xref: /linux/arch/powerpc/kernel/traps.c (revision cbac924200b838cfb8d8b1415113d788089dc50b)
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 #include <linux/debugfs.h>
41 
42 #include <asm/emulated_ops.h>
43 #include <linux/uaccess.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 #include <asm/disassemble.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 	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 	make_task_dead(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 	/*
344 	 * Must not enable interrupts even for user-mode exception, because
345 	 * this can be called from machine check, which may be a NMI or IRQ
346 	 * which don't like interrupts being enabled. Could check for
347 	 * in_hardirq || in_nmi perhaps, but there doesn't seem to be a good
348 	 * reason why _exception() should enable irqs for an exception handler,
349 	 * the handlers themselves do that directly.
350 	 */
351 
352 	show_signal_msg(signr, regs, code, addr);
353 
354 	current->thread.trap_nr = code;
355 
356 	return true;
357 }
358 
359 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key)
360 {
361 	if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr))
362 		return;
363 
364 	force_sig_pkuerr((void __user *) addr, key);
365 }
366 
367 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
368 {
369 	if (!exception_common(signr, regs, code, addr))
370 		return;
371 
372 	force_sig_fault(signr, code, (void __user *)addr);
373 }
374 
375 /*
376  * The interrupt architecture has a quirk in that the HV interrupts excluding
377  * the NMIs (0x100 and 0x200) do not clear MSR[RI] at entry. The first thing
378  * that an interrupt handler must do is save off a GPR into a scratch register,
379  * and all interrupts on POWERNV (HV=1) use the HSPRG1 register as scratch.
380  * Therefore an NMI can clobber an HV interrupt's live HSPRG1 without noticing
381  * that it is non-reentrant, which leads to random data corruption.
382  *
383  * The solution is for NMI interrupts in HV mode to check if they originated
384  * from these critical HV interrupt regions. If so, then mark them not
385  * recoverable.
386  *
387  * An alternative would be for HV NMIs to use SPRG for scratch to avoid the
388  * HSPRG1 clobber, however this would cause guest SPRG to be clobbered. Linux
389  * guests should always have MSR[RI]=0 when its scratch SPRG is in use, so
390  * that would work. However any other guest OS that may have the SPRG live
391  * and MSR[RI]=1 could encounter silent corruption.
392  *
393  * Builds that do not support KVM could take this second option to increase
394  * the recoverability of NMIs.
395  */
396 void hv_nmi_check_nonrecoverable(struct pt_regs *regs)
397 {
398 #ifdef CONFIG_PPC_POWERNV
399 	unsigned long kbase = (unsigned long)_stext;
400 	unsigned long nip = regs->nip;
401 
402 	if (!(regs->msr & MSR_RI))
403 		return;
404 	if (!(regs->msr & MSR_HV))
405 		return;
406 	if (regs->msr & MSR_PR)
407 		return;
408 
409 	/*
410 	 * Now test if the interrupt has hit a range that may be using
411 	 * HSPRG1 without having RI=0 (i.e., an HSRR interrupt). The
412 	 * problem ranges all run un-relocated. Test real and virt modes
413 	 * at the same time by dropping the high bit of the nip (virt mode
414 	 * entry points still have the +0x4000 offset).
415 	 */
416 	nip &= ~0xc000000000000000ULL;
417 	if ((nip >= 0x500 && nip < 0x600) || (nip >= 0x4500 && nip < 0x4600))
418 		goto nonrecoverable;
419 	if ((nip >= 0x980 && nip < 0xa00) || (nip >= 0x4980 && nip < 0x4a00))
420 		goto nonrecoverable;
421 	if ((nip >= 0xe00 && nip < 0xec0) || (nip >= 0x4e00 && nip < 0x4ec0))
422 		goto nonrecoverable;
423 	if ((nip >= 0xf80 && nip < 0xfa0) || (nip >= 0x4f80 && nip < 0x4fa0))
424 		goto nonrecoverable;
425 
426 	/* Trampoline code runs un-relocated so subtract kbase. */
427 	if (nip >= (unsigned long)(start_real_trampolines - kbase) &&
428 			nip < (unsigned long)(end_real_trampolines - kbase))
429 		goto nonrecoverable;
430 	if (nip >= (unsigned long)(start_virt_trampolines - kbase) &&
431 			nip < (unsigned long)(end_virt_trampolines - kbase))
432 		goto nonrecoverable;
433 	return;
434 
435 nonrecoverable:
436 	regs_set_unrecoverable(regs);
437 #endif
438 }
439 DEFINE_INTERRUPT_HANDLER_NMI(system_reset_exception)
440 {
441 	unsigned long hsrr0, hsrr1;
442 	bool saved_hsrrs = false;
443 
444 	/*
445 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
446 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
447 	 * OPAL), so save them here and restore them before returning.
448 	 *
449 	 * Machine checks don't need to save HSRRs, as the real mode handler
450 	 * is careful to avoid them, and the regular handler is not delivered
451 	 * as an NMI.
452 	 */
453 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
454 		hsrr0 = mfspr(SPRN_HSRR0);
455 		hsrr1 = mfspr(SPRN_HSRR1);
456 		saved_hsrrs = true;
457 	}
458 
459 	hv_nmi_check_nonrecoverable(regs);
460 
461 	__this_cpu_inc(irq_stat.sreset_irqs);
462 
463 	/* See if any machine dependent calls */
464 	if (ppc_md.system_reset_exception) {
465 		if (ppc_md.system_reset_exception(regs))
466 			goto out;
467 	}
468 
469 	if (debugger(regs))
470 		goto out;
471 
472 	kmsg_dump(KMSG_DUMP_OOPS);
473 	/*
474 	 * A system reset is a request to dump, so we always send
475 	 * it through the crashdump code (if fadump or kdump are
476 	 * registered).
477 	 */
478 	crash_fadump(regs, "System Reset");
479 
480 	crash_kexec(regs);
481 
482 	/*
483 	 * We aren't the primary crash CPU. We need to send it
484 	 * to a holding pattern to avoid it ending up in the panic
485 	 * code.
486 	 */
487 	crash_kexec_secondary(regs);
488 
489 	/*
490 	 * No debugger or crash dump registered, print logs then
491 	 * panic.
492 	 */
493 	die("System Reset", regs, SIGABRT);
494 
495 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
496 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
497 	nmi_panic(regs, "System Reset");
498 
499 out:
500 #ifdef CONFIG_PPC_BOOK3S_64
501 	BUG_ON(get_paca()->in_nmi == 0);
502 	if (get_paca()->in_nmi > 1)
503 		die("Unrecoverable nested System Reset", regs, SIGABRT);
504 #endif
505 	/* Must die if the interrupt is not recoverable */
506 	if (regs_is_unrecoverable(regs)) {
507 		/* For the reason explained in die_mce, nmi_exit before die */
508 		nmi_exit();
509 		die("Unrecoverable System Reset", regs, SIGABRT);
510 	}
511 
512 	if (saved_hsrrs) {
513 		mtspr(SPRN_HSRR0, hsrr0);
514 		mtspr(SPRN_HSRR1, hsrr1);
515 	}
516 
517 	/* What should we do here? We could issue a shutdown or hard reset. */
518 
519 	return 0;
520 }
521 
522 /*
523  * I/O accesses can cause machine checks on powermacs.
524  * Check if the NIP corresponds to the address of a sync
525  * instruction for which there is an entry in the exception
526  * table.
527  *  -- paulus.
528  */
529 static inline int check_io_access(struct pt_regs *regs)
530 {
531 #ifdef CONFIG_PPC32
532 	unsigned long msr = regs->msr;
533 	const struct exception_table_entry *entry;
534 	unsigned int *nip = (unsigned int *)regs->nip;
535 
536 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
537 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
538 		/*
539 		 * Check that it's a sync instruction, or somewhere
540 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
541 		 * As the address is in the exception table
542 		 * we should be able to read the instr there.
543 		 * For the debug message, we look at the preceding
544 		 * load or store.
545 		 */
546 		if (*nip == PPC_RAW_NOP())
547 			nip -= 2;
548 		else if (*nip == PPC_RAW_ISYNC())
549 			--nip;
550 		if (*nip == PPC_RAW_SYNC() || get_op(*nip) == OP_TRAP) {
551 			unsigned int rb;
552 
553 			--nip;
554 			rb = (*nip >> 11) & 0x1f;
555 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
556 			       (*nip & 0x100)? "OUT to": "IN from",
557 			       regs->gpr[rb] - _IO_BASE, nip);
558 			regs_set_recoverable(regs);
559 			regs_set_return_ip(regs, extable_fixup(entry));
560 			return 1;
561 		}
562 	}
563 #endif /* CONFIG_PPC32 */
564 	return 0;
565 }
566 
567 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
568 /* On 4xx, the reason for the machine check or program exception
569    is in the ESR. */
570 #define get_reason(regs)	((regs)->esr)
571 #define REASON_FP		ESR_FP
572 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
573 #define REASON_PRIVILEGED	ESR_PPR
574 #define REASON_TRAP		ESR_PTR
575 #define REASON_PREFIXED		0
576 #define REASON_BOUNDARY		0
577 
578 /* single-step stuff */
579 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
580 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
581 #define clear_br_trace(regs)	do {} while(0)
582 #else
583 /* On non-4xx, the reason for the machine check or program
584    exception is in the MSR. */
585 #define get_reason(regs)	((regs)->msr)
586 #define REASON_TM		SRR1_PROGTM
587 #define REASON_FP		SRR1_PROGFPE
588 #define REASON_ILLEGAL		SRR1_PROGILL
589 #define REASON_PRIVILEGED	SRR1_PROGPRIV
590 #define REASON_TRAP		SRR1_PROGTRAP
591 #define REASON_PREFIXED		SRR1_PREFIXED
592 #define REASON_BOUNDARY		SRR1_BOUNDARY
593 
594 #define single_stepping(regs)	((regs)->msr & MSR_SE)
595 #define clear_single_step(regs)	(regs_set_return_msr((regs), (regs)->msr & ~MSR_SE))
596 #define clear_br_trace(regs)	(regs_set_return_msr((regs), (regs)->msr & ~MSR_BE))
597 #endif
598 
599 #define inst_length(reason)	(((reason) & REASON_PREFIXED) ? 8 : 4)
600 
601 #if defined(CONFIG_E500)
602 int machine_check_e500mc(struct pt_regs *regs)
603 {
604 	unsigned long mcsr = mfspr(SPRN_MCSR);
605 	unsigned long pvr = mfspr(SPRN_PVR);
606 	unsigned long reason = mcsr;
607 	int recoverable = 1;
608 
609 	if (reason & MCSR_LD) {
610 		recoverable = fsl_rio_mcheck_exception(regs);
611 		if (recoverable == 1)
612 			goto silent_out;
613 	}
614 
615 	printk("Machine check in kernel mode.\n");
616 	printk("Caused by (from MCSR=%lx): ", reason);
617 
618 	if (reason & MCSR_MCP)
619 		pr_cont("Machine Check Signal\n");
620 
621 	if (reason & MCSR_ICPERR) {
622 		pr_cont("Instruction Cache Parity Error\n");
623 
624 		/*
625 		 * This is recoverable by invalidating the i-cache.
626 		 */
627 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
628 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
629 			;
630 
631 		/*
632 		 * This will generally be accompanied by an instruction
633 		 * fetch error report -- only treat MCSR_IF as fatal
634 		 * if it wasn't due to an L1 parity error.
635 		 */
636 		reason &= ~MCSR_IF;
637 	}
638 
639 	if (reason & MCSR_DCPERR_MC) {
640 		pr_cont("Data Cache Parity Error\n");
641 
642 		/*
643 		 * In write shadow mode we auto-recover from the error, but it
644 		 * may still get logged and cause a machine check.  We should
645 		 * only treat the non-write shadow case as non-recoverable.
646 		 */
647 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
648 		 * is not implemented but L1 data cache always runs in write
649 		 * shadow mode. Hence on data cache parity errors HW will
650 		 * automatically invalidate the L1 Data Cache.
651 		 */
652 		if (PVR_VER(pvr) != PVR_VER_E6500) {
653 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
654 				recoverable = 0;
655 		}
656 	}
657 
658 	if (reason & MCSR_L2MMU_MHIT) {
659 		pr_cont("Hit on multiple TLB entries\n");
660 		recoverable = 0;
661 	}
662 
663 	if (reason & MCSR_NMI)
664 		pr_cont("Non-maskable interrupt\n");
665 
666 	if (reason & MCSR_IF) {
667 		pr_cont("Instruction Fetch Error Report\n");
668 		recoverable = 0;
669 	}
670 
671 	if (reason & MCSR_LD) {
672 		pr_cont("Load Error Report\n");
673 		recoverable = 0;
674 	}
675 
676 	if (reason & MCSR_ST) {
677 		pr_cont("Store Error Report\n");
678 		recoverable = 0;
679 	}
680 
681 	if (reason & MCSR_LDG) {
682 		pr_cont("Guarded Load Error Report\n");
683 		recoverable = 0;
684 	}
685 
686 	if (reason & MCSR_TLBSYNC)
687 		pr_cont("Simultaneous tlbsync operations\n");
688 
689 	if (reason & MCSR_BSL2_ERR) {
690 		pr_cont("Level 2 Cache Error\n");
691 		recoverable = 0;
692 	}
693 
694 	if (reason & MCSR_MAV) {
695 		u64 addr;
696 
697 		addr = mfspr(SPRN_MCAR);
698 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
699 
700 		pr_cont("Machine Check %s Address: %#llx\n",
701 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
702 	}
703 
704 silent_out:
705 	mtspr(SPRN_MCSR, mcsr);
706 	return mfspr(SPRN_MCSR) == 0 && recoverable;
707 }
708 
709 int machine_check_e500(struct pt_regs *regs)
710 {
711 	unsigned long reason = mfspr(SPRN_MCSR);
712 
713 	if (reason & MCSR_BUS_RBERR) {
714 		if (fsl_rio_mcheck_exception(regs))
715 			return 1;
716 		if (fsl_pci_mcheck_exception(regs))
717 			return 1;
718 	}
719 
720 	printk("Machine check in kernel mode.\n");
721 	printk("Caused by (from MCSR=%lx): ", reason);
722 
723 	if (reason & MCSR_MCP)
724 		pr_cont("Machine Check Signal\n");
725 	if (reason & MCSR_ICPERR)
726 		pr_cont("Instruction Cache Parity Error\n");
727 	if (reason & MCSR_DCP_PERR)
728 		pr_cont("Data Cache Push Parity Error\n");
729 	if (reason & MCSR_DCPERR)
730 		pr_cont("Data Cache Parity Error\n");
731 	if (reason & MCSR_BUS_IAERR)
732 		pr_cont("Bus - Instruction Address Error\n");
733 	if (reason & MCSR_BUS_RAERR)
734 		pr_cont("Bus - Read Address Error\n");
735 	if (reason & MCSR_BUS_WAERR)
736 		pr_cont("Bus - Write Address Error\n");
737 	if (reason & MCSR_BUS_IBERR)
738 		pr_cont("Bus - Instruction Data Error\n");
739 	if (reason & MCSR_BUS_RBERR)
740 		pr_cont("Bus - Read Data Bus Error\n");
741 	if (reason & MCSR_BUS_WBERR)
742 		pr_cont("Bus - Write Data Bus Error\n");
743 	if (reason & MCSR_BUS_IPERR)
744 		pr_cont("Bus - Instruction Parity Error\n");
745 	if (reason & MCSR_BUS_RPERR)
746 		pr_cont("Bus - Read Parity Error\n");
747 
748 	return 0;
749 }
750 
751 int machine_check_generic(struct pt_regs *regs)
752 {
753 	return 0;
754 }
755 #elif defined(CONFIG_PPC32)
756 int machine_check_generic(struct pt_regs *regs)
757 {
758 	unsigned long reason = regs->msr;
759 
760 	printk("Machine check in kernel mode.\n");
761 	printk("Caused by (from SRR1=%lx): ", reason);
762 	switch (reason & 0x601F0000) {
763 	case 0x80000:
764 		pr_cont("Machine check signal\n");
765 		break;
766 	case 0x40000:
767 	case 0x140000:	/* 7450 MSS error and TEA */
768 		pr_cont("Transfer error ack signal\n");
769 		break;
770 	case 0x20000:
771 		pr_cont("Data parity error signal\n");
772 		break;
773 	case 0x10000:
774 		pr_cont("Address parity error signal\n");
775 		break;
776 	case 0x20000000:
777 		pr_cont("L1 Data Cache error\n");
778 		break;
779 	case 0x40000000:
780 		pr_cont("L1 Instruction Cache error\n");
781 		break;
782 	case 0x00100000:
783 		pr_cont("L2 data cache parity error\n");
784 		break;
785 	default:
786 		pr_cont("Unknown values in msr\n");
787 	}
788 	return 0;
789 }
790 #endif /* everything else */
791 
792 void die_mce(const char *str, struct pt_regs *regs, long err)
793 {
794 	/*
795 	 * The machine check wants to kill the interrupted context,
796 	 * but make_task_dead() checks for in_interrupt() and panics
797 	 * in that case, so exit the irq/nmi before calling die.
798 	 */
799 	if (in_nmi())
800 		nmi_exit();
801 	else
802 		irq_exit();
803 	die(str, regs, err);
804 }
805 
806 /*
807  * BOOK3S_64 does not usually call this handler as a non-maskable interrupt
808  * (it uses its own early real-mode handler to handle the MCE proper
809  * and then raises irq_work to call this handler when interrupts are
810  * enabled). The only time when this is not true is if the early handler
811  * is unrecoverable, then it does call this directly to try to get a
812  * message out.
813  */
814 static void __machine_check_exception(struct pt_regs *regs)
815 {
816 	int recover = 0;
817 
818 	__this_cpu_inc(irq_stat.mce_exceptions);
819 
820 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
821 
822 	/* See if any machine dependent calls. In theory, we would want
823 	 * to call the CPU first, and call the ppc_md. one if the CPU
824 	 * one returns a positive number. However there is existing code
825 	 * that assumes the board gets a first chance, so let's keep it
826 	 * that way for now and fix things later. --BenH.
827 	 */
828 	if (ppc_md.machine_check_exception)
829 		recover = ppc_md.machine_check_exception(regs);
830 	else if (cur_cpu_spec->machine_check)
831 		recover = cur_cpu_spec->machine_check(regs);
832 
833 	if (recover > 0)
834 		goto bail;
835 
836 	if (debugger_fault_handler(regs))
837 		goto bail;
838 
839 	if (check_io_access(regs))
840 		goto bail;
841 
842 	die_mce("Machine check", regs, SIGBUS);
843 
844 bail:
845 	/* Must die if the interrupt is not recoverable */
846 	if (regs_is_unrecoverable(regs))
847 		die_mce("Unrecoverable Machine check", regs, SIGBUS);
848 }
849 
850 #ifdef CONFIG_PPC_BOOK3S_64
851 DEFINE_INTERRUPT_HANDLER_ASYNC(machine_check_exception_async)
852 {
853 	__machine_check_exception(regs);
854 }
855 #endif
856 DEFINE_INTERRUPT_HANDLER_NMI(machine_check_exception)
857 {
858 	__machine_check_exception(regs);
859 
860 	return 0;
861 }
862 
863 DEFINE_INTERRUPT_HANDLER(SMIException) /* async? */
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] __aligned(16), *vdst;
874 	unsigned long ea, msr, msr_mask;
875 	bool swap;
876 
877 	if (__get_user(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_add_return_ip(regs, 4);
1046 }
1047 #endif /* CONFIG_VSX */
1048 
1049 DEFINE_INTERRUPT_HANDLER_ASYNC(handle_hmi_exception)
1050 {
1051 	struct pt_regs *old_regs;
1052 
1053 	old_regs = set_irq_regs(regs);
1054 
1055 #ifdef CONFIG_VSX
1056 	/* Real mode flagged P9 special emu is needed */
1057 	if (local_paca->hmi_p9_special_emu) {
1058 		local_paca->hmi_p9_special_emu = 0;
1059 
1060 		/*
1061 		 * We don't want to take page faults while doing the
1062 		 * emulation, we just replay the instruction if necessary.
1063 		 */
1064 		pagefault_disable();
1065 		p9_hmi_special_emu(regs);
1066 		pagefault_enable();
1067 	}
1068 #endif /* CONFIG_VSX */
1069 
1070 	if (ppc_md.handle_hmi_exception)
1071 		ppc_md.handle_hmi_exception(regs);
1072 
1073 	set_irq_regs(old_regs);
1074 }
1075 
1076 DEFINE_INTERRUPT_HANDLER(unknown_exception)
1077 {
1078 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1079 	       regs->nip, regs->msr, regs->trap);
1080 
1081 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1082 }
1083 
1084 DEFINE_INTERRUPT_HANDLER_ASYNC(unknown_async_exception)
1085 {
1086 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1087 	       regs->nip, regs->msr, regs->trap);
1088 
1089 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1090 }
1091 
1092 DEFINE_INTERRUPT_HANDLER_NMI(unknown_nmi_exception)
1093 {
1094 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1095 	       regs->nip, regs->msr, regs->trap);
1096 
1097 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1098 
1099 	return 0;
1100 }
1101 
1102 DEFINE_INTERRUPT_HANDLER(instruction_breakpoint_exception)
1103 {
1104 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1105 					5, SIGTRAP) == NOTIFY_STOP)
1106 		return;
1107 	if (debugger_iabr_match(regs))
1108 		return;
1109 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1110 }
1111 
1112 DEFINE_INTERRUPT_HANDLER(RunModeException)
1113 {
1114 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1115 }
1116 
1117 static void __single_step_exception(struct pt_regs *regs)
1118 {
1119 	clear_single_step(regs);
1120 	clear_br_trace(regs);
1121 
1122 	if (kprobe_post_handler(regs))
1123 		return;
1124 
1125 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1126 					5, SIGTRAP) == NOTIFY_STOP)
1127 		return;
1128 	if (debugger_sstep(regs))
1129 		return;
1130 
1131 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1132 }
1133 
1134 DEFINE_INTERRUPT_HANDLER(single_step_exception)
1135 {
1136 	__single_step_exception(regs);
1137 }
1138 
1139 /*
1140  * After we have successfully emulated an instruction, we have to
1141  * check if the instruction was being single-stepped, and if so,
1142  * pretend we got a single-step exception.  This was pointed out
1143  * by Kumar Gala.  -- paulus
1144  */
1145 static void emulate_single_step(struct pt_regs *regs)
1146 {
1147 	if (single_stepping(regs))
1148 		__single_step_exception(regs);
1149 }
1150 
1151 static inline int __parse_fpscr(unsigned long fpscr)
1152 {
1153 	int ret = FPE_FLTUNK;
1154 
1155 	/* Invalid operation */
1156 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1157 		ret = FPE_FLTINV;
1158 
1159 	/* Overflow */
1160 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1161 		ret = FPE_FLTOVF;
1162 
1163 	/* Underflow */
1164 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1165 		ret = FPE_FLTUND;
1166 
1167 	/* Divide by zero */
1168 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1169 		ret = FPE_FLTDIV;
1170 
1171 	/* Inexact result */
1172 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1173 		ret = FPE_FLTRES;
1174 
1175 	return ret;
1176 }
1177 
1178 static void parse_fpe(struct pt_regs *regs)
1179 {
1180 	int code = 0;
1181 
1182 	flush_fp_to_thread(current);
1183 
1184 #ifdef CONFIG_PPC_FPU_REGS
1185 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1186 #endif
1187 
1188 	_exception(SIGFPE, regs, code, regs->nip);
1189 }
1190 
1191 /*
1192  * Illegal instruction emulation support.  Originally written to
1193  * provide the PVR to user applications using the mfspr rd, PVR.
1194  * Return non-zero if we can't emulate, or -EFAULT if the associated
1195  * memory access caused an access fault.  Return zero on success.
1196  *
1197  * There are a couple of ways to do this, either "decode" the instruction
1198  * or directly match lots of bits.  In this case, matching lots of
1199  * bits is faster and easier.
1200  *
1201  */
1202 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1203 {
1204 	u8 rT = (instword >> 21) & 0x1f;
1205 	u8 rA = (instword >> 16) & 0x1f;
1206 	u8 NB_RB = (instword >> 11) & 0x1f;
1207 	u32 num_bytes;
1208 	unsigned long EA;
1209 	int pos = 0;
1210 
1211 	/* Early out if we are an invalid form of lswx */
1212 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1213 		if ((rT == rA) || (rT == NB_RB))
1214 			return -EINVAL;
1215 
1216 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1217 
1218 	switch (instword & PPC_INST_STRING_MASK) {
1219 		case PPC_INST_LSWX:
1220 		case PPC_INST_STSWX:
1221 			EA += NB_RB;
1222 			num_bytes = regs->xer & 0x7f;
1223 			break;
1224 		case PPC_INST_LSWI:
1225 		case PPC_INST_STSWI:
1226 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1227 			break;
1228 		default:
1229 			return -EINVAL;
1230 	}
1231 
1232 	while (num_bytes != 0)
1233 	{
1234 		u8 val;
1235 		u32 shift = 8 * (3 - (pos & 0x3));
1236 
1237 		/* if process is 32-bit, clear upper 32 bits of EA */
1238 		if ((regs->msr & MSR_64BIT) == 0)
1239 			EA &= 0xFFFFFFFF;
1240 
1241 		switch ((instword & PPC_INST_STRING_MASK)) {
1242 			case PPC_INST_LSWX:
1243 			case PPC_INST_LSWI:
1244 				if (get_user(val, (u8 __user *)EA))
1245 					return -EFAULT;
1246 				/* first time updating this reg,
1247 				 * zero it out */
1248 				if (pos == 0)
1249 					regs->gpr[rT] = 0;
1250 				regs->gpr[rT] |= val << shift;
1251 				break;
1252 			case PPC_INST_STSWI:
1253 			case PPC_INST_STSWX:
1254 				val = regs->gpr[rT] >> shift;
1255 				if (put_user(val, (u8 __user *)EA))
1256 					return -EFAULT;
1257 				break;
1258 		}
1259 		/* move EA to next address */
1260 		EA += 1;
1261 		num_bytes--;
1262 
1263 		/* manage our position within the register */
1264 		if (++pos == 4) {
1265 			pos = 0;
1266 			if (++rT == 32)
1267 				rT = 0;
1268 		}
1269 	}
1270 
1271 	return 0;
1272 }
1273 
1274 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1275 {
1276 	u32 ra,rs;
1277 	unsigned long tmp;
1278 
1279 	ra = (instword >> 16) & 0x1f;
1280 	rs = (instword >> 21) & 0x1f;
1281 
1282 	tmp = regs->gpr[rs];
1283 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1284 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1285 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1286 	regs->gpr[ra] = tmp;
1287 
1288 	return 0;
1289 }
1290 
1291 static int emulate_isel(struct pt_regs *regs, u32 instword)
1292 {
1293 	u8 rT = (instword >> 21) & 0x1f;
1294 	u8 rA = (instword >> 16) & 0x1f;
1295 	u8 rB = (instword >> 11) & 0x1f;
1296 	u8 BC = (instword >> 6) & 0x1f;
1297 	u8 bit;
1298 	unsigned long tmp;
1299 
1300 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1301 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1302 
1303 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1304 
1305 	return 0;
1306 }
1307 
1308 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1309 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1310 {
1311         /* If we're emulating a load/store in an active transaction, we cannot
1312          * emulate it as the kernel operates in transaction suspended context.
1313          * We need to abort the transaction.  This creates a persistent TM
1314          * abort so tell the user what caused it with a new code.
1315 	 */
1316 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1317 		tm_enable();
1318 		tm_abort(cause);
1319 		return true;
1320 	}
1321 	return false;
1322 }
1323 #else
1324 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1325 {
1326 	return false;
1327 }
1328 #endif
1329 
1330 static int emulate_instruction(struct pt_regs *regs)
1331 {
1332 	u32 instword;
1333 	u32 rd;
1334 
1335 	if (!user_mode(regs))
1336 		return -EINVAL;
1337 
1338 	if (get_user(instword, (u32 __user *)(regs->nip)))
1339 		return -EFAULT;
1340 
1341 	/* Emulate the mfspr rD, PVR. */
1342 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1343 		PPC_WARN_EMULATED(mfpvr, regs);
1344 		rd = (instword >> 21) & 0x1f;
1345 		regs->gpr[rd] = mfspr(SPRN_PVR);
1346 		return 0;
1347 	}
1348 
1349 	/* Emulating the dcba insn is just a no-op.  */
1350 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1351 		PPC_WARN_EMULATED(dcba, regs);
1352 		return 0;
1353 	}
1354 
1355 	/* Emulate the mcrxr insn.  */
1356 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1357 		int shift = (instword >> 21) & 0x1c;
1358 		unsigned long msk = 0xf0000000UL >> shift;
1359 
1360 		PPC_WARN_EMULATED(mcrxr, regs);
1361 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1362 		regs->xer &= ~0xf0000000UL;
1363 		return 0;
1364 	}
1365 
1366 	/* Emulate load/store string insn. */
1367 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1368 		if (tm_abort_check(regs,
1369 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1370 			return -EINVAL;
1371 		PPC_WARN_EMULATED(string, regs);
1372 		return emulate_string_inst(regs, instword);
1373 	}
1374 
1375 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1376 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1377 		PPC_WARN_EMULATED(popcntb, regs);
1378 		return emulate_popcntb_inst(regs, instword);
1379 	}
1380 
1381 	/* Emulate isel (Integer Select) instruction */
1382 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1383 		PPC_WARN_EMULATED(isel, regs);
1384 		return emulate_isel(regs, instword);
1385 	}
1386 
1387 	/* Emulate sync instruction variants */
1388 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1389 		PPC_WARN_EMULATED(sync, regs);
1390 		asm volatile("sync");
1391 		return 0;
1392 	}
1393 
1394 #ifdef CONFIG_PPC64
1395 	/* Emulate the mfspr rD, DSCR. */
1396 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1397 		PPC_INST_MFSPR_DSCR_USER) ||
1398 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1399 		PPC_INST_MFSPR_DSCR)) &&
1400 			cpu_has_feature(CPU_FTR_DSCR)) {
1401 		PPC_WARN_EMULATED(mfdscr, regs);
1402 		rd = (instword >> 21) & 0x1f;
1403 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1404 		return 0;
1405 	}
1406 	/* Emulate the mtspr DSCR, rD. */
1407 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1408 		PPC_INST_MTSPR_DSCR_USER) ||
1409 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1410 		PPC_INST_MTSPR_DSCR)) &&
1411 			cpu_has_feature(CPU_FTR_DSCR)) {
1412 		PPC_WARN_EMULATED(mtdscr, regs);
1413 		rd = (instword >> 21) & 0x1f;
1414 		current->thread.dscr = regs->gpr[rd];
1415 		current->thread.dscr_inherit = 1;
1416 		mtspr(SPRN_DSCR, current->thread.dscr);
1417 		return 0;
1418 	}
1419 #endif
1420 
1421 	return -EINVAL;
1422 }
1423 
1424 int is_valid_bugaddr(unsigned long addr)
1425 {
1426 	return is_kernel_addr(addr);
1427 }
1428 
1429 #ifdef CONFIG_MATH_EMULATION
1430 static int emulate_math(struct pt_regs *regs)
1431 {
1432 	int ret;
1433 
1434 	ret = do_mathemu(regs);
1435 	if (ret >= 0)
1436 		PPC_WARN_EMULATED(math, regs);
1437 
1438 	switch (ret) {
1439 	case 0:
1440 		emulate_single_step(regs);
1441 		return 0;
1442 	case 1: {
1443 			int code = 0;
1444 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1445 			_exception(SIGFPE, regs, code, regs->nip);
1446 			return 0;
1447 		}
1448 	case -EFAULT:
1449 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1450 		return 0;
1451 	}
1452 
1453 	return -1;
1454 }
1455 #else
1456 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1457 #endif
1458 
1459 static void do_program_check(struct pt_regs *regs)
1460 {
1461 	unsigned int reason = get_reason(regs);
1462 
1463 	/* We can now get here via a FP Unavailable exception if the core
1464 	 * has no FPU, in that case the reason flags will be 0 */
1465 
1466 	if (reason & REASON_FP) {
1467 		/* IEEE FP exception */
1468 		parse_fpe(regs);
1469 		return;
1470 	}
1471 	if (reason & REASON_TRAP) {
1472 		unsigned long bugaddr;
1473 		/* Debugger is first in line to stop recursive faults in
1474 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1475 		if (debugger_bpt(regs))
1476 			return;
1477 
1478 		if (kprobe_handler(regs))
1479 			return;
1480 
1481 		/* trap exception */
1482 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1483 				== NOTIFY_STOP)
1484 			return;
1485 
1486 		bugaddr = regs->nip;
1487 		/*
1488 		 * Fixup bugaddr for BUG_ON() in real mode
1489 		 */
1490 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1491 			bugaddr += PAGE_OFFSET;
1492 
1493 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1494 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1495 			const struct exception_table_entry *entry;
1496 
1497 			entry = search_exception_tables(bugaddr);
1498 			if (entry) {
1499 				regs_set_return_ip(regs, extable_fixup(entry) + regs->nip - bugaddr);
1500 				return;
1501 			}
1502 		}
1503 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1504 		return;
1505 	}
1506 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1507 	if (reason & REASON_TM) {
1508 		/* This is a TM "Bad Thing Exception" program check.
1509 		 * This occurs when:
1510 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1511 		 *    transition in TM states.
1512 		 * -  A trechkpt is attempted when transactional.
1513 		 * -  A treclaim is attempted when non transactional.
1514 		 * -  A tend is illegally attempted.
1515 		 * -  writing a TM SPR when transactional.
1516 		 *
1517 		 * If usermode caused this, it's done something illegal and
1518 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1519 		 * operand to distinguish from the instruction just being bad
1520 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1521 		 * illegal /placement/ of a valid instruction.
1522 		 */
1523 		if (user_mode(regs)) {
1524 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1525 			return;
1526 		} else {
1527 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1528 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1529 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1530 			die("Unrecoverable exception", regs, SIGABRT);
1531 		}
1532 	}
1533 #endif
1534 
1535 	/*
1536 	 * If we took the program check in the kernel skip down to sending a
1537 	 * SIGILL. The subsequent cases all relate to emulating instructions
1538 	 * which we should only do for userspace. We also do not want to enable
1539 	 * interrupts for kernel faults because that might lead to further
1540 	 * faults, and loose the context of the original exception.
1541 	 */
1542 	if (!user_mode(regs))
1543 		goto sigill;
1544 
1545 	interrupt_cond_local_irq_enable(regs);
1546 
1547 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1548 	 * but there seems to be a hardware bug on the 405GP (RevD)
1549 	 * that means ESR is sometimes set incorrectly - either to
1550 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1551 	 * hardware people - not sure if it can happen on any illegal
1552 	 * instruction or only on FP instructions, whether there is a
1553 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1554 	 */
1555 	if (!emulate_math(regs))
1556 		return;
1557 
1558 	/* Try to emulate it if we should. */
1559 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1560 		switch (emulate_instruction(regs)) {
1561 		case 0:
1562 			regs_add_return_ip(regs, 4);
1563 			emulate_single_step(regs);
1564 			return;
1565 		case -EFAULT:
1566 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1567 			return;
1568 		}
1569 	}
1570 
1571 sigill:
1572 	if (reason & REASON_PRIVILEGED)
1573 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1574 	else
1575 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1576 
1577 }
1578 
1579 DEFINE_INTERRUPT_HANDLER(program_check_exception)
1580 {
1581 	do_program_check(regs);
1582 }
1583 
1584 /*
1585  * This occurs when running in hypervisor mode on POWER6 or later
1586  * and an illegal instruction is encountered.
1587  */
1588 DEFINE_INTERRUPT_HANDLER(emulation_assist_interrupt)
1589 {
1590 	regs_set_return_msr(regs, regs->msr | REASON_ILLEGAL);
1591 	do_program_check(regs);
1592 }
1593 
1594 DEFINE_INTERRUPT_HANDLER(alignment_exception)
1595 {
1596 	int sig, code, fixed = 0;
1597 	unsigned long  reason;
1598 
1599 	interrupt_cond_local_irq_enable(regs);
1600 
1601 	reason = get_reason(regs);
1602 	if (reason & REASON_BOUNDARY) {
1603 		sig = SIGBUS;
1604 		code = BUS_ADRALN;
1605 		goto bad;
1606 	}
1607 
1608 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1609 		return;
1610 
1611 	/* we don't implement logging of alignment exceptions */
1612 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1613 		fixed = fix_alignment(regs);
1614 
1615 	if (fixed == 1) {
1616 		/* skip over emulated instruction */
1617 		regs_add_return_ip(regs, inst_length(reason));
1618 		emulate_single_step(regs);
1619 		return;
1620 	}
1621 
1622 	/* Operand address was bad */
1623 	if (fixed == -EFAULT) {
1624 		sig = SIGSEGV;
1625 		code = SEGV_ACCERR;
1626 	} else {
1627 		sig = SIGBUS;
1628 		code = BUS_ADRALN;
1629 	}
1630 bad:
1631 	if (user_mode(regs))
1632 		_exception(sig, regs, code, regs->dar);
1633 	else
1634 		bad_page_fault(regs, sig);
1635 }
1636 
1637 DEFINE_INTERRUPT_HANDLER(stack_overflow_exception)
1638 {
1639 	die("Kernel stack overflow", regs, SIGSEGV);
1640 }
1641 
1642 DEFINE_INTERRUPT_HANDLER(kernel_fp_unavailable_exception)
1643 {
1644 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1645 			  "%lx at %lx\n", regs->trap, regs->nip);
1646 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1647 }
1648 
1649 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_exception)
1650 {
1651 	if (user_mode(regs)) {
1652 		/* A user program has executed an altivec instruction,
1653 		   but this kernel doesn't support altivec. */
1654 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1655 		return;
1656 	}
1657 
1658 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1659 			"%lx at %lx\n", regs->trap, regs->nip);
1660 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1661 }
1662 
1663 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_exception)
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_set_return_msr(regs, 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 DEFINE_INTERRUPT_HANDLER(facility_unavailable_exception)
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 		[FSCR_PREFIX_LG] = "PREFIX",
1708 	};
1709 	char *facility = "unknown";
1710 	u64 value;
1711 	u32 instword, rd;
1712 	u8 status;
1713 	bool hv;
1714 
1715 	hv = (TRAP(regs) == INTERRUPT_H_FAC_UNAVAIL);
1716 	if (hv)
1717 		value = mfspr(SPRN_HFSCR);
1718 	else
1719 		value = mfspr(SPRN_FSCR);
1720 
1721 	status = value >> 56;
1722 	if ((hv || status >= 2) &&
1723 	    (status < ARRAY_SIZE(facility_strings)) &&
1724 	    facility_strings[status])
1725 		facility = facility_strings[status];
1726 
1727 	/* We should not have taken this interrupt in kernel */
1728 	if (!user_mode(regs)) {
1729 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1730 			 facility, status, regs->nip);
1731 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1732 	}
1733 
1734 	interrupt_cond_local_irq_enable(regs);
1735 
1736 	if (status == FSCR_DSCR_LG) {
1737 		/*
1738 		 * User is accessing the DSCR register using the problem
1739 		 * state only SPR number (0x03) either through a mfspr or
1740 		 * a mtspr instruction. If it is a write attempt through
1741 		 * a mtspr, then we set the inherit bit. This also allows
1742 		 * the user to write or read the register directly in the
1743 		 * future by setting via the FSCR DSCR bit. But in case it
1744 		 * is a read DSCR attempt through a mfspr instruction, we
1745 		 * just emulate the instruction instead. This code path will
1746 		 * always emulate all the mfspr instructions till the user
1747 		 * has attempted at least one mtspr instruction. This way it
1748 		 * preserves the same behaviour when the user is accessing
1749 		 * the DSCR through privilege level only SPR number (0x11)
1750 		 * which is emulated through illegal instruction exception.
1751 		 * We always leave HFSCR DSCR set.
1752 		 */
1753 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1754 			pr_err("Failed to fetch the user instruction\n");
1755 			return;
1756 		}
1757 
1758 		/* Write into DSCR (mtspr 0x03, RS) */
1759 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1760 				== PPC_INST_MTSPR_DSCR_USER) {
1761 			rd = (instword >> 21) & 0x1f;
1762 			current->thread.dscr = regs->gpr[rd];
1763 			current->thread.dscr_inherit = 1;
1764 			current->thread.fscr |= FSCR_DSCR;
1765 			mtspr(SPRN_FSCR, current->thread.fscr);
1766 		}
1767 
1768 		/* Read from DSCR (mfspr RT, 0x03) */
1769 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1770 				== PPC_INST_MFSPR_DSCR_USER) {
1771 			if (emulate_instruction(regs)) {
1772 				pr_err("DSCR based mfspr emulation failed\n");
1773 				return;
1774 			}
1775 			regs_add_return_ip(regs, 4);
1776 			emulate_single_step(regs);
1777 		}
1778 		return;
1779 	}
1780 
1781 	if (status == FSCR_TM_LG) {
1782 		/*
1783 		 * If we're here then the hardware is TM aware because it
1784 		 * generated an exception with FSRM_TM set.
1785 		 *
1786 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1787 		 * told us not to do TM, or the kernel is not built with TM
1788 		 * support.
1789 		 *
1790 		 * If both of those things are true, then userspace can spam the
1791 		 * console by triggering the printk() below just by continually
1792 		 * doing tbegin (or any TM instruction). So in that case just
1793 		 * send the process a SIGILL immediately.
1794 		 */
1795 		if (!cpu_has_feature(CPU_FTR_TM))
1796 			goto out;
1797 
1798 		tm_unavailable(regs);
1799 		return;
1800 	}
1801 
1802 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1803 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1804 
1805 out:
1806 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1807 }
1808 #endif
1809 
1810 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1811 
1812 DEFINE_INTERRUPT_HANDLER(fp_unavailable_tm)
1813 {
1814 	/* Note:  This does not handle any kind of FP laziness. */
1815 
1816 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1817 		 regs->nip, regs->msr);
1818 
1819         /* We can only have got here if the task started using FP after
1820          * beginning the transaction.  So, the transactional regs are just a
1821          * copy of the checkpointed ones.  But, we still need to recheckpoint
1822          * as we're enabling FP for the process; it will return, abort the
1823          * transaction, and probably retry but now with FP enabled.  So the
1824          * checkpointed FP registers need to be loaded.
1825 	 */
1826 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1827 
1828 	/*
1829 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1830 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1831 	 *
1832 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1833 	 * recheckpoint.
1834 	 */
1835 
1836 	/* Enable FP for the task: */
1837 	current->thread.load_fp = 1;
1838 
1839 	/*
1840 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1841 	 */
1842 	tm_recheckpoint(&current->thread);
1843 }
1844 
1845 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_tm)
1846 {
1847 	/* See the comments in fp_unavailable_tm().  This function operates
1848 	 * the same way.
1849 	 */
1850 
1851 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1852 		 "MSR=%lx\n",
1853 		 regs->nip, regs->msr);
1854 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1855 	current->thread.load_vec = 1;
1856 	tm_recheckpoint(&current->thread);
1857 	current->thread.used_vr = 1;
1858 }
1859 
1860 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_tm)
1861 {
1862 	/* See the comments in fp_unavailable_tm().  This works similarly,
1863 	 * though we're loading both FP and VEC registers in here.
1864 	 *
1865 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1866 	 * regs.  Either way, set MSR_VSX.
1867 	 */
1868 
1869 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1870 		 "MSR=%lx\n",
1871 		 regs->nip, regs->msr);
1872 
1873 	current->thread.used_vsr = 1;
1874 
1875 	/* This reclaims FP and/or VR regs if they're already enabled */
1876 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1877 
1878 	current->thread.load_vec = 1;
1879 	current->thread.load_fp = 1;
1880 
1881 	tm_recheckpoint(&current->thread);
1882 }
1883 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1884 
1885 #ifdef CONFIG_PPC64
1886 DECLARE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi);
1887 DEFINE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi)
1888 {
1889 	__this_cpu_inc(irq_stat.pmu_irqs);
1890 
1891 	perf_irq(regs);
1892 
1893 	return 0;
1894 }
1895 #endif
1896 
1897 DECLARE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async);
1898 DEFINE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async)
1899 {
1900 	__this_cpu_inc(irq_stat.pmu_irqs);
1901 
1902 	perf_irq(regs);
1903 }
1904 
1905 DEFINE_INTERRUPT_HANDLER_RAW(performance_monitor_exception)
1906 {
1907 	/*
1908 	 * On 64-bit, if perf interrupts hit in a local_irq_disable
1909 	 * (soft-masked) region, we consider them as NMIs. This is required to
1910 	 * prevent hash faults on user addresses when reading callchains (and
1911 	 * looks better from an irq tracing perspective).
1912 	 */
1913 	if (IS_ENABLED(CONFIG_PPC64) && unlikely(arch_irq_disabled_regs(regs)))
1914 		performance_monitor_exception_nmi(regs);
1915 	else
1916 		performance_monitor_exception_async(regs);
1917 
1918 	return 0;
1919 }
1920 
1921 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1922 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1923 {
1924 	int changed = 0;
1925 	/*
1926 	 * Determine the cause of the debug event, clear the
1927 	 * event flags and send a trap to the handler. Torez
1928 	 */
1929 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1930 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1931 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1932 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1933 #endif
1934 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1935 			     5);
1936 		changed |= 0x01;
1937 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1938 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1939 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1940 			     6);
1941 		changed |= 0x01;
1942 	}  else if (debug_status & DBSR_IAC1) {
1943 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1944 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1945 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1946 			     1);
1947 		changed |= 0x01;
1948 	}  else if (debug_status & DBSR_IAC2) {
1949 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1950 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1951 			     2);
1952 		changed |= 0x01;
1953 	}  else if (debug_status & DBSR_IAC3) {
1954 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1955 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1956 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1957 			     3);
1958 		changed |= 0x01;
1959 	}  else if (debug_status & DBSR_IAC4) {
1960 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1961 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1962 			     4);
1963 		changed |= 0x01;
1964 	}
1965 	/*
1966 	 * At the point this routine was called, the MSR(DE) was turned off.
1967 	 * Check all other debug flags and see if that bit needs to be turned
1968 	 * back on or not.
1969 	 */
1970 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1971 			       current->thread.debug.dbcr1))
1972 		regs_set_return_msr(regs, regs->msr | MSR_DE);
1973 	else
1974 		/* Make sure the IDM flag is off */
1975 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1976 
1977 	if (changed & 0x01)
1978 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1979 }
1980 
1981 DEFINE_INTERRUPT_HANDLER(DebugException)
1982 {
1983 	unsigned long debug_status = regs->dsisr;
1984 
1985 	current->thread.debug.dbsr = debug_status;
1986 
1987 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1988 	 * on server, it stops on the target of the branch. In order to simulate
1989 	 * the server behaviour, we thus restart right away with a single step
1990 	 * instead of stopping here when hitting a BT
1991 	 */
1992 	if (debug_status & DBSR_BT) {
1993 		regs_set_return_msr(regs, regs->msr & ~MSR_DE);
1994 
1995 		/* Disable BT */
1996 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1997 		/* Clear the BT event */
1998 		mtspr(SPRN_DBSR, DBSR_BT);
1999 
2000 		/* Do the single step trick only when coming from userspace */
2001 		if (user_mode(regs)) {
2002 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
2003 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
2004 			regs_set_return_msr(regs, regs->msr | MSR_DE);
2005 			return;
2006 		}
2007 
2008 		if (kprobe_post_handler(regs))
2009 			return;
2010 
2011 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
2012 			       5, SIGTRAP) == NOTIFY_STOP) {
2013 			return;
2014 		}
2015 		if (debugger_sstep(regs))
2016 			return;
2017 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
2018 		regs_set_return_msr(regs, regs->msr & ~MSR_DE);
2019 
2020 		/* Disable instruction completion */
2021 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
2022 		/* Clear the instruction completion event */
2023 		mtspr(SPRN_DBSR, DBSR_IC);
2024 
2025 		if (kprobe_post_handler(regs))
2026 			return;
2027 
2028 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2029 			       5, SIGTRAP) == NOTIFY_STOP) {
2030 			return;
2031 		}
2032 
2033 		if (debugger_sstep(regs))
2034 			return;
2035 
2036 		if (user_mode(regs)) {
2037 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2038 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2039 					       current->thread.debug.dbcr1))
2040 				regs_set_return_msr(regs, regs->msr | MSR_DE);
2041 			else
2042 				/* Make sure the IDM bit is off */
2043 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2044 		}
2045 
2046 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2047 	} else
2048 		handle_debug(regs, debug_status);
2049 }
2050 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2051 
2052 #ifdef CONFIG_ALTIVEC
2053 DEFINE_INTERRUPT_HANDLER(altivec_assist_exception)
2054 {
2055 	int err;
2056 
2057 	if (!user_mode(regs)) {
2058 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2059 		       " at %lx\n", regs->nip);
2060 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2061 	}
2062 
2063 	flush_altivec_to_thread(current);
2064 
2065 	PPC_WARN_EMULATED(altivec, regs);
2066 	err = emulate_altivec(regs);
2067 	if (err == 0) {
2068 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2069 		emulate_single_step(regs);
2070 		return;
2071 	}
2072 
2073 	if (err == -EFAULT) {
2074 		/* got an error reading the instruction */
2075 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2076 	} else {
2077 		/* didn't recognize the instruction */
2078 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2079 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2080 				   "in %s at %lx\n", current->comm, regs->nip);
2081 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2082 	}
2083 }
2084 #endif /* CONFIG_ALTIVEC */
2085 
2086 #ifdef CONFIG_FSL_BOOKE
2087 DEFINE_INTERRUPT_HANDLER(CacheLockingException)
2088 {
2089 	unsigned long error_code = regs->dsisr;
2090 
2091 	/* We treat cache locking instructions from the user
2092 	 * as priv ops, in the future we could try to do
2093 	 * something smarter
2094 	 */
2095 	if (error_code & (ESR_DLK|ESR_ILK))
2096 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2097 	return;
2098 }
2099 #endif /* CONFIG_FSL_BOOKE */
2100 
2101 #ifdef CONFIG_SPE
2102 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointException)
2103 {
2104 	extern int do_spe_mathemu(struct pt_regs *regs);
2105 	unsigned long spefscr;
2106 	int fpexc_mode;
2107 	int code = FPE_FLTUNK;
2108 	int err;
2109 
2110 	interrupt_cond_local_irq_enable(regs);
2111 
2112 	flush_spe_to_thread(current);
2113 
2114 	spefscr = current->thread.spefscr;
2115 	fpexc_mode = current->thread.fpexc_mode;
2116 
2117 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2118 		code = FPE_FLTOVF;
2119 	}
2120 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2121 		code = FPE_FLTUND;
2122 	}
2123 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2124 		code = FPE_FLTDIV;
2125 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2126 		code = FPE_FLTINV;
2127 	}
2128 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2129 		code = FPE_FLTRES;
2130 
2131 	err = do_spe_mathemu(regs);
2132 	if (err == 0) {
2133 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2134 		emulate_single_step(regs);
2135 		return;
2136 	}
2137 
2138 	if (err == -EFAULT) {
2139 		/* got an error reading the instruction */
2140 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2141 	} else if (err == -EINVAL) {
2142 		/* didn't recognize the instruction */
2143 		printk(KERN_ERR "unrecognized spe instruction "
2144 		       "in %s at %lx\n", current->comm, regs->nip);
2145 	} else {
2146 		_exception(SIGFPE, regs, code, regs->nip);
2147 	}
2148 
2149 	return;
2150 }
2151 
2152 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointRoundException)
2153 {
2154 	extern int speround_handler(struct pt_regs *regs);
2155 	int err;
2156 
2157 	interrupt_cond_local_irq_enable(regs);
2158 
2159 	preempt_disable();
2160 	if (regs->msr & MSR_SPE)
2161 		giveup_spe(current);
2162 	preempt_enable();
2163 
2164 	regs_add_return_ip(regs, -4);
2165 	err = speround_handler(regs);
2166 	if (err == 0) {
2167 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2168 		emulate_single_step(regs);
2169 		return;
2170 	}
2171 
2172 	if (err == -EFAULT) {
2173 		/* got an error reading the instruction */
2174 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2175 	} else if (err == -EINVAL) {
2176 		/* didn't recognize the instruction */
2177 		printk(KERN_ERR "unrecognized spe instruction "
2178 		       "in %s at %lx\n", current->comm, regs->nip);
2179 	} else {
2180 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2181 		return;
2182 	}
2183 }
2184 #endif
2185 
2186 /*
2187  * We enter here if we get an unrecoverable exception, that is, one
2188  * that happened at a point where the RI (recoverable interrupt) bit
2189  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2190  * we therefore lost state by taking this exception.
2191  */
2192 void __noreturn unrecoverable_exception(struct pt_regs *regs)
2193 {
2194 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2195 		 regs->trap, regs->nip, regs->msr);
2196 	die("Unrecoverable exception", regs, SIGABRT);
2197 	/* die() should not return */
2198 	for (;;)
2199 		;
2200 }
2201 
2202 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2203 /*
2204  * Default handler for a Watchdog exception,
2205  * spins until a reboot occurs
2206  */
2207 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2208 {
2209 	/* Generic WatchdogHandler, implement your own */
2210 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2211 	return;
2212 }
2213 
2214 DEFINE_INTERRUPT_HANDLER_NMI(WatchdogException)
2215 {
2216 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2217 	WatchdogHandler(regs);
2218 	return 0;
2219 }
2220 #endif
2221 
2222 /*
2223  * We enter here if we discover during exception entry that we are
2224  * running in supervisor mode with a userspace value in the stack pointer.
2225  */
2226 DEFINE_INTERRUPT_HANDLER(kernel_bad_stack)
2227 {
2228 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2229 	       regs->gpr[1], regs->nip);
2230 	die("Bad kernel stack pointer", regs, SIGABRT);
2231 }
2232 
2233 #ifdef CONFIG_PPC_EMULATED_STATS
2234 
2235 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2236 
2237 struct ppc_emulated ppc_emulated = {
2238 #ifdef CONFIG_ALTIVEC
2239 	WARN_EMULATED_SETUP(altivec),
2240 #endif
2241 	WARN_EMULATED_SETUP(dcba),
2242 	WARN_EMULATED_SETUP(dcbz),
2243 	WARN_EMULATED_SETUP(fp_pair),
2244 	WARN_EMULATED_SETUP(isel),
2245 	WARN_EMULATED_SETUP(mcrxr),
2246 	WARN_EMULATED_SETUP(mfpvr),
2247 	WARN_EMULATED_SETUP(multiple),
2248 	WARN_EMULATED_SETUP(popcntb),
2249 	WARN_EMULATED_SETUP(spe),
2250 	WARN_EMULATED_SETUP(string),
2251 	WARN_EMULATED_SETUP(sync),
2252 	WARN_EMULATED_SETUP(unaligned),
2253 #ifdef CONFIG_MATH_EMULATION
2254 	WARN_EMULATED_SETUP(math),
2255 #endif
2256 #ifdef CONFIG_VSX
2257 	WARN_EMULATED_SETUP(vsx),
2258 #endif
2259 #ifdef CONFIG_PPC64
2260 	WARN_EMULATED_SETUP(mfdscr),
2261 	WARN_EMULATED_SETUP(mtdscr),
2262 	WARN_EMULATED_SETUP(lq_stq),
2263 	WARN_EMULATED_SETUP(lxvw4x),
2264 	WARN_EMULATED_SETUP(lxvh8x),
2265 	WARN_EMULATED_SETUP(lxvd2x),
2266 	WARN_EMULATED_SETUP(lxvb16x),
2267 #endif
2268 };
2269 
2270 u32 ppc_warn_emulated;
2271 
2272 void ppc_warn_emulated_print(const char *type)
2273 {
2274 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2275 			    type);
2276 }
2277 
2278 static int __init ppc_warn_emulated_init(void)
2279 {
2280 	struct dentry *dir;
2281 	unsigned int i;
2282 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2283 
2284 	dir = debugfs_create_dir("emulated_instructions",
2285 				 arch_debugfs_dir);
2286 
2287 	debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated);
2288 
2289 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++)
2290 		debugfs_create_u32(entries[i].name, 0644, dir,
2291 				   (u32 *)&entries[i].val.counter);
2292 
2293 	return 0;
2294 }
2295 
2296 device_initcall(ppc_warn_emulated_init);
2297 
2298 #endif /* CONFIG_PPC_EMULATED_STATS */
2299