xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision de528723)
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 <asm/pgtable.h>
43 #include <linux/uaccess.h>
44 #include <asm/debugfs.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 #include <asm/tm.h>
57 #endif
58 #include <asm/kexec.h>
59 #include <asm/ppc-opcode.h>
60 #include <asm/rio.h>
61 #include <asm/fadump.h>
62 #include <asm/switch_to.h>
63 #include <asm/tm.h>
64 #include <asm/debug.h>
65 #include <asm/asm-prototypes.h>
66 #include <asm/hmi.h>
67 #include <sysdev/fsl_pci.h>
68 #include <asm/kprobes.h>
69 #include <asm/stacktrace.h>
70 #include <asm/nmi.h>
71 
72 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
73 int (*__debugger)(struct pt_regs *regs) __read_mostly;
74 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
75 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
79 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
80 
81 EXPORT_SYMBOL(__debugger);
82 EXPORT_SYMBOL(__debugger_ipi);
83 EXPORT_SYMBOL(__debugger_bpt);
84 EXPORT_SYMBOL(__debugger_sstep);
85 EXPORT_SYMBOL(__debugger_iabr_match);
86 EXPORT_SYMBOL(__debugger_break_match);
87 EXPORT_SYMBOL(__debugger_fault_handler);
88 #endif
89 
90 /* Transactional Memory trap debug */
91 #ifdef TM_DEBUG_SW
92 #define TM_DEBUG(x...) printk(KERN_INFO x)
93 #else
94 #define TM_DEBUG(x...) do { } while(0)
95 #endif
96 
97 static const char *signame(int signr)
98 {
99 	switch (signr) {
100 	case SIGBUS:	return "bus error";
101 	case SIGFPE:	return "floating point exception";
102 	case SIGILL:	return "illegal instruction";
103 	case SIGSEGV:	return "segfault";
104 	case SIGTRAP:	return "unhandled trap";
105 	}
106 
107 	return "unknown signal";
108 }
109 
110 /*
111  * Trap & Exception support
112  */
113 
114 #ifdef CONFIG_PMAC_BACKLIGHT
115 static void pmac_backlight_unblank(void)
116 {
117 	mutex_lock(&pmac_backlight_mutex);
118 	if (pmac_backlight) {
119 		struct backlight_properties *props;
120 
121 		props = &pmac_backlight->props;
122 		props->brightness = props->max_brightness;
123 		props->power = FB_BLANK_UNBLANK;
124 		backlight_update_status(pmac_backlight);
125 	}
126 	mutex_unlock(&pmac_backlight_mutex);
127 }
128 #else
129 static inline void pmac_backlight_unblank(void) { }
130 #endif
131 
132 /*
133  * If oops/die is expected to crash the machine, return true here.
134  *
135  * This should not be expected to be 100% accurate, there may be
136  * notifiers registered or other unexpected conditions that may bring
137  * down the kernel. Or if the current process in the kernel is holding
138  * locks or has other critical state, the kernel may become effectively
139  * unusable anyway.
140  */
141 bool die_will_crash(void)
142 {
143 	if (should_fadump_crash())
144 		return true;
145 	if (kexec_should_crash(current))
146 		return true;
147 	if (in_interrupt() || panic_on_oops ||
148 			!current->pid || is_global_init(current))
149 		return true;
150 
151 	return false;
152 }
153 
154 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
155 static int die_owner = -1;
156 static unsigned int die_nest_count;
157 static int die_counter;
158 
159 extern void panic_flush_kmsg_start(void)
160 {
161 	/*
162 	 * These are mostly taken from kernel/panic.c, but tries to do
163 	 * relatively minimal work. Don't use delay functions (TB may
164 	 * be broken), don't crash dump (need to set a firmware log),
165 	 * don't run notifiers. We do want to get some information to
166 	 * Linux console.
167 	 */
168 	console_verbose();
169 	bust_spinlocks(1);
170 }
171 
172 extern void panic_flush_kmsg_end(void)
173 {
174 	printk_safe_flush_on_panic();
175 	kmsg_dump(KMSG_DUMP_PANIC);
176 	bust_spinlocks(0);
177 	debug_locks_off();
178 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
179 }
180 
181 static unsigned long oops_begin(struct pt_regs *regs)
182 {
183 	int cpu;
184 	unsigned long flags;
185 
186 	oops_enter();
187 
188 	/* racy, but better than risking deadlock. */
189 	raw_local_irq_save(flags);
190 	cpu = smp_processor_id();
191 	if (!arch_spin_trylock(&die_lock)) {
192 		if (cpu == die_owner)
193 			/* nested oops. should stop eventually */;
194 		else
195 			arch_spin_lock(&die_lock);
196 	}
197 	die_nest_count++;
198 	die_owner = cpu;
199 	console_verbose();
200 	bust_spinlocks(1);
201 	if (machine_is(powermac))
202 		pmac_backlight_unblank();
203 	return flags;
204 }
205 NOKPROBE_SYMBOL(oops_begin);
206 
207 static void oops_end(unsigned long flags, struct pt_regs *regs,
208 			       int signr)
209 {
210 	bust_spinlocks(0);
211 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
212 	die_nest_count--;
213 	oops_exit();
214 	printk("\n");
215 	if (!die_nest_count) {
216 		/* Nest count reaches zero, release the lock. */
217 		die_owner = -1;
218 		arch_spin_unlock(&die_lock);
219 	}
220 	raw_local_irq_restore(flags);
221 
222 	/*
223 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
224 	 */
225 	if (TRAP(regs) == 0x100)
226 		return;
227 
228 	crash_fadump(regs, "die oops");
229 
230 	if (kexec_should_crash(current))
231 		crash_kexec(regs);
232 
233 	if (!signr)
234 		return;
235 
236 	/*
237 	 * While our oops output is serialised by a spinlock, output
238 	 * from panic() called below can race and corrupt it. If we
239 	 * know we are going to panic, delay for 1 second so we have a
240 	 * chance to get clean backtraces from all CPUs that are oopsing.
241 	 */
242 	if (in_interrupt() || panic_on_oops || !current->pid ||
243 	    is_global_init(current)) {
244 		mdelay(MSEC_PER_SEC);
245 	}
246 
247 	if (panic_on_oops)
248 		panic("Fatal exception");
249 	do_exit(signr);
250 }
251 NOKPROBE_SYMBOL(oops_end);
252 
253 static int __die(const char *str, struct pt_regs *regs, long err)
254 {
255 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
256 
257 	printk("%s PAGE_SIZE=%luK%s%s%s%s%s%s%s %s\n",
258 	       IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE",
259 	       PAGE_SIZE / 1024,
260 	       early_radix_enabled() ? " MMU=Radix" : "",
261 	       early_mmu_has_feature(MMU_FTR_HPTE_TABLE) ? " MMU=Hash" : "",
262 	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
263 	       IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
264 	       IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "",
265 	       debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
266 	       IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "",
267 	       ppc_md.name ? ppc_md.name : "");
268 
269 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
270 		return 1;
271 
272 	print_modules();
273 	show_regs(regs);
274 
275 	return 0;
276 }
277 NOKPROBE_SYMBOL(__die);
278 
279 void die(const char *str, struct pt_regs *regs, long err)
280 {
281 	unsigned long flags;
282 
283 	/*
284 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
285 	 */
286 	if (TRAP(regs) != 0x100) {
287 		if (debugger(regs))
288 			return;
289 	}
290 
291 	flags = oops_begin(regs);
292 	if (__die(str, regs, err))
293 		err = 0;
294 	oops_end(flags, regs, err);
295 }
296 NOKPROBE_SYMBOL(die);
297 
298 void user_single_step_report(struct pt_regs *regs)
299 {
300 	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip);
301 }
302 
303 static void show_signal_msg(int signr, struct pt_regs *regs, int code,
304 			    unsigned long addr)
305 {
306 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
307 				      DEFAULT_RATELIMIT_BURST);
308 
309 	if (!show_unhandled_signals)
310 		return;
311 
312 	if (!unhandled_signal(current, signr))
313 		return;
314 
315 	if (!__ratelimit(&rs))
316 		return;
317 
318 	pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x",
319 		current->comm, current->pid, signame(signr), signr,
320 		addr, regs->nip, regs->link, code);
321 
322 	print_vma_addr(KERN_CONT " in ", regs->nip);
323 
324 	pr_cont("\n");
325 
326 	show_user_instructions(regs);
327 }
328 
329 static bool exception_common(int signr, struct pt_regs *regs, int code,
330 			      unsigned long addr)
331 {
332 	if (!user_mode(regs)) {
333 		die("Exception in kernel mode", regs, signr);
334 		return false;
335 	}
336 
337 	show_signal_msg(signr, regs, code, addr);
338 
339 	if (arch_irqs_disabled() && !arch_irq_disabled_regs(regs))
340 		local_irq_enable();
341 
342 	current->thread.trap_nr = code;
343 
344 	/*
345 	 * Save all the pkey registers AMR/IAMR/UAMOR. Eg: Core dumps need
346 	 * to capture the content, if the task gets killed.
347 	 */
348 	thread_pkey_regs_save(&current->thread);
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 droping 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 
434 void system_reset_exception(struct pt_regs *regs)
435 {
436 	unsigned long hsrr0, hsrr1;
437 	bool nested = in_nmi();
438 	bool saved_hsrrs = false;
439 
440 	/*
441 	 * Avoid crashes in case of nested NMI exceptions. Recoverability
442 	 * is determined by RI and in_nmi
443 	 */
444 	if (!nested)
445 		nmi_enter();
446 
447 	/*
448 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
449 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
450 	 * OPAL), so save them here and restore them before returning.
451 	 *
452 	 * Machine checks don't need to save HSRRs, as the real mode handler
453 	 * is careful to avoid them, and the regular handler is not delivered
454 	 * as an NMI.
455 	 */
456 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
457 		hsrr0 = mfspr(SPRN_HSRR0);
458 		hsrr1 = mfspr(SPRN_HSRR1);
459 		saved_hsrrs = true;
460 	}
461 
462 	hv_nmi_check_nonrecoverable(regs);
463 
464 	__this_cpu_inc(irq_stat.sreset_irqs);
465 
466 	/* See if any machine dependent calls */
467 	if (ppc_md.system_reset_exception) {
468 		if (ppc_md.system_reset_exception(regs))
469 			goto out;
470 	}
471 
472 	if (debugger(regs))
473 		goto out;
474 
475 	kmsg_dump(KMSG_DUMP_OOPS);
476 	/*
477 	 * A system reset is a request to dump, so we always send
478 	 * it through the crashdump code (if fadump or kdump are
479 	 * registered).
480 	 */
481 	crash_fadump(regs, "System Reset");
482 
483 	crash_kexec(regs);
484 
485 	/*
486 	 * We aren't the primary crash CPU. We need to send it
487 	 * to a holding pattern to avoid it ending up in the panic
488 	 * code.
489 	 */
490 	crash_kexec_secondary(regs);
491 
492 	/*
493 	 * No debugger or crash dump registered, print logs then
494 	 * panic.
495 	 */
496 	die("System Reset", regs, SIGABRT);
497 
498 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
499 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
500 	nmi_panic(regs, "System Reset");
501 
502 out:
503 #ifdef CONFIG_PPC_BOOK3S_64
504 	BUG_ON(get_paca()->in_nmi == 0);
505 	if (get_paca()->in_nmi > 1)
506 		nmi_panic(regs, "Unrecoverable nested System Reset");
507 #endif
508 	/* Must die if the interrupt is not recoverable */
509 	if (!(regs->msr & MSR_RI))
510 		nmi_panic(regs, "Unrecoverable System Reset");
511 
512 	if (saved_hsrrs) {
513 		mtspr(SPRN_HSRR0, hsrr0);
514 		mtspr(SPRN_HSRR1, hsrr1);
515 	}
516 
517 	if (!nested)
518 		nmi_exit();
519 
520 	/* What should we do here? We could issue a shutdown or hard reset. */
521 }
522 
523 /*
524  * I/O accesses can cause machine checks on powermacs.
525  * Check if the NIP corresponds to the address of a sync
526  * instruction for which there is an entry in the exception
527  * table.
528  * Note that the 601 only takes a machine check on TEA
529  * (transfer error ack) signal assertion, and does not
530  * set any of the top 16 bits of SRR1.
531  *  -- paulus.
532  */
533 static inline int check_io_access(struct pt_regs *regs)
534 {
535 #ifdef CONFIG_PPC32
536 	unsigned long msr = regs->msr;
537 	const struct exception_table_entry *entry;
538 	unsigned int *nip = (unsigned int *)regs->nip;
539 
540 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
541 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
542 		/*
543 		 * Check that it's a sync instruction, or somewhere
544 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
545 		 * As the address is in the exception table
546 		 * we should be able to read the instr there.
547 		 * For the debug message, we look at the preceding
548 		 * load or store.
549 		 */
550 		if (*nip == PPC_INST_NOP)
551 			nip -= 2;
552 		else if (*nip == PPC_INST_ISYNC)
553 			--nip;
554 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
555 			unsigned int rb;
556 
557 			--nip;
558 			rb = (*nip >> 11) & 0x1f;
559 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
560 			       (*nip & 0x100)? "OUT to": "IN from",
561 			       regs->gpr[rb] - _IO_BASE, nip);
562 			regs->msr |= MSR_RI;
563 			regs->nip = extable_fixup(entry);
564 			return 1;
565 		}
566 	}
567 #endif /* CONFIG_PPC32 */
568 	return 0;
569 }
570 
571 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
572 /* On 4xx, the reason for the machine check or program exception
573    is in the ESR. */
574 #define get_reason(regs)	((regs)->dsisr)
575 #define REASON_FP		ESR_FP
576 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
577 #define REASON_PRIVILEGED	ESR_PPR
578 #define REASON_TRAP		ESR_PTR
579 
580 /* single-step stuff */
581 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
582 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
583 #define clear_br_trace(regs)	do {} while(0)
584 #else
585 /* On non-4xx, the reason for the machine check or program
586    exception is in the MSR. */
587 #define get_reason(regs)	((regs)->msr)
588 #define REASON_TM		SRR1_PROGTM
589 #define REASON_FP		SRR1_PROGFPE
590 #define REASON_ILLEGAL		SRR1_PROGILL
591 #define REASON_PRIVILEGED	SRR1_PROGPRIV
592 #define REASON_TRAP		SRR1_PROGTRAP
593 
594 #define single_stepping(regs)	((regs)->msr & MSR_SE)
595 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
596 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
597 #endif
598 
599 #if defined(CONFIG_E500)
600 int machine_check_e500mc(struct pt_regs *regs)
601 {
602 	unsigned long mcsr = mfspr(SPRN_MCSR);
603 	unsigned long pvr = mfspr(SPRN_PVR);
604 	unsigned long reason = mcsr;
605 	int recoverable = 1;
606 
607 	if (reason & MCSR_LD) {
608 		recoverable = fsl_rio_mcheck_exception(regs);
609 		if (recoverable == 1)
610 			goto silent_out;
611 	}
612 
613 	printk("Machine check in kernel mode.\n");
614 	printk("Caused by (from MCSR=%lx): ", reason);
615 
616 	if (reason & MCSR_MCP)
617 		pr_cont("Machine Check Signal\n");
618 
619 	if (reason & MCSR_ICPERR) {
620 		pr_cont("Instruction Cache Parity Error\n");
621 
622 		/*
623 		 * This is recoverable by invalidating the i-cache.
624 		 */
625 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
626 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
627 			;
628 
629 		/*
630 		 * This will generally be accompanied by an instruction
631 		 * fetch error report -- only treat MCSR_IF as fatal
632 		 * if it wasn't due to an L1 parity error.
633 		 */
634 		reason &= ~MCSR_IF;
635 	}
636 
637 	if (reason & MCSR_DCPERR_MC) {
638 		pr_cont("Data Cache Parity Error\n");
639 
640 		/*
641 		 * In write shadow mode we auto-recover from the error, but it
642 		 * may still get logged and cause a machine check.  We should
643 		 * only treat the non-write shadow case as non-recoverable.
644 		 */
645 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
646 		 * is not implemented but L1 data cache always runs in write
647 		 * shadow mode. Hence on data cache parity errors HW will
648 		 * automatically invalidate the L1 Data Cache.
649 		 */
650 		if (PVR_VER(pvr) != PVR_VER_E6500) {
651 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
652 				recoverable = 0;
653 		}
654 	}
655 
656 	if (reason & MCSR_L2MMU_MHIT) {
657 		pr_cont("Hit on multiple TLB entries\n");
658 		recoverable = 0;
659 	}
660 
661 	if (reason & MCSR_NMI)
662 		pr_cont("Non-maskable interrupt\n");
663 
664 	if (reason & MCSR_IF) {
665 		pr_cont("Instruction Fetch Error Report\n");
666 		recoverable = 0;
667 	}
668 
669 	if (reason & MCSR_LD) {
670 		pr_cont("Load Error Report\n");
671 		recoverable = 0;
672 	}
673 
674 	if (reason & MCSR_ST) {
675 		pr_cont("Store Error Report\n");
676 		recoverable = 0;
677 	}
678 
679 	if (reason & MCSR_LDG) {
680 		pr_cont("Guarded Load Error Report\n");
681 		recoverable = 0;
682 	}
683 
684 	if (reason & MCSR_TLBSYNC)
685 		pr_cont("Simultaneous tlbsync operations\n");
686 
687 	if (reason & MCSR_BSL2_ERR) {
688 		pr_cont("Level 2 Cache Error\n");
689 		recoverable = 0;
690 	}
691 
692 	if (reason & MCSR_MAV) {
693 		u64 addr;
694 
695 		addr = mfspr(SPRN_MCAR);
696 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
697 
698 		pr_cont("Machine Check %s Address: %#llx\n",
699 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
700 	}
701 
702 silent_out:
703 	mtspr(SPRN_MCSR, mcsr);
704 	return mfspr(SPRN_MCSR) == 0 && recoverable;
705 }
706 
707 int machine_check_e500(struct pt_regs *regs)
708 {
709 	unsigned long reason = mfspr(SPRN_MCSR);
710 
711 	if (reason & MCSR_BUS_RBERR) {
712 		if (fsl_rio_mcheck_exception(regs))
713 			return 1;
714 		if (fsl_pci_mcheck_exception(regs))
715 			return 1;
716 	}
717 
718 	printk("Machine check in kernel mode.\n");
719 	printk("Caused by (from MCSR=%lx): ", reason);
720 
721 	if (reason & MCSR_MCP)
722 		pr_cont("Machine Check Signal\n");
723 	if (reason & MCSR_ICPERR)
724 		pr_cont("Instruction Cache Parity Error\n");
725 	if (reason & MCSR_DCP_PERR)
726 		pr_cont("Data Cache Push Parity Error\n");
727 	if (reason & MCSR_DCPERR)
728 		pr_cont("Data Cache Parity Error\n");
729 	if (reason & MCSR_BUS_IAERR)
730 		pr_cont("Bus - Instruction Address Error\n");
731 	if (reason & MCSR_BUS_RAERR)
732 		pr_cont("Bus - Read Address Error\n");
733 	if (reason & MCSR_BUS_WAERR)
734 		pr_cont("Bus - Write Address Error\n");
735 	if (reason & MCSR_BUS_IBERR)
736 		pr_cont("Bus - Instruction Data Error\n");
737 	if (reason & MCSR_BUS_RBERR)
738 		pr_cont("Bus - Read Data Bus Error\n");
739 	if (reason & MCSR_BUS_WBERR)
740 		pr_cont("Bus - Write Data Bus Error\n");
741 	if (reason & MCSR_BUS_IPERR)
742 		pr_cont("Bus - Instruction Parity Error\n");
743 	if (reason & MCSR_BUS_RPERR)
744 		pr_cont("Bus - Read Parity Error\n");
745 
746 	return 0;
747 }
748 
749 int machine_check_generic(struct pt_regs *regs)
750 {
751 	return 0;
752 }
753 #elif defined(CONFIG_E200)
754 int machine_check_e200(struct pt_regs *regs)
755 {
756 	unsigned long reason = mfspr(SPRN_MCSR);
757 
758 	printk("Machine check in kernel mode.\n");
759 	printk("Caused by (from MCSR=%lx): ", reason);
760 
761 	if (reason & MCSR_MCP)
762 		pr_cont("Machine Check Signal\n");
763 	if (reason & MCSR_CP_PERR)
764 		pr_cont("Cache Push Parity Error\n");
765 	if (reason & MCSR_CPERR)
766 		pr_cont("Cache Parity Error\n");
767 	if (reason & MCSR_EXCP_ERR)
768 		pr_cont("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
769 	if (reason & MCSR_BUS_IRERR)
770 		pr_cont("Bus - Read Bus Error on instruction fetch\n");
771 	if (reason & MCSR_BUS_DRERR)
772 		pr_cont("Bus - Read Bus Error on data load\n");
773 	if (reason & MCSR_BUS_WRERR)
774 		pr_cont("Bus - Write Bus Error on buffered store or cache line push\n");
775 
776 	return 0;
777 }
778 #elif defined(CONFIG_PPC32)
779 int machine_check_generic(struct pt_regs *regs)
780 {
781 	unsigned long reason = regs->msr;
782 
783 	printk("Machine check in kernel mode.\n");
784 	printk("Caused by (from SRR1=%lx): ", reason);
785 	switch (reason & 0x601F0000) {
786 	case 0x80000:
787 		pr_cont("Machine check signal\n");
788 		break;
789 	case 0:		/* for 601 */
790 	case 0x40000:
791 	case 0x140000:	/* 7450 MSS error and TEA */
792 		pr_cont("Transfer error ack signal\n");
793 		break;
794 	case 0x20000:
795 		pr_cont("Data parity error signal\n");
796 		break;
797 	case 0x10000:
798 		pr_cont("Address parity error signal\n");
799 		break;
800 	case 0x20000000:
801 		pr_cont("L1 Data Cache error\n");
802 		break;
803 	case 0x40000000:
804 		pr_cont("L1 Instruction Cache error\n");
805 		break;
806 	case 0x00100000:
807 		pr_cont("L2 data cache parity error\n");
808 		break;
809 	default:
810 		pr_cont("Unknown values in msr\n");
811 	}
812 	return 0;
813 }
814 #endif /* everything else */
815 
816 void machine_check_exception(struct pt_regs *regs)
817 {
818 	int recover = 0;
819 	bool nested = in_nmi();
820 	if (!nested)
821 		nmi_enter();
822 
823 	__this_cpu_inc(irq_stat.mce_exceptions);
824 
825 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
826 
827 	/* See if any machine dependent calls. In theory, we would want
828 	 * to call the CPU first, and call the ppc_md. one if the CPU
829 	 * one returns a positive number. However there is existing code
830 	 * that assumes the board gets a first chance, so let's keep it
831 	 * that way for now and fix things later. --BenH.
832 	 */
833 	if (ppc_md.machine_check_exception)
834 		recover = ppc_md.machine_check_exception(regs);
835 	else if (cur_cpu_spec->machine_check)
836 		recover = cur_cpu_spec->machine_check(regs);
837 
838 	if (recover > 0)
839 		goto bail;
840 
841 	if (debugger_fault_handler(regs))
842 		goto bail;
843 
844 	if (check_io_access(regs))
845 		goto bail;
846 
847 	if (!nested)
848 		nmi_exit();
849 
850 	die("Machine check", regs, SIGBUS);
851 
852 	/* Must die if the interrupt is not recoverable */
853 	if (!(regs->msr & MSR_RI))
854 		nmi_panic(regs, "Unrecoverable Machine check");
855 
856 	return;
857 
858 bail:
859 	if (!nested)
860 		nmi_exit();
861 }
862 
863 void SMIException(struct pt_regs *regs)
864 {
865 	die("System Management Interrupt", regs, SIGABRT);
866 }
867 
868 #ifdef CONFIG_VSX
869 static void p9_hmi_special_emu(struct pt_regs *regs)
870 {
871 	unsigned int ra, rb, t, i, sel, instr, rc;
872 	const void __user *addr;
873 	u8 vbuf[16], *vdst;
874 	unsigned long ea, msr, msr_mask;
875 	bool swap;
876 
877 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
878 		return;
879 
880 	/*
881 	 * lxvb16x	opcode: 0x7c0006d8
882 	 * lxvd2x	opcode: 0x7c000698
883 	 * lxvh8x	opcode: 0x7c000658
884 	 * lxvw4x	opcode: 0x7c000618
885 	 */
886 	if ((instr & 0xfc00073e) != 0x7c000618) {
887 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
888 			 " instr=%08x\n",
889 			 smp_processor_id(), current->comm, current->pid,
890 			 regs->nip, instr);
891 		return;
892 	}
893 
894 	/* Grab vector registers into the task struct */
895 	msr = regs->msr; /* Grab msr before we flush the bits */
896 	flush_vsx_to_thread(current);
897 	enable_kernel_altivec();
898 
899 	/*
900 	 * Is userspace running with a different endian (this is rare but
901 	 * not impossible)
902 	 */
903 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
904 
905 	/* Decode the instruction */
906 	ra = (instr >> 16) & 0x1f;
907 	rb = (instr >> 11) & 0x1f;
908 	t = (instr >> 21) & 0x1f;
909 	if (instr & 1)
910 		vdst = (u8 *)&current->thread.vr_state.vr[t];
911 	else
912 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
913 
914 	/* Grab the vector address */
915 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
916 	if (is_32bit_task())
917 		ea &= 0xfffffffful;
918 	addr = (__force const void __user *)ea;
919 
920 	/* Check it */
921 	if (!access_ok(addr, 16)) {
922 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
923 			 " instr=%08x addr=%016lx\n",
924 			 smp_processor_id(), current->comm, current->pid,
925 			 regs->nip, instr, (unsigned long)addr);
926 		return;
927 	}
928 
929 	/* Read the vector */
930 	rc = 0;
931 	if ((unsigned long)addr & 0xfUL)
932 		/* unaligned case */
933 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
934 	else
935 		__get_user_atomic_128_aligned(vbuf, addr, rc);
936 	if (rc) {
937 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
938 			 " instr=%08x addr=%016lx\n",
939 			 smp_processor_id(), current->comm, current->pid,
940 			 regs->nip, instr, (unsigned long)addr);
941 		return;
942 	}
943 
944 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
945 		 " instr=%08x addr=%016lx\n",
946 		 smp_processor_id(), current->comm, current->pid, regs->nip,
947 		 instr, (unsigned long) addr);
948 
949 	/* Grab instruction "selector" */
950 	sel = (instr >> 6) & 3;
951 
952 	/*
953 	 * Check to make sure the facility is actually enabled. This
954 	 * could happen if we get a false positive hit.
955 	 *
956 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
957 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
958 	 */
959 	msr_mask = MSR_VSX;
960 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
961 		msr_mask = MSR_VEC;
962 	if (!(msr & msr_mask)) {
963 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
964 			 " instr=%08x msr:%016lx\n",
965 			 smp_processor_id(), current->comm, current->pid,
966 			 regs->nip, instr, msr);
967 		return;
968 	}
969 
970 	/* Do logging here before we modify sel based on endian */
971 	switch (sel) {
972 	case 0:	/* lxvw4x */
973 		PPC_WARN_EMULATED(lxvw4x, regs);
974 		break;
975 	case 1: /* lxvh8x */
976 		PPC_WARN_EMULATED(lxvh8x, regs);
977 		break;
978 	case 2: /* lxvd2x */
979 		PPC_WARN_EMULATED(lxvd2x, regs);
980 		break;
981 	case 3: /* lxvb16x */
982 		PPC_WARN_EMULATED(lxvb16x, regs);
983 		break;
984 	}
985 
986 #ifdef __LITTLE_ENDIAN__
987 	/*
988 	 * An LE kernel stores the vector in the task struct as an LE
989 	 * byte array (effectively swapping both the components and
990 	 * the content of the components). Those instructions expect
991 	 * the components to remain in ascending address order, so we
992 	 * swap them back.
993 	 *
994 	 * If we are running a BE user space, the expectation is that
995 	 * of a simple memcpy, so forcing the emulation to look like
996 	 * a lxvb16x should do the trick.
997 	 */
998 	if (swap)
999 		sel = 3;
1000 
1001 	switch (sel) {
1002 	case 0:	/* lxvw4x */
1003 		for (i = 0; i < 4; i++)
1004 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1005 		break;
1006 	case 1: /* lxvh8x */
1007 		for (i = 0; i < 8; i++)
1008 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1009 		break;
1010 	case 2: /* lxvd2x */
1011 		for (i = 0; i < 2; i++)
1012 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1013 		break;
1014 	case 3: /* lxvb16x */
1015 		for (i = 0; i < 16; i++)
1016 			vdst[i] = vbuf[15-i];
1017 		break;
1018 	}
1019 #else /* __LITTLE_ENDIAN__ */
1020 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1021 	if (!swap)
1022 		sel = 3;
1023 
1024 	/* Otherwise, we need to swap the content of the components */
1025 	switch (sel) {
1026 	case 0:	/* lxvw4x */
1027 		for (i = 0; i < 4; i++)
1028 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1029 		break;
1030 	case 1: /* lxvh8x */
1031 		for (i = 0; i < 8; i++)
1032 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1033 		break;
1034 	case 2: /* lxvd2x */
1035 		for (i = 0; i < 2; i++)
1036 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1037 		break;
1038 	case 3: /* lxvb16x */
1039 		memcpy(vdst, vbuf, 16);
1040 		break;
1041 	}
1042 #endif /* !__LITTLE_ENDIAN__ */
1043 
1044 	/* Go to next instruction */
1045 	regs->nip += 4;
1046 }
1047 #endif /* CONFIG_VSX */
1048 
1049 void handle_hmi_exception(struct pt_regs *regs)
1050 {
1051 	struct pt_regs *old_regs;
1052 
1053 	old_regs = set_irq_regs(regs);
1054 	irq_enter();
1055 
1056 #ifdef CONFIG_VSX
1057 	/* Real mode flagged P9 special emu is needed */
1058 	if (local_paca->hmi_p9_special_emu) {
1059 		local_paca->hmi_p9_special_emu = 0;
1060 
1061 		/*
1062 		 * We don't want to take page faults while doing the
1063 		 * emulation, we just replay the instruction if necessary.
1064 		 */
1065 		pagefault_disable();
1066 		p9_hmi_special_emu(regs);
1067 		pagefault_enable();
1068 	}
1069 #endif /* CONFIG_VSX */
1070 
1071 	if (ppc_md.handle_hmi_exception)
1072 		ppc_md.handle_hmi_exception(regs);
1073 
1074 	irq_exit();
1075 	set_irq_regs(old_regs);
1076 }
1077 
1078 void unknown_exception(struct pt_regs *regs)
1079 {
1080 	enum ctx_state prev_state = exception_enter();
1081 
1082 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1083 	       regs->nip, regs->msr, regs->trap);
1084 
1085 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1086 
1087 	exception_exit(prev_state);
1088 }
1089 
1090 void instruction_breakpoint_exception(struct pt_regs *regs)
1091 {
1092 	enum ctx_state prev_state = exception_enter();
1093 
1094 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1095 					5, SIGTRAP) == NOTIFY_STOP)
1096 		goto bail;
1097 	if (debugger_iabr_match(regs))
1098 		goto bail;
1099 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1100 
1101 bail:
1102 	exception_exit(prev_state);
1103 }
1104 
1105 void RunModeException(struct pt_regs *regs)
1106 {
1107 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1108 }
1109 
1110 void single_step_exception(struct pt_regs *regs)
1111 {
1112 	enum ctx_state prev_state = exception_enter();
1113 
1114 	clear_single_step(regs);
1115 	clear_br_trace(regs);
1116 
1117 	if (kprobe_post_handler(regs))
1118 		return;
1119 
1120 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1121 					5, SIGTRAP) == NOTIFY_STOP)
1122 		goto bail;
1123 	if (debugger_sstep(regs))
1124 		goto bail;
1125 
1126 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1127 
1128 bail:
1129 	exception_exit(prev_state);
1130 }
1131 NOKPROBE_SYMBOL(single_step_exception);
1132 
1133 /*
1134  * After we have successfully emulated an instruction, we have to
1135  * check if the instruction was being single-stepped, and if so,
1136  * pretend we got a single-step exception.  This was pointed out
1137  * by Kumar Gala.  -- paulus
1138  */
1139 static void emulate_single_step(struct pt_regs *regs)
1140 {
1141 	if (single_stepping(regs))
1142 		single_step_exception(regs);
1143 }
1144 
1145 static inline int __parse_fpscr(unsigned long fpscr)
1146 {
1147 	int ret = FPE_FLTUNK;
1148 
1149 	/* Invalid operation */
1150 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1151 		ret = FPE_FLTINV;
1152 
1153 	/* Overflow */
1154 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1155 		ret = FPE_FLTOVF;
1156 
1157 	/* Underflow */
1158 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1159 		ret = FPE_FLTUND;
1160 
1161 	/* Divide by zero */
1162 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1163 		ret = FPE_FLTDIV;
1164 
1165 	/* Inexact result */
1166 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1167 		ret = FPE_FLTRES;
1168 
1169 	return ret;
1170 }
1171 
1172 static void parse_fpe(struct pt_regs *regs)
1173 {
1174 	int code = 0;
1175 
1176 	flush_fp_to_thread(current);
1177 
1178 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1179 
1180 	_exception(SIGFPE, regs, code, regs->nip);
1181 }
1182 
1183 /*
1184  * Illegal instruction emulation support.  Originally written to
1185  * provide the PVR to user applications using the mfspr rd, PVR.
1186  * Return non-zero if we can't emulate, or -EFAULT if the associated
1187  * memory access caused an access fault.  Return zero on success.
1188  *
1189  * There are a couple of ways to do this, either "decode" the instruction
1190  * or directly match lots of bits.  In this case, matching lots of
1191  * bits is faster and easier.
1192  *
1193  */
1194 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1195 {
1196 	u8 rT = (instword >> 21) & 0x1f;
1197 	u8 rA = (instword >> 16) & 0x1f;
1198 	u8 NB_RB = (instword >> 11) & 0x1f;
1199 	u32 num_bytes;
1200 	unsigned long EA;
1201 	int pos = 0;
1202 
1203 	/* Early out if we are an invalid form of lswx */
1204 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1205 		if ((rT == rA) || (rT == NB_RB))
1206 			return -EINVAL;
1207 
1208 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1209 
1210 	switch (instword & PPC_INST_STRING_MASK) {
1211 		case PPC_INST_LSWX:
1212 		case PPC_INST_STSWX:
1213 			EA += NB_RB;
1214 			num_bytes = regs->xer & 0x7f;
1215 			break;
1216 		case PPC_INST_LSWI:
1217 		case PPC_INST_STSWI:
1218 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1219 			break;
1220 		default:
1221 			return -EINVAL;
1222 	}
1223 
1224 	while (num_bytes != 0)
1225 	{
1226 		u8 val;
1227 		u32 shift = 8 * (3 - (pos & 0x3));
1228 
1229 		/* if process is 32-bit, clear upper 32 bits of EA */
1230 		if ((regs->msr & MSR_64BIT) == 0)
1231 			EA &= 0xFFFFFFFF;
1232 
1233 		switch ((instword & PPC_INST_STRING_MASK)) {
1234 			case PPC_INST_LSWX:
1235 			case PPC_INST_LSWI:
1236 				if (get_user(val, (u8 __user *)EA))
1237 					return -EFAULT;
1238 				/* first time updating this reg,
1239 				 * zero it out */
1240 				if (pos == 0)
1241 					regs->gpr[rT] = 0;
1242 				regs->gpr[rT] |= val << shift;
1243 				break;
1244 			case PPC_INST_STSWI:
1245 			case PPC_INST_STSWX:
1246 				val = regs->gpr[rT] >> shift;
1247 				if (put_user(val, (u8 __user *)EA))
1248 					return -EFAULT;
1249 				break;
1250 		}
1251 		/* move EA to next address */
1252 		EA += 1;
1253 		num_bytes--;
1254 
1255 		/* manage our position within the register */
1256 		if (++pos == 4) {
1257 			pos = 0;
1258 			if (++rT == 32)
1259 				rT = 0;
1260 		}
1261 	}
1262 
1263 	return 0;
1264 }
1265 
1266 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1267 {
1268 	u32 ra,rs;
1269 	unsigned long tmp;
1270 
1271 	ra = (instword >> 16) & 0x1f;
1272 	rs = (instword >> 21) & 0x1f;
1273 
1274 	tmp = regs->gpr[rs];
1275 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1276 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1277 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1278 	regs->gpr[ra] = tmp;
1279 
1280 	return 0;
1281 }
1282 
1283 static int emulate_isel(struct pt_regs *regs, u32 instword)
1284 {
1285 	u8 rT = (instword >> 21) & 0x1f;
1286 	u8 rA = (instword >> 16) & 0x1f;
1287 	u8 rB = (instword >> 11) & 0x1f;
1288 	u8 BC = (instword >> 6) & 0x1f;
1289 	u8 bit;
1290 	unsigned long tmp;
1291 
1292 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1293 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1294 
1295 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1296 
1297 	return 0;
1298 }
1299 
1300 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1301 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1302 {
1303         /* If we're emulating a load/store in an active transaction, we cannot
1304          * emulate it as the kernel operates in transaction suspended context.
1305          * We need to abort the transaction.  This creates a persistent TM
1306          * abort so tell the user what caused it with a new code.
1307 	 */
1308 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1309 		tm_enable();
1310 		tm_abort(cause);
1311 		return true;
1312 	}
1313 	return false;
1314 }
1315 #else
1316 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1317 {
1318 	return false;
1319 }
1320 #endif
1321 
1322 static int emulate_instruction(struct pt_regs *regs)
1323 {
1324 	u32 instword;
1325 	u32 rd;
1326 
1327 	if (!user_mode(regs))
1328 		return -EINVAL;
1329 	CHECK_FULL_REGS(regs);
1330 
1331 	if (get_user(instword, (u32 __user *)(regs->nip)))
1332 		return -EFAULT;
1333 
1334 	/* Emulate the mfspr rD, PVR. */
1335 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1336 		PPC_WARN_EMULATED(mfpvr, regs);
1337 		rd = (instword >> 21) & 0x1f;
1338 		regs->gpr[rd] = mfspr(SPRN_PVR);
1339 		return 0;
1340 	}
1341 
1342 	/* Emulating the dcba insn is just a no-op.  */
1343 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1344 		PPC_WARN_EMULATED(dcba, regs);
1345 		return 0;
1346 	}
1347 
1348 	/* Emulate the mcrxr insn.  */
1349 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1350 		int shift = (instword >> 21) & 0x1c;
1351 		unsigned long msk = 0xf0000000UL >> shift;
1352 
1353 		PPC_WARN_EMULATED(mcrxr, regs);
1354 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1355 		regs->xer &= ~0xf0000000UL;
1356 		return 0;
1357 	}
1358 
1359 	/* Emulate load/store string insn. */
1360 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1361 		if (tm_abort_check(regs,
1362 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1363 			return -EINVAL;
1364 		PPC_WARN_EMULATED(string, regs);
1365 		return emulate_string_inst(regs, instword);
1366 	}
1367 
1368 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1369 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1370 		PPC_WARN_EMULATED(popcntb, regs);
1371 		return emulate_popcntb_inst(regs, instword);
1372 	}
1373 
1374 	/* Emulate isel (Integer Select) instruction */
1375 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1376 		PPC_WARN_EMULATED(isel, regs);
1377 		return emulate_isel(regs, instword);
1378 	}
1379 
1380 	/* Emulate sync instruction variants */
1381 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1382 		PPC_WARN_EMULATED(sync, regs);
1383 		asm volatile("sync");
1384 		return 0;
1385 	}
1386 
1387 #ifdef CONFIG_PPC64
1388 	/* Emulate the mfspr rD, DSCR. */
1389 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1390 		PPC_INST_MFSPR_DSCR_USER) ||
1391 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1392 		PPC_INST_MFSPR_DSCR)) &&
1393 			cpu_has_feature(CPU_FTR_DSCR)) {
1394 		PPC_WARN_EMULATED(mfdscr, regs);
1395 		rd = (instword >> 21) & 0x1f;
1396 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1397 		return 0;
1398 	}
1399 	/* Emulate the mtspr DSCR, rD. */
1400 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1401 		PPC_INST_MTSPR_DSCR_USER) ||
1402 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1403 		PPC_INST_MTSPR_DSCR)) &&
1404 			cpu_has_feature(CPU_FTR_DSCR)) {
1405 		PPC_WARN_EMULATED(mtdscr, regs);
1406 		rd = (instword >> 21) & 0x1f;
1407 		current->thread.dscr = regs->gpr[rd];
1408 		current->thread.dscr_inherit = 1;
1409 		mtspr(SPRN_DSCR, current->thread.dscr);
1410 		return 0;
1411 	}
1412 #endif
1413 
1414 	return -EINVAL;
1415 }
1416 
1417 int is_valid_bugaddr(unsigned long addr)
1418 {
1419 	return is_kernel_addr(addr);
1420 }
1421 
1422 #ifdef CONFIG_MATH_EMULATION
1423 static int emulate_math(struct pt_regs *regs)
1424 {
1425 	int ret;
1426 	extern int do_mathemu(struct pt_regs *regs);
1427 
1428 	ret = do_mathemu(regs);
1429 	if (ret >= 0)
1430 		PPC_WARN_EMULATED(math, regs);
1431 
1432 	switch (ret) {
1433 	case 0:
1434 		emulate_single_step(regs);
1435 		return 0;
1436 	case 1: {
1437 			int code = 0;
1438 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1439 			_exception(SIGFPE, regs, code, regs->nip);
1440 			return 0;
1441 		}
1442 	case -EFAULT:
1443 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1444 		return 0;
1445 	}
1446 
1447 	return -1;
1448 }
1449 #else
1450 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1451 #endif
1452 
1453 void program_check_exception(struct pt_regs *regs)
1454 {
1455 	enum ctx_state prev_state = exception_enter();
1456 	unsigned int reason = get_reason(regs);
1457 
1458 	/* We can now get here via a FP Unavailable exception if the core
1459 	 * has no FPU, in that case the reason flags will be 0 */
1460 
1461 	if (reason & REASON_FP) {
1462 		/* IEEE FP exception */
1463 		parse_fpe(regs);
1464 		goto bail;
1465 	}
1466 	if (reason & REASON_TRAP) {
1467 		unsigned long bugaddr;
1468 		/* Debugger is first in line to stop recursive faults in
1469 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1470 		if (debugger_bpt(regs))
1471 			goto bail;
1472 
1473 		if (kprobe_handler(regs))
1474 			goto bail;
1475 
1476 		/* trap exception */
1477 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1478 				== NOTIFY_STOP)
1479 			goto bail;
1480 
1481 		bugaddr = regs->nip;
1482 		/*
1483 		 * Fixup bugaddr for BUG_ON() in real mode
1484 		 */
1485 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1486 			bugaddr += PAGE_OFFSET;
1487 
1488 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1489 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1490 			regs->nip += 4;
1491 			goto bail;
1492 		}
1493 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1494 		goto bail;
1495 	}
1496 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1497 	if (reason & REASON_TM) {
1498 		/* This is a TM "Bad Thing Exception" program check.
1499 		 * This occurs when:
1500 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1501 		 *    transition in TM states.
1502 		 * -  A trechkpt is attempted when transactional.
1503 		 * -  A treclaim is attempted when non transactional.
1504 		 * -  A tend is illegally attempted.
1505 		 * -  writing a TM SPR when transactional.
1506 		 *
1507 		 * If usermode caused this, it's done something illegal and
1508 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1509 		 * operand to distinguish from the instruction just being bad
1510 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1511 		 * illegal /placement/ of a valid instruction.
1512 		 */
1513 		if (user_mode(regs)) {
1514 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1515 			goto bail;
1516 		} else {
1517 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1518 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1519 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1520 			die("Unrecoverable exception", regs, SIGABRT);
1521 		}
1522 	}
1523 #endif
1524 
1525 	/*
1526 	 * If we took the program check in the kernel skip down to sending a
1527 	 * SIGILL. The subsequent cases all relate to emulating instructions
1528 	 * which we should only do for userspace. We also do not want to enable
1529 	 * interrupts for kernel faults because that might lead to further
1530 	 * faults, and loose the context of the original exception.
1531 	 */
1532 	if (!user_mode(regs))
1533 		goto sigill;
1534 
1535 	/* We restore the interrupt state now */
1536 	if (!arch_irq_disabled_regs(regs))
1537 		local_irq_enable();
1538 
1539 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1540 	 * but there seems to be a hardware bug on the 405GP (RevD)
1541 	 * that means ESR is sometimes set incorrectly - either to
1542 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1543 	 * hardware people - not sure if it can happen on any illegal
1544 	 * instruction or only on FP instructions, whether there is a
1545 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1546 	 */
1547 	if (!emulate_math(regs))
1548 		goto bail;
1549 
1550 	/* Try to emulate it if we should. */
1551 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1552 		switch (emulate_instruction(regs)) {
1553 		case 0:
1554 			regs->nip += 4;
1555 			emulate_single_step(regs);
1556 			goto bail;
1557 		case -EFAULT:
1558 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1559 			goto bail;
1560 		}
1561 	}
1562 
1563 sigill:
1564 	if (reason & REASON_PRIVILEGED)
1565 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1566 	else
1567 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1568 
1569 bail:
1570 	exception_exit(prev_state);
1571 }
1572 NOKPROBE_SYMBOL(program_check_exception);
1573 
1574 /*
1575  * This occurs when running in hypervisor mode on POWER6 or later
1576  * and an illegal instruction is encountered.
1577  */
1578 void emulation_assist_interrupt(struct pt_regs *regs)
1579 {
1580 	regs->msr |= REASON_ILLEGAL;
1581 	program_check_exception(regs);
1582 }
1583 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1584 
1585 void alignment_exception(struct pt_regs *regs)
1586 {
1587 	enum ctx_state prev_state = exception_enter();
1588 	int sig, code, fixed = 0;
1589 
1590 	/* We restore the interrupt state now */
1591 	if (!arch_irq_disabled_regs(regs))
1592 		local_irq_enable();
1593 
1594 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1595 		goto bail;
1596 
1597 	/* we don't implement logging of alignment exceptions */
1598 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1599 		fixed = fix_alignment(regs);
1600 
1601 	if (fixed == 1) {
1602 		regs->nip += 4;	/* skip over emulated instruction */
1603 		emulate_single_step(regs);
1604 		goto bail;
1605 	}
1606 
1607 	/* Operand address was bad */
1608 	if (fixed == -EFAULT) {
1609 		sig = SIGSEGV;
1610 		code = SEGV_ACCERR;
1611 	} else {
1612 		sig = SIGBUS;
1613 		code = BUS_ADRALN;
1614 	}
1615 	if (user_mode(regs))
1616 		_exception(sig, regs, code, regs->dar);
1617 	else
1618 		bad_page_fault(regs, regs->dar, sig);
1619 
1620 bail:
1621 	exception_exit(prev_state);
1622 }
1623 
1624 void StackOverflow(struct pt_regs *regs)
1625 {
1626 	pr_crit("Kernel stack overflow in process %s[%d], r1=%lx\n",
1627 		current->comm, task_pid_nr(current), regs->gpr[1]);
1628 	debugger(regs);
1629 	show_regs(regs);
1630 	panic("kernel stack overflow");
1631 }
1632 
1633 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1634 {
1635 	enum ctx_state prev_state = exception_enter();
1636 
1637 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1638 			  "%lx at %lx\n", regs->trap, regs->nip);
1639 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1640 
1641 	exception_exit(prev_state);
1642 }
1643 
1644 void altivec_unavailable_exception(struct pt_regs *regs)
1645 {
1646 	enum ctx_state prev_state = exception_enter();
1647 
1648 	if (user_mode(regs)) {
1649 		/* A user program has executed an altivec instruction,
1650 		   but this kernel doesn't support altivec. */
1651 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1652 		goto bail;
1653 	}
1654 
1655 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1656 			"%lx at %lx\n", regs->trap, regs->nip);
1657 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1658 
1659 bail:
1660 	exception_exit(prev_state);
1661 }
1662 
1663 void vsx_unavailable_exception(struct pt_regs *regs)
1664 {
1665 	if (user_mode(regs)) {
1666 		/* A user program has executed an vsx instruction,
1667 		   but this kernel doesn't support vsx. */
1668 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1669 		return;
1670 	}
1671 
1672 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1673 			"%lx at %lx\n", regs->trap, regs->nip);
1674 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1675 }
1676 
1677 #ifdef CONFIG_PPC64
1678 static void tm_unavailable(struct pt_regs *regs)
1679 {
1680 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1681 	if (user_mode(regs)) {
1682 		current->thread.load_tm++;
1683 		regs->msr |= MSR_TM;
1684 		tm_enable();
1685 		tm_restore_sprs(&current->thread);
1686 		return;
1687 	}
1688 #endif
1689 	pr_emerg("Unrecoverable TM Unavailable Exception "
1690 			"%lx at %lx\n", regs->trap, regs->nip);
1691 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1692 }
1693 
1694 void facility_unavailable_exception(struct pt_regs *regs)
1695 {
1696 	static char *facility_strings[] = {
1697 		[FSCR_FP_LG] = "FPU",
1698 		[FSCR_VECVSX_LG] = "VMX/VSX",
1699 		[FSCR_DSCR_LG] = "DSCR",
1700 		[FSCR_PM_LG] = "PMU SPRs",
1701 		[FSCR_BHRB_LG] = "BHRB",
1702 		[FSCR_TM_LG] = "TM",
1703 		[FSCR_EBB_LG] = "EBB",
1704 		[FSCR_TAR_LG] = "TAR",
1705 		[FSCR_MSGP_LG] = "MSGP",
1706 		[FSCR_SCV_LG] = "SCV",
1707 	};
1708 	char *facility = "unknown";
1709 	u64 value;
1710 	u32 instword, rd;
1711 	u8 status;
1712 	bool hv;
1713 
1714 	hv = (TRAP(regs) == 0xf80);
1715 	if (hv)
1716 		value = mfspr(SPRN_HFSCR);
1717 	else
1718 		value = mfspr(SPRN_FSCR);
1719 
1720 	status = value >> 56;
1721 	if ((hv || status >= 2) &&
1722 	    (status < ARRAY_SIZE(facility_strings)) &&
1723 	    facility_strings[status])
1724 		facility = facility_strings[status];
1725 
1726 	/* We should not have taken this interrupt in kernel */
1727 	if (!user_mode(regs)) {
1728 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1729 			 facility, status, regs->nip);
1730 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1731 	}
1732 
1733 	/* We restore the interrupt state now */
1734 	if (!arch_irq_disabled_regs(regs))
1735 		local_irq_enable();
1736 
1737 	if (status == FSCR_DSCR_LG) {
1738 		/*
1739 		 * User is accessing the DSCR register using the problem
1740 		 * state only SPR number (0x03) either through a mfspr or
1741 		 * a mtspr instruction. If it is a write attempt through
1742 		 * a mtspr, then we set the inherit bit. This also allows
1743 		 * the user to write or read the register directly in the
1744 		 * future by setting via the FSCR DSCR bit. But in case it
1745 		 * is a read DSCR attempt through a mfspr instruction, we
1746 		 * just emulate the instruction instead. This code path will
1747 		 * always emulate all the mfspr instructions till the user
1748 		 * has attempted at least one mtspr instruction. This way it
1749 		 * preserves the same behaviour when the user is accessing
1750 		 * the DSCR through privilege level only SPR number (0x11)
1751 		 * which is emulated through illegal instruction exception.
1752 		 * We always leave HFSCR DSCR set.
1753 		 */
1754 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1755 			pr_err("Failed to fetch the user instruction\n");
1756 			return;
1757 		}
1758 
1759 		/* Write into DSCR (mtspr 0x03, RS) */
1760 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1761 				== PPC_INST_MTSPR_DSCR_USER) {
1762 			rd = (instword >> 21) & 0x1f;
1763 			current->thread.dscr = regs->gpr[rd];
1764 			current->thread.dscr_inherit = 1;
1765 			current->thread.fscr |= FSCR_DSCR;
1766 			mtspr(SPRN_FSCR, current->thread.fscr);
1767 		}
1768 
1769 		/* Read from DSCR (mfspr RT, 0x03) */
1770 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1771 				== PPC_INST_MFSPR_DSCR_USER) {
1772 			if (emulate_instruction(regs)) {
1773 				pr_err("DSCR based mfspr emulation failed\n");
1774 				return;
1775 			}
1776 			regs->nip += 4;
1777 			emulate_single_step(regs);
1778 		}
1779 		return;
1780 	}
1781 
1782 	if (status == FSCR_TM_LG) {
1783 		/*
1784 		 * If we're here then the hardware is TM aware because it
1785 		 * generated an exception with FSRM_TM set.
1786 		 *
1787 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1788 		 * told us not to do TM, or the kernel is not built with TM
1789 		 * support.
1790 		 *
1791 		 * If both of those things are true, then userspace can spam the
1792 		 * console by triggering the printk() below just by continually
1793 		 * doing tbegin (or any TM instruction). So in that case just
1794 		 * send the process a SIGILL immediately.
1795 		 */
1796 		if (!cpu_has_feature(CPU_FTR_TM))
1797 			goto out;
1798 
1799 		tm_unavailable(regs);
1800 		return;
1801 	}
1802 
1803 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1804 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1805 
1806 out:
1807 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1808 }
1809 #endif
1810 
1811 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1812 
1813 void fp_unavailable_tm(struct pt_regs *regs)
1814 {
1815 	/* Note:  This does not handle any kind of FP laziness. */
1816 
1817 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1818 		 regs->nip, regs->msr);
1819 
1820         /* We can only have got here if the task started using FP after
1821          * beginning the transaction.  So, the transactional regs are just a
1822          * copy of the checkpointed ones.  But, we still need to recheckpoint
1823          * as we're enabling FP for the process; it will return, abort the
1824          * transaction, and probably retry but now with FP enabled.  So the
1825          * checkpointed FP registers need to be loaded.
1826 	 */
1827 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1828 
1829 	/*
1830 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1831 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1832 	 *
1833 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1834 	 * recheckpoint.
1835 	 */
1836 
1837 	/* Enable FP for the task: */
1838 	current->thread.load_fp = 1;
1839 
1840 	/*
1841 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1842 	 */
1843 	tm_recheckpoint(&current->thread);
1844 }
1845 
1846 void altivec_unavailable_tm(struct pt_regs *regs)
1847 {
1848 	/* See the comments in fp_unavailable_tm().  This function operates
1849 	 * the same way.
1850 	 */
1851 
1852 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1853 		 "MSR=%lx\n",
1854 		 regs->nip, regs->msr);
1855 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1856 	current->thread.load_vec = 1;
1857 	tm_recheckpoint(&current->thread);
1858 	current->thread.used_vr = 1;
1859 }
1860 
1861 void vsx_unavailable_tm(struct pt_regs *regs)
1862 {
1863 	/* See the comments in fp_unavailable_tm().  This works similarly,
1864 	 * though we're loading both FP and VEC registers in here.
1865 	 *
1866 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1867 	 * regs.  Either way, set MSR_VSX.
1868 	 */
1869 
1870 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1871 		 "MSR=%lx\n",
1872 		 regs->nip, regs->msr);
1873 
1874 	current->thread.used_vsr = 1;
1875 
1876 	/* This reclaims FP and/or VR regs if they're already enabled */
1877 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1878 
1879 	current->thread.load_vec = 1;
1880 	current->thread.load_fp = 1;
1881 
1882 	tm_recheckpoint(&current->thread);
1883 }
1884 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1885 
1886 void performance_monitor_exception(struct pt_regs *regs)
1887 {
1888 	__this_cpu_inc(irq_stat.pmu_irqs);
1889 
1890 	perf_irq(regs);
1891 }
1892 
1893 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1894 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1895 {
1896 	int changed = 0;
1897 	/*
1898 	 * Determine the cause of the debug event, clear the
1899 	 * event flags and send a trap to the handler. Torez
1900 	 */
1901 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1902 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1903 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1904 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1905 #endif
1906 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1907 			     5);
1908 		changed |= 0x01;
1909 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1910 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1911 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1912 			     6);
1913 		changed |= 0x01;
1914 	}  else if (debug_status & DBSR_IAC1) {
1915 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1916 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1917 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1918 			     1);
1919 		changed |= 0x01;
1920 	}  else if (debug_status & DBSR_IAC2) {
1921 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1922 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1923 			     2);
1924 		changed |= 0x01;
1925 	}  else if (debug_status & DBSR_IAC3) {
1926 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1927 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1928 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1929 			     3);
1930 		changed |= 0x01;
1931 	}  else if (debug_status & DBSR_IAC4) {
1932 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1933 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1934 			     4);
1935 		changed |= 0x01;
1936 	}
1937 	/*
1938 	 * At the point this routine was called, the MSR(DE) was turned off.
1939 	 * Check all other debug flags and see if that bit needs to be turned
1940 	 * back on or not.
1941 	 */
1942 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1943 			       current->thread.debug.dbcr1))
1944 		regs->msr |= MSR_DE;
1945 	else
1946 		/* Make sure the IDM flag is off */
1947 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1948 
1949 	if (changed & 0x01)
1950 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1951 }
1952 
1953 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1954 {
1955 	current->thread.debug.dbsr = debug_status;
1956 
1957 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1958 	 * on server, it stops on the target of the branch. In order to simulate
1959 	 * the server behaviour, we thus restart right away with a single step
1960 	 * instead of stopping here when hitting a BT
1961 	 */
1962 	if (debug_status & DBSR_BT) {
1963 		regs->msr &= ~MSR_DE;
1964 
1965 		/* Disable BT */
1966 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1967 		/* Clear the BT event */
1968 		mtspr(SPRN_DBSR, DBSR_BT);
1969 
1970 		/* Do the single step trick only when coming from userspace */
1971 		if (user_mode(regs)) {
1972 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1973 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1974 			regs->msr |= MSR_DE;
1975 			return;
1976 		}
1977 
1978 		if (kprobe_post_handler(regs))
1979 			return;
1980 
1981 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1982 			       5, SIGTRAP) == NOTIFY_STOP) {
1983 			return;
1984 		}
1985 		if (debugger_sstep(regs))
1986 			return;
1987 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1988 		regs->msr &= ~MSR_DE;
1989 
1990 		/* Disable instruction completion */
1991 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
1992 		/* Clear the instruction completion event */
1993 		mtspr(SPRN_DBSR, DBSR_IC);
1994 
1995 		if (kprobe_post_handler(regs))
1996 			return;
1997 
1998 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1999 			       5, SIGTRAP) == NOTIFY_STOP) {
2000 			return;
2001 		}
2002 
2003 		if (debugger_sstep(regs))
2004 			return;
2005 
2006 		if (user_mode(regs)) {
2007 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2008 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2009 					       current->thread.debug.dbcr1))
2010 				regs->msr |= MSR_DE;
2011 			else
2012 				/* Make sure the IDM bit is off */
2013 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2014 		}
2015 
2016 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2017 	} else
2018 		handle_debug(regs, debug_status);
2019 }
2020 NOKPROBE_SYMBOL(DebugException);
2021 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2022 
2023 #if !defined(CONFIG_TAU_INT)
2024 void TAUException(struct pt_regs *regs)
2025 {
2026 	printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
2027 	       regs->nip, regs->msr, regs->trap, print_tainted());
2028 }
2029 #endif /* CONFIG_INT_TAU */
2030 
2031 #ifdef CONFIG_ALTIVEC
2032 void altivec_assist_exception(struct pt_regs *regs)
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 void CacheLockingException(struct pt_regs *regs, unsigned long address,
2067 			   unsigned long error_code)
2068 {
2069 	/* We treat cache locking instructions from the user
2070 	 * as priv ops, in the future we could try to do
2071 	 * something smarter
2072 	 */
2073 	if (error_code & (ESR_DLK|ESR_ILK))
2074 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2075 	return;
2076 }
2077 #endif /* CONFIG_FSL_BOOKE */
2078 
2079 #ifdef CONFIG_SPE
2080 void SPEFloatingPointException(struct pt_regs *regs)
2081 {
2082 	extern int do_spe_mathemu(struct pt_regs *regs);
2083 	unsigned long spefscr;
2084 	int fpexc_mode;
2085 	int code = FPE_FLTUNK;
2086 	int err;
2087 
2088 	/* We restore the interrupt state now */
2089 	if (!arch_irq_disabled_regs(regs))
2090 		local_irq_enable();
2091 
2092 	flush_spe_to_thread(current);
2093 
2094 	spefscr = current->thread.spefscr;
2095 	fpexc_mode = current->thread.fpexc_mode;
2096 
2097 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2098 		code = FPE_FLTOVF;
2099 	}
2100 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2101 		code = FPE_FLTUND;
2102 	}
2103 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2104 		code = FPE_FLTDIV;
2105 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2106 		code = FPE_FLTINV;
2107 	}
2108 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2109 		code = FPE_FLTRES;
2110 
2111 	err = do_spe_mathemu(regs);
2112 	if (err == 0) {
2113 		regs->nip += 4;		/* skip emulated instruction */
2114 		emulate_single_step(regs);
2115 		return;
2116 	}
2117 
2118 	if (err == -EFAULT) {
2119 		/* got an error reading the instruction */
2120 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2121 	} else if (err == -EINVAL) {
2122 		/* didn't recognize the instruction */
2123 		printk(KERN_ERR "unrecognized spe instruction "
2124 		       "in %s at %lx\n", current->comm, regs->nip);
2125 	} else {
2126 		_exception(SIGFPE, regs, code, regs->nip);
2127 	}
2128 
2129 	return;
2130 }
2131 
2132 void SPEFloatingPointRoundException(struct pt_regs *regs)
2133 {
2134 	extern int speround_handler(struct pt_regs *regs);
2135 	int err;
2136 
2137 	/* We restore the interrupt state now */
2138 	if (!arch_irq_disabled_regs(regs))
2139 		local_irq_enable();
2140 
2141 	preempt_disable();
2142 	if (regs->msr & MSR_SPE)
2143 		giveup_spe(current);
2144 	preempt_enable();
2145 
2146 	regs->nip -= 4;
2147 	err = speround_handler(regs);
2148 	if (err == 0) {
2149 		regs->nip += 4;		/* skip emulated instruction */
2150 		emulate_single_step(regs);
2151 		return;
2152 	}
2153 
2154 	if (err == -EFAULT) {
2155 		/* got an error reading the instruction */
2156 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2157 	} else if (err == -EINVAL) {
2158 		/* didn't recognize the instruction */
2159 		printk(KERN_ERR "unrecognized spe instruction "
2160 		       "in %s at %lx\n", current->comm, regs->nip);
2161 	} else {
2162 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2163 		return;
2164 	}
2165 }
2166 #endif
2167 
2168 /*
2169  * We enter here if we get an unrecoverable exception, that is, one
2170  * that happened at a point where the RI (recoverable interrupt) bit
2171  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2172  * we therefore lost state by taking this exception.
2173  */
2174 void unrecoverable_exception(struct pt_regs *regs)
2175 {
2176 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2177 		 regs->trap, regs->nip, regs->msr);
2178 	die("Unrecoverable exception", regs, SIGABRT);
2179 }
2180 NOKPROBE_SYMBOL(unrecoverable_exception);
2181 
2182 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2183 /*
2184  * Default handler for a Watchdog exception,
2185  * spins until a reboot occurs
2186  */
2187 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2188 {
2189 	/* Generic WatchdogHandler, implement your own */
2190 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2191 	return;
2192 }
2193 
2194 void WatchdogException(struct pt_regs *regs)
2195 {
2196 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2197 	WatchdogHandler(regs);
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 void kernel_bad_stack(struct pt_regs *regs)
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 NOKPROBE_SYMBOL(kernel_bad_stack);
2212 
2213 void __init trap_init(void)
2214 {
2215 }
2216 
2217 
2218 #ifdef CONFIG_PPC_EMULATED_STATS
2219 
2220 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2221 
2222 struct ppc_emulated ppc_emulated = {
2223 #ifdef CONFIG_ALTIVEC
2224 	WARN_EMULATED_SETUP(altivec),
2225 #endif
2226 	WARN_EMULATED_SETUP(dcba),
2227 	WARN_EMULATED_SETUP(dcbz),
2228 	WARN_EMULATED_SETUP(fp_pair),
2229 	WARN_EMULATED_SETUP(isel),
2230 	WARN_EMULATED_SETUP(mcrxr),
2231 	WARN_EMULATED_SETUP(mfpvr),
2232 	WARN_EMULATED_SETUP(multiple),
2233 	WARN_EMULATED_SETUP(popcntb),
2234 	WARN_EMULATED_SETUP(spe),
2235 	WARN_EMULATED_SETUP(string),
2236 	WARN_EMULATED_SETUP(sync),
2237 	WARN_EMULATED_SETUP(unaligned),
2238 #ifdef CONFIG_MATH_EMULATION
2239 	WARN_EMULATED_SETUP(math),
2240 #endif
2241 #ifdef CONFIG_VSX
2242 	WARN_EMULATED_SETUP(vsx),
2243 #endif
2244 #ifdef CONFIG_PPC64
2245 	WARN_EMULATED_SETUP(mfdscr),
2246 	WARN_EMULATED_SETUP(mtdscr),
2247 	WARN_EMULATED_SETUP(lq_stq),
2248 	WARN_EMULATED_SETUP(lxvw4x),
2249 	WARN_EMULATED_SETUP(lxvh8x),
2250 	WARN_EMULATED_SETUP(lxvd2x),
2251 	WARN_EMULATED_SETUP(lxvb16x),
2252 #endif
2253 };
2254 
2255 u32 ppc_warn_emulated;
2256 
2257 void ppc_warn_emulated_print(const char *type)
2258 {
2259 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2260 			    type);
2261 }
2262 
2263 static int __init ppc_warn_emulated_init(void)
2264 {
2265 	struct dentry *dir, *d;
2266 	unsigned int i;
2267 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2268 
2269 	if (!powerpc_debugfs_root)
2270 		return -ENODEV;
2271 
2272 	dir = debugfs_create_dir("emulated_instructions",
2273 				 powerpc_debugfs_root);
2274 	if (!dir)
2275 		return -ENOMEM;
2276 
2277 	d = debugfs_create_u32("do_warn", 0644, dir,
2278 			       &ppc_warn_emulated);
2279 	if (!d)
2280 		goto fail;
2281 
2282 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) {
2283 		d = debugfs_create_u32(entries[i].name, 0644, dir,
2284 				       (u32 *)&entries[i].val.counter);
2285 		if (!d)
2286 			goto fail;
2287 	}
2288 
2289 	return 0;
2290 
2291 fail:
2292 	debugfs_remove_recursive(dir);
2293 	return -ENOMEM;
2294 }
2295 
2296 device_initcall(ppc_warn_emulated_init);
2297 
2298 #endif /* CONFIG_PPC_EMULATED_STATS */
2299