpr[ra] = tmp;
1271 
1272 	return 0;
1273 }
1274 
1275 static int emulate_isel(struct pt_regs *regs, u32 instword)
1276 {
1277 	u8 rT = (instword >> 21) & 0x1f;
1278 	u8 rA = (instword >> 16) & 0x1f;
1279 	u8 rB = (instword >> 11) & 0x1f;
1280 	u8 BC = (instword >> 6) & 0x1f;
1281 	u8 bit;
1282 	unsigned long tmp;
1283 
1284 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1285 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1286 
1287 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1288 
1289 	return 0;
1290 }
1291 
1292 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1293 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1294 {
1295         /* If we're emulating a load/store in an active transaction, we cannot
1296          * emulate it as the kernel operates in transaction suspended context.
1297          * We need to abort the transaction.  This creates a persistent TM
1298          * abort so tell the user what caused it with a new code.
1299 	 */
1300 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1301 		tm_enable();
1302 		tm_abort(cause);
1303 		return true;
1304 	}
1305 	return false;
1306 }
1307 #else
1308 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1309 {
1310 	return false;
1311 }
1312 #endif
1313 
1314 static int emulate_instruction(struct pt_regs *regs)
1315 {
1316 	u32 instword;
1317 	u32 rd;
1318 
1319 	if (!user_mode(regs))
1320 		return -EINVAL;
1321 
1322 	if (get_user(instword, (u32 __user *)(regs->nip)))
1323 		return -EFAULT;
1324 
1325 	/* Emulate the mfspr rD, PVR. */
1326 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1327 		PPC_WARN_EMULATED(mfpvr, regs);
1328 		rd = (instword >> 21) & 0x1f;
1329 		regs->gpr[rd] = mfspr(SPRN_PVR);
1330 		return 0;
1331 	}
1332 
1333 	/* Emulating the dcba insn is just a no-op.  */
1334 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1335 		PPC_WARN_EMULATED(dcba, regs);
1336 		return 0;
1337 	}
1338 
1339 	/* Emulate the mcrxr insn.  */
1340 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1341 		int shift = (instword >> 21) & 0x1c;
1342 		unsigned long msk = 0xf0000000UL >> shift;
1343 
1344 		PPC_WARN_EMULATED(mcrxr, regs);
1345 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1346 		regs->xer &= ~0xf0000000UL;
1347 		return 0;
1348 	}
1349 
1350 	/* Emulate load/store string insn. */
1351 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1352 		if (tm_abort_check(regs,
1353 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1354 			return -EINVAL;
1355 		PPC_WARN_EMULATED(string, regs);
1356 		return emulate_string_inst(regs, instword);
1357 	}
1358 
1359 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1360 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1361 		PPC_WARN_EMULATED(popcntb, regs);
1362 		return emulate_popcntb_inst(regs, instword);
1363 	}
1364 
1365 	/* Emulate isel (Integer Select) instruction */
1366 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1367 		PPC_WARN_EMULATED(isel, regs);
1368 		return emulate_isel(regs, instword);
1369 	}
1370 
1371 	/* Emulate sync instruction variants */
1372 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1373 		PPC_WARN_EMULATED(sync, regs);
1374 		asm volatile("sync");
1375 		return 0;
1376 	}
1377 
1378 #ifdef CONFIG_PPC64
1379 	/* Emulate the mfspr rD, DSCR. */
1380 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1381 		PPC_INST_MFSPR_DSCR_USER) ||
1382 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1383 		PPC_INST_MFSPR_DSCR)) &&
1384 			cpu_has_feature(CPU_FTR_DSCR)) {
1385 		PPC_WARN_EMULATED(mfdscr, regs);
1386 		rd = (instword >> 21) & 0x1f;
1387 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1388 		return 0;
1389 	}
1390 	/* Emulate the mtspr DSCR, rD. */
1391 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1392 		PPC_INST_MTSPR_DSCR_USER) ||
1393 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1394 		PPC_INST_MTSPR_DSCR)) &&
1395 			cpu_has_feature(CPU_FTR_DSCR)) {
1396 		PPC_WARN_EMULATED(mtdscr, regs);
1397 		rd = (instword >> 21) & 0x1f;
1398 		current->thread.dscr = regs->gpr[rd];
1399 		current->thread.dscr_inherit = 1;
1400 		mtspr(SPRN_DSCR, current->thread.dscr);
1401 		return 0;
1402 	}
1403 #endif
1404 
1405 	return -EINVAL;
1406 }
1407 
1408 int is_valid_bugaddr(unsigned long addr)
1409 {
1410 	return is_kernel_addr(addr);
1411 }
1412 
1413 #ifdef CONFIG_MATH_EMULATION
1414 static int emulate_math(struct pt_regs *regs)
1415 {
1416 	int ret;
1417 
1418 	ret = do_mathemu(regs);
1419 	if (ret >= 0)
1420 		PPC_WARN_EMULATED(math, regs);
1421 
1422 	switch (ret) {
1423 	case 0:
1424 		emulate_single_step(regs);
1425 		return 0;
1426 	case 1: {
1427 			int code = 0;
1428 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1429 			_exception(SIGFPE, regs, code, regs->nip);
1430 			return 0;
1431 		}
1432 	case -EFAULT:
1433 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1434 		return 0;
1435 	}
1436 
1437 	return -1;
1438 }
1439 #else
1440 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1441 #endif
1442 
1443 static void do_program_check(struct pt_regs *regs)
1444 {
1445 	unsigned int reason = get_reason(regs);
1446 
1447 	/* We can now get here via a FP Unavailable exception if the core
1448 	 * has no FPU, in that case the reason flags will be 0 */
1449 
1450 	if (reason & REASON_FP) {
1451 		/* IEEE FP exception */
1452 		parse_fpe(regs);
1453 		return;
1454 	}
1455 	if (reason & REASON_TRAP) {
1456 		unsigned long bugaddr;
1457 		/* Debugger is first in line to stop recursive faults in
1458 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1459 		if (debugger_bpt(regs))
1460 			return;
1461 
1462 		if (kprobe_handler(regs))
1463 			return;
1464 
1465 		/* trap exception */
1466 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1467 				== NOTIFY_STOP)
1468 			return;
1469 
1470 		bugaddr = regs->nip;
1471 		/*
1472 		 * Fixup bugaddr for BUG_ON() in real mode
1473 		 */
1474 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1475 			bugaddr += PAGE_OFFSET;
1476 
1477 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1478 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1479 			regs->nip += 4;
1480 			return;
1481 		}
1482 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1483 		return;
1484 	}
1485 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1486 	if (reason & REASON_TM) {
1487 		/* This is a TM "Bad Thing Exception" program check.
1488 		 * This occurs when:
1489 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1490 		 *    transition in TM states.
1491 		 * -  A trechkpt is attempted when transactional.
1492 		 * -  A treclaim is attempted when non transactional.
1493 		 * -  A tend is illegally attempted.
1494 		 * -  writing a TM SPR when transactional.
1495 		 *
1496 		 * If usermode caused this, it's done something illegal and
1497 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1498 		 * operand to distinguish from the instruction just being bad
1499 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1500 		 * illegal /placement/ of a valid instruction.
1501 		 */
1502 		if (user_mode(regs)) {
1503 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1504 			return;
1505 		} else {
1506 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1507 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1508 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1509 			die("Unrecoverable exception", regs, SIGABRT);
1510 		}
1511 	}
1512 #endif
1513 
1514 	/*
1515 	 * If we took the program check in the kernel skip down to sending a
1516 	 * SIGILL. The subsequent cases all relate to emulating instructions
1517 	 * which we should only do for userspace. We also do not want to enable
1518 	 * interrupts for kernel faults because that might lead to further
1519 	 * faults, and loose the context of the original exception.
1520 	 */
1521 	if (!user_mode(regs))
1522 		goto sigill;
1523 
1524 	interrupt_cond_local_irq_enable(regs);
1525 
1526 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1527 	 * but there seems to be a hardware bug on the 405GP (RevD)
1528 	 * that means ESR is sometimes set incorrectly - either to
1529 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1530 	 * hardware people - not sure if it can happen on any illegal
1531 	 * instruction or only on FP instructions, whether there is a
1532 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1533 	 */
1534 	if (!emulate_math(regs))
1535 		return;
1536 
1537 	/* Try to emulate it if we should. */
1538 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1539 		switch (emulate_instruction(regs)) {
1540 		case 0:
1541 			regs->nip += 4;
1542 			emulate_single_step(regs);
1543 			return;
1544 		case -EFAULT:
1545 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1546 			return;
1547 		}
1548 	}
1549 
1550 sigill:
1551 	if (reason & REASON_PRIVILEGED)
1552 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1553 	else
1554 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1555 
1556 }
1557 
1558 DEFINE_INTERRUPT_HANDLER(program_check_exception)
1559 {
1560 	do_program_check(regs);
1561 }
1562 
1563 /*
1564  * This occurs when running in hypervisor mode on POWER6 or later
1565  * and an illegal instruction is encountered.
1566  */
1567 DEFINE_INTERRUPT_HANDLER(emulation_assist_interrupt)
1568 {
1569 	regs->msr |= REASON_ILLEGAL;
1570 	do_program_check(regs);
1571 }
1572 
1573 DEFINE_INTERRUPT_HANDLER(alignment_exception)
1574 {
1575 	int sig, code, fixed = 0;
1576 	unsigned long  reason;
1577 
1578 	interrupt_cond_local_irq_enable(regs);
1579 
1580 	reason = get_reason(regs);
1581 	if (reason & REASON_BOUNDARY) {
1582 		sig = SIGBUS;
1583 		code = BUS_ADRALN;
1584 		goto bad;
1585 	}
1586 
1587 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1588 		return;
1589 
1590 	/* we don't implement logging of alignment exceptions */
1591 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1592 		fixed = fix_alignment(regs);
1593 
1594 	if (fixed == 1) {
1595 		/* skip over emulated instruction */
1596 		regs->nip += inst_length(reason);
1597 		emulate_single_step(regs);
1598 		return;
1599 	}
1600 
1601 	/* Operand address was bad */
1602 	if (fixed == -EFAULT) {
1603 		sig = SIGSEGV;
1604 		code = SEGV_ACCERR;
1605 	} else {
1606 		sig = SIGBUS;
1607 		code = BUS_ADRALN;
1608 	}
1609 bad:
1610 	if (user_mode(regs))
1611 		_exception(sig, regs, code, regs->dar);
1612 	else
1613 		bad_page_fault(regs, sig);
1614 }
1615 
1616 DEFINE_INTERRUPT_HANDLER(stack_overflow_exception)
1617 {
1618 	die("Kernel stack overflow", regs, SIGSEGV);
1619 }
1620 
1621 DEFINE_INTERRUPT_HANDLER(kernel_fp_unavailable_exception)
1622 {
1623 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1624 			  "%lx at %lx\n", regs->trap, regs->nip);
1625 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1626 }
1627 
1628 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_exception)
1629 {
1630 	if (user_mode(regs)) {
1631 		/* A user program has executed an altivec instruction,
1632 		   but this kernel doesn't support altivec. */
1633 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1634 		return;
1635 	}
1636 
1637 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1638 			"%lx at %lx\n", regs->trap, regs->nip);
1639 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1640 }
1641 
1642 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_exception)
1643 {
1644 	if (user_mode(regs)) {
1645 		/* A user program has executed an vsx instruction,
1646 		   but this kernel doesn't support vsx. */
1647 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1648 		return;
1649 	}
1650 
1651 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1652 			"%lx at %lx\n", regs->trap, regs->nip);
1653 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1654 }
1655 
1656 #ifdef CONFIG_PPC64
1657 static void tm_unavailable(struct pt_regs *regs)
1658 {
1659 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1660 	if (user_mode(regs)) {
1661 		current->thread.load_tm++;
1662 		regs->msr |= MSR_TM;
1663 		tm_enable();
1664 		tm_restore_sprs(&current->thread);
1665 		return;
1666 	}
1667 #endif
1668 	pr_emerg("Unrecoverable TM Unavailable Exception "
1669 			"%lx at %lx\n", regs->trap, regs->nip);
1670 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1671 }
1672 
1673 DEFINE_INTERRUPT_HANDLER(facility_unavailable_exception)
1674 {
1675 	static char *facility_strings[] = {
1676 		[FSCR_FP_LG] = "FPU",
1677 		[FSCR_VECVSX_LG] = "VMX/VSX",
1678 		[FSCR_DSCR_LG] = "DSCR",
1679 		[FSCR_PM_LG] = "PMU SPRs",
1680 		[FSCR_BHRB_LG] = "BHRB",
1681 		[FSCR_TM_LG] = "TM",
1682 		[FSCR_EBB_LG] = "EBB",
1683 		[FSCR_TAR_LG] = "TAR",
1684 		[FSCR_MSGP_LG] = "MSGP",
1685 		[FSCR_SCV_LG] = "SCV",
1686 		[FSCR_PREFIX_LG] = "PREFIX",
1687 	};
1688 	char *facility = "unknown";
1689 	u64 value;
1690 	u32 instword, rd;
1691 	u8 status;
1692 	bool hv;
1693 
1694 	hv = (TRAP(regs) == INTERRUPT_H_FAC_UNAVAIL);
1695 	if (hv)
1696 		value = mfspr(SPRN_HFSCR);
1697 	else
1698 		value = mfspr(SPRN_FSCR);
1699 
1700 	status = value >> 56;
1701 	if ((hv || status >= 2) &&
1702 	    (status < ARRAY_SIZE(facility_strings)) &&
1703 	    facility_strings[status])
1704 		facility = facility_strings[status];
1705 
1706 	/* We should not have taken this interrupt in kernel */
1707 	if (!user_mode(regs)) {
1708 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1709 			 facility, status, regs->nip);
1710 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1711 	}
1712 
1713 	interrupt_cond_local_irq_enable(regs);
1714 
1715 	if (status == FSCR_DSCR_LG) {
1716 		/*
1717 		 * User is accessing the DSCR register using the problem
1718 		 * state only SPR number (0x03) either through a mfspr or
1719 		 * a mtspr instruction. If it is a write attempt through
1720 		 * a mtspr, then we set the inherit bit. This also allows
1721 		 * the user to write or read the register directly in the
1722 		 * future by setting via the FSCR DSCR bit. But in case it
1723 		 * is a read DSCR attempt through a mfspr instruction, we
1724 		 * just emulate the instruction instead. This code path will
1725 		 * always emulate all the mfspr instructions till the user
1726 		 * has attempted at least one mtspr instruction. This way it
1727 		 * preserves the same behaviour when the user is accessing
1728 		 * the DSCR through privilege level only SPR number (0x11)
1729 		 * which is emulated through illegal instruction exception.
1730 		 * We always leave HFSCR DSCR set.
1731 		 */
1732 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1733 			pr_err("Failed to fetch the user instruction\n");
1734 			return;
1735 		}
1736 
1737 		/* Write into DSCR (mtspr 0x03, RS) */
1738 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1739 				== PPC_INST_MTSPR_DSCR_USER) {
1740 			rd = (instword >> 21) & 0x1f;
1741 			current->thread.dscr = regs->gpr[rd];
1742 			current->thread.dscr_inherit = 1;
1743 			current->thread.fscr |= FSCR_DSCR;
1744 			mtspr(SPRN_FSCR, current->thread.fscr);
1745 		}
1746 
1747 		/* Read from DSCR (mfspr RT, 0x03) */
1748 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1749 				== PPC_INST_MFSPR_DSCR_USER) {
1750 			if (emulate_instruction(regs)) {
1751 				pr_err("DSCR based mfspr emulation failed\n");
1752 				return;
1753 			}
1754 			regs->nip += 4;
1755 			emulate_single_step(regs);
1756 		}
1757 		return;
1758 	}
1759 
1760 	if (status == FSCR_TM_LG) {
1761 		/*
1762 		 * If we're here then the hardware is TM aware because it
1763 		 * generated an exception with FSRM_TM set.
1764 		 *
1765 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1766 		 * told us not to do TM, or the kernel is not built with TM
1767 		 * support.
1768 		 *
1769 		 * If both of those things are true, then userspace can spam the
1770 		 * console by triggering the printk() below just by continually
1771 		 * doing tbegin (or any TM instruction). So in that case just
1772 		 * send the process a SIGILL immediately.
1773 		 */
1774 		if (!cpu_has_feature(CPU_FTR_TM))
1775 			goto out;
1776 
1777 		tm_unavailable(regs);
1778 		return;
1779 	}
1780 
1781 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1782 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1783 
1784 out:
1785 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1786 }
1787 #endif
1788 
1789 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1790 
1791 DEFINE_INTERRUPT_HANDLER(fp_unavailable_tm)
1792 {
1793 	/* Note:  This does not handle any kind of FP laziness. */
1794 
1795 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1796 		 regs->nip, regs->msr);
1797 
1798         /* We can only have got here if the task started using FP after
1799          * beginning the transaction.  So, the transactional regs are just a
1800          * copy of the checkpointed ones.  But, we still need to recheckpoint
1801          * as we're enabling FP for the process; it will return, abort the
1802          * transaction, and probably retry but now with FP enabled.  So the
1803          * checkpointed FP registers need to be loaded.
1804 	 */
1805 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1806 
1807 	/*
1808 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1809 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1810 	 *
1811 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1812 	 * recheckpoint.
1813 	 */
1814 
1815 	/* Enable FP for the task: */
1816 	current->thread.load_fp = 1;
1817 
1818 	/*
1819 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1820 	 */
1821 	tm_recheckpoint(&current->thread);
1822 }
1823 
1824 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_tm)
1825 {
1826 	/* See the comments in fp_unavailable_tm().  This function operates
1827 	 * the same way.
1828 	 */
1829 
1830 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1831 		 "MSR=%lx\n",
1832 		 regs->nip, regs->msr);
1833 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1834 	current->thread.load_vec = 1;
1835 	tm_recheckpoint(&current->thread);
1836 	current->thread.used_vr = 1;
1837 }
1838 
1839 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_tm)
1840 {
1841 	/* See the comments in fp_unavailable_tm().  This works similarly,
1842 	 * though we're loading both FP and VEC registers in here.
1843 	 *
1844 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1845 	 * regs.  Either way, set MSR_VSX.
1846 	 */
1847 
1848 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1849 		 "MSR=%lx\n",
1850 		 regs->nip, regs->msr);
1851 
1852 	current->thread.used_vsr = 1;
1853 
1854 	/* This reclaims FP and/or VR regs if they're already enabled */
1855 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1856 
1857 	current->thread.load_vec = 1;
1858 	current->thread.load_fp = 1;
1859 
1860 	tm_recheckpoint(&current->thread);
1861 }
1862 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1863 
1864 #ifdef CONFIG_PPC64
1865 DECLARE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi);
1866 DEFINE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi)
1867 {
1868 	__this_cpu_inc(irq_stat.pmu_irqs);
1869 
1870 	perf_irq(regs);
1871 
1872 	return 0;
1873 }
1874 #endif
1875 
1876 DECLARE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async);
1877 DEFINE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async)
1878 {
1879 	__this_cpu_inc(irq_stat.pmu_irqs);
1880 
1881 	perf_irq(regs);
1882 }
1883 
1884 DEFINE_INTERRUPT_HANDLER_RAW(performance_monitor_exception)
1885 {
1886 	/*
1887 	 * On 64-bit, if perf interrupts hit in a local_irq_disable
1888 	 * (soft-masked) region, we consider them as NMIs. This is required to
1889 	 * prevent hash faults on user addresses when reading callchains (and
1890 	 * looks better from an irq tracing perspective).
1891 	 */
1892 	if (IS_ENABLED(CONFIG_PPC64) && unlikely(arch_irq_disabled_regs(regs)))
1893 		performance_monitor_exception_nmi(regs);
1894 	else
1895 		performance_monitor_exception_async(regs);
1896 
1897 	return 0;
1898 }
1899 
1900 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1901 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1902 {
1903 	int changed = 0;
1904 	/*
1905 	 * Determine the cause of the debug event, clear the
1906 	 * event flags and send a trap to the handler. Torez
1907 	 */
1908 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1909 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1910 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1911 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1912 #endif
1913 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1914 			     5);
1915 		changed |= 0x01;
1916 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1917 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1918 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1919 			     6);
1920 		changed |= 0x01;
1921 	}  else if (debug_status & DBSR_IAC1) {
1922 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1923 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1924 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1925 			     1);
1926 		changed |= 0x01;
1927 	}  else if (debug_status & DBSR_IAC2) {
1928 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1929 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1930 			     2);
1931 		changed |= 0x01;
1932 	}  else if (debug_status & DBSR_IAC3) {
1933 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1934 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1935 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1936 			     3);
1937 		changed |= 0x01;
1938 	}  else if (debug_status & DBSR_IAC4) {
1939 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1940 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1941 			     4);
1942 		changed |= 0x01;
1943 	}
1944 	/*
1945 	 * At the point this routine was called, the MSR(DE) was turned off.
1946 	 * Check all other debug flags and see if that bit needs to be turned
1947 	 * back on or not.
1948 	 */
1949 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1950 			       current->thread.debug.dbcr1))
1951 		regs->msr |= MSR_DE;
1952 	else
1953 		/* Make sure the IDM flag is off */
1954 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1955 
1956 	if (changed & 0x01)
1957 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1958 }
1959 
1960 DEFINE_INTERRUPT_HANDLER(DebugException)
1961 {
1962 	unsigned long debug_status = regs->dsisr;
1963 
1964 	current->thread.debug.dbsr = debug_status;
1965 
1966 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1967 	 * on server, it stops on the target of the branch. In order to simulate
1968 	 * the server behaviour, we thus restart right away with a single step
1969 	 * instead of stopping here when hitting a BT
1970 	 */
1971 	if (debug_status & DBSR_BT) {
1972 		regs->msr &= ~MSR_DE;
1973 
1974 		/* Disable BT */
1975 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1976 		/* Clear the BT event */
1977 		mtspr(SPRN_DBSR, DBSR_BT);
1978 
1979 		/* Do the single step trick only when coming from userspace */
1980 		if (user_mode(regs)) {
1981 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1982 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1983 			regs->msr |= MSR_DE;
1984 			return;
1985 		}
1986 
1987 		if (kprobe_post_handler(regs))
1988 			return;
1989 
1990 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1991 			       5, SIGTRAP) == NOTIFY_STOP) {
1992 			return;
1993 		}
1994 		if (debugger_sstep(regs))
1995 			return;
1996 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1997 		regs->msr &= ~MSR_DE;
1998 
1999 		/* Disable instruction completion */
2000 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
2001 		/* Clear the instruction completion event */
2002 		mtspr(SPRN_DBSR, DBSR_IC);
2003 
2004 		if (kprobe_post_handler(regs))
2005 			return;
2006 
2007 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2008 			       5, SIGTRAP) == NOTIFY_STOP) {
2009 			return;
2010 		}
2011 
2012 		if (debugger_sstep(regs))
2013 			return;
2014 
2015 		if (user_mode(regs)) {
2016 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2017 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2018 					       current->thread.debug.dbcr1))
2019 				regs->msr |= MSR_DE;
2020 			else
2021 				/* Make sure the IDM bit is off */
2022 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2023 		}
2024 
2025 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2026 	} else
2027 		handle_debug(regs, debug_status);
2028 }
2029 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2030 
2031 #ifdef CONFIG_ALTIVEC
2032 DEFINE_INTERRUPT_HANDLER(altivec_assist_exception)
2033 {
2034 	int err;
2035 
2036 	if (!user_mode(regs)) {
2037 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2038 		       " at %lx\n", regs->nip);
2039 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2040 	}
2041 
2042 	flush_altivec_to_thread(current);
2043 
2044 	PPC_WARN_EMULATED(altivec, regs);
2045 	err = emulate_altivec(regs);
2046 	if (err == 0) {
2047 		regs->nip += 4;		/* skip emulated instruction */
2048 		emulate_single_step(regs);
2049 		return;
2050 	}
2051 
2052 	if (err == -EFAULT) {
2053 		/* got an error reading the instruction */
2054 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2055 	} else {
2056 		/* didn't recognize the instruction */
2057 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2058 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2059 				   "in %s at %lx\n", current->comm, regs->nip);
2060 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2061 	}
2062 }
2063 #endif /* CONFIG_ALTIVEC */
2064 
2065 #ifdef CONFIG_FSL_BOOKE
2066 DEFINE_INTERRUPT_HANDLER(CacheLockingException)
2067 {
2068 	unsigned long error_code = regs->dsisr;
2069 
2070 	/* We treat cache locking instructions from the user
2071 	 * as priv ops, in the future we could try to do
2072 	 * something smarter
2073 	 */
2074 	if (error_code & (ESR_DLK|ESR_ILK))
2075 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2076 	return;
2077 }
2078 #endif /* CONFIG_FSL_BOOKE */
2079 
2080 #ifdef CONFIG_SPE
2081 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointException)
2082 {
2083 	extern int do_spe_mathemu(struct pt_regs *regs);
2084 	unsigned long spefscr;
2085 	int fpexc_mode;
2086 	int code = FPE_FLTUNK;
2087 	int err;
2088 
2089 	interrupt_cond_local_irq_enable(regs);
2090 
2091 	flush_spe_to_thread(current);
2092 
2093 	spefscr = current->thread.spefscr;
2094 	fpexc_mode = current->thread.fpexc_mode;
2095 
2096 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2097 		code = FPE_FLTOVF;
2098 	}
2099 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2100 		code = FPE_FLTUND;
2101 	}
2102 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2103 		code = FPE_FLTDIV;
2104 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2105 		code = FPE_FLTINV;
2106 	}
2107 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2108 		code = FPE_FLTRES;
2109 
2110 	err = do_spe_mathemu(regs);
2111 	if (err == 0) {
2112 		regs->nip += 4;		/* skip emulated instruction */
2113 		emulate_single_step(regs);
2114 		return;
2115 	}
2116 
2117 	if (err == -EFAULT) {
2118 		/* got an error reading the instruction */
2119 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2120 	} else if (err == -EINVAL) {
2121 		/* didn't recognize the instruction */
2122 		printk(KERN_ERR "unrecognized spe instruction "
2123 		       "in %s at %lx\n", current->comm, regs->nip);
2124 	} else {
2125 		_exception(SIGFPE, regs, code, regs->nip);
2126 	}
2127 
2128 	return;
2129 }
2130 
2131 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointRoundException)
2132 {
2133 	extern int speround_handler(struct pt_regs *regs);
2134 	int err;
2135 
2136 	interrupt_cond_local_irq_enable(regs);
2137 
2138 	preempt_disable();
2139 	if (regs->msr & MSR_SPE)
2140 		giveup_spe(current);
2141 	preempt_enable();
2142 
2143 	regs->nip -= 4;
2144 	err = speround_handler(regs);
2145 	if (err == 0) {
2146 		regs->nip += 4;		/* skip emulated instruction */
2147 		emulate_single_step(regs);
2148 		return;
2149 	}
2150 
2151 	if (err == -EFAULT) {
2152 		/* got an error reading the instruction */
2153 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2154 	} else if (err == -EINVAL) {
2155 		/* didn't recognize the instruction */
2156 		printk(KERN_ERR "unrecognized spe instruction "
2157 		       "in %s at %lx\n", current->comm, regs->nip);
2158 	} else {
2159 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2160 		return;
2161 	}
2162 }
2163 #endif
2164 
2165 /*
2166  * We enter here if we get an unrecoverable exception, that is, one
2167  * that happened at a point where the RI (recoverable interrupt) bit
2168  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2169  * we therefore lost state by taking this exception.
2170  */
2171 void __noreturn unrecoverable_exception(struct pt_regs *regs)
2172 {
2173 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2174 		 regs->trap, regs->nip, regs->msr);
2175 	die("Unrecoverable exception", regs, SIGABRT);
2176 	/* die() should not return */
2177 	for (;;)
2178 		;
2179 }
2180 
2181 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2182 /*
2183  * Default handler for a Watchdog exception,
2184  * spins until a reboot occurs
2185  */
2186 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2187 {
2188 	/* Generic WatchdogHandler, implement your own */
2189 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2190 	return;
2191 }
2192 
2193 DEFINE_INTERRUPT_HANDLER_NMI(WatchdogException)
2194 {
2195 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2196 	WatchdogHandler(regs);
2197 	return 0;
2198 }
2199 #endif
2200 
2201 /*
2202  * We enter here if we discover during exception entry that we are
2203  * running in supervisor mode with a userspace value in the stack pointer.
2204  */
2205 DEFINE_INTERRUPT_HANDLER(kernel_bad_stack)
2206 {
2207 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2208 	       regs->gpr[1], regs->nip);
2209 	die("Bad kernel stack pointer", regs, SIGABRT);
2210 }
2211 
2212 void __init trap_init(void)
2213 {
2214 }
2215 
2216 
2217 #ifdef CONFIG_PPC_EMULATED_STATS
2218 
2219 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2220 
2221 struct ppc_emulated ppc_emulated = {
2222 #ifdef CONFIG_ALTIVEC
2223 	WARN_EMULATED_SETUP(altivec),
2224 #endif
2225 	WARN_EMULATED_SETUP(dcba),
2226 	WARN_EMULATED_SETUP(dcbz),
2227 	WARN_EMULATED_SETUP(fp_pair),
2228 	WARN_EMULATED_SETUP(isel),
2229 	WARN_EMULATED_SETUP(mcrxr),
2230 	WARN_EMULATED_SETUP(mfpvr),
2231 	WARN_EMULATED_SETUP(multiple),
2232 	WARN_EMULATED_SETUP(popcntb),
2233 	WARN_EMULATED_SETUP(spe),
2234 	WARN_EMULATED_SETUP(string),
2235 	WARN_EMULATED_SETUP(sync),
2236 	WARN_EMULATED_SETUP(unaligned),
2237 #ifdef CONFIG_MATH_EMULATION
2238 	WARN_EMULATED_SETUP(math),
2239 #endif
2240 #ifdef CONFIG_VSX
2241 	WARN_EMULATED_SETUP(vsx),
2242 #endif
2243 #ifdef CONFIG_PPC64
2244 	WARN_EMULATED_SETUP(mfdscr),
2245 	WARN_EMULATED_SETUP(mtdscr),
2246 	WARN_EMULATED_SETUP(lq_stq),
2247 	WARN_EMULATED_SETUP(lxvw4x),
2248 	WARN_EMULATED_SETUP(lxvh8x),
2249 	WARN_EMULATED_SETUP(lxvd2x),
2250 	WARN_EMULATED_SETUP(lxvb16x),
2251 #endif
2252 };
2253 
2254 u32 ppc_warn_emulated;
2255 
2256 void ppc_warn_emulated_print(const char *type)
2257 {
2258 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2259 			    type);
2260 }
2261 
2262 static int __init ppc_warn_emulated_init(void)
2263 {
2264 	struct dentry *dir;
2265 	unsigned int i;
2266 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2267 
2268 	dir = debugfs_create_dir("emulated_instructions",
2269 				 powerpc_debugfs_root);
2270 
2271 	debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated);
2272 
2273 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++)
2274 		debugfs_create_u32(entries[i].name, 0644, dir,
2275 				   (u32 *)&entries[i].val.counter);
2276 
2277 	return 0;
2278 }
2279 
2280 device_initcall(ppc_warn_emulated_init);
2281 
2282 #endif /* CONFIG_PPC_EMULATED_STATS */
2283