xref: /linux/arch/powerpc/kernel/traps.c (revision 288440de9e5fdb4a3ff73864850f080c1250fc81)
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 noinstr 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->msr &= ~MSR_RI;
437 	local_paca->hsrr_valid = 0;
438 	local_paca->srr_valid = 0;
439 #endif
440 }
441 DEFINE_INTERRUPT_HANDLER_NMI(system_reset_exception)
442 {
443 	unsigned long hsrr0, hsrr1;
444 	bool saved_hsrrs = false;
445 
446 	/*
447 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
448 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
449 	 * OPAL), so save them here and restore them before returning.
450 	 *
451 	 * Machine checks don't need to save HSRRs, as the real mode handler
452 	 * is careful to avoid them, and the regular handler is not delivered
453 	 * as an NMI.
454 	 */
455 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
456 		hsrr0 = mfspr(SPRN_HSRR0);
457 		hsrr1 = mfspr(SPRN_HSRR1);
458 		saved_hsrrs = true;
459 	}
460 
461 	hv_nmi_check_nonrecoverable(regs);
462 
463 	__this_cpu_inc(irq_stat.sreset_irqs);
464 
465 	/* See if any machine dependent calls */
466 	if (ppc_md.system_reset_exception) {
467 		if (ppc_md.system_reset_exception(regs))
468 			goto out;
469 	}
470 
471 	if (debugger(regs))
472 		goto out;
473 
474 	kmsg_dump(KMSG_DUMP_OOPS);
475 	/*
476 	 * A system reset is a request to dump, so we always send
477 	 * it through the crashdump code (if fadump or kdump are
478 	 * registered).
479 	 */
480 	crash_fadump(regs, "System Reset");
481 
482 	crash_kexec(regs);
483 
484 	/*
485 	 * We aren't the primary crash CPU. We need to send it
486 	 * to a holding pattern to avoid it ending up in the panic
487 	 * code.
488 	 */
489 	crash_kexec_secondary(regs);
490 
491 	/*
492 	 * No debugger or crash dump registered, print logs then
493 	 * panic.
494 	 */
495 	die("System Reset", regs, SIGABRT);
496 
497 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
498 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
499 	nmi_panic(regs, "System Reset");
500 
501 out:
502 #ifdef CONFIG_PPC_BOOK3S_64
503 	BUG_ON(get_paca()->in_nmi == 0);
504 	if (get_paca()->in_nmi > 1)
505 		die("Unrecoverable nested System Reset", regs, SIGABRT);
506 #endif
507 	/* Must die if the interrupt is not recoverable */
508 	if (regs_is_unrecoverable(regs)) {
509 		/* For the reason explained in die_mce, nmi_exit before die */
510 		nmi_exit();
511 		die("Unrecoverable System Reset", regs, SIGABRT);
512 	}
513 
514 	if (saved_hsrrs) {
515 		mtspr(SPRN_HSRR0, hsrr0);
516 		mtspr(SPRN_HSRR1, hsrr1);
517 	}
518 
519 	/* What should we do here? We could issue a shutdown or hard reset. */
520 
521 	return 0;
522 }
523 
524 /*
525  * I/O accesses can cause machine checks on powermacs.
526  * Check if the NIP corresponds to the address of a sync
527  * instruction for which there is an entry in the exception
528  * table.
529  *  -- paulus.
530  */
531 static inline int check_io_access(struct pt_regs *regs)
532 {
533 #ifdef CONFIG_PPC32
534 	unsigned long msr = regs->msr;
535 	const struct exception_table_entry *entry;
536 	unsigned int *nip = (unsigned int *)regs->nip;
537 
538 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
539 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
540 		/*
541 		 * Check that it's a sync instruction, or somewhere
542 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
543 		 * As the address is in the exception table
544 		 * we should be able to read the instr there.
545 		 * For the debug message, we look at the preceding
546 		 * load or store.
547 		 */
548 		if (*nip == PPC_RAW_NOP())
549 			nip -= 2;
550 		else if (*nip == PPC_RAW_ISYNC())
551 			--nip;
552 		if (*nip == PPC_RAW_SYNC() || get_op(*nip) == OP_TRAP) {
553 			unsigned int rb;
554 
555 			--nip;
556 			rb = (*nip >> 11) & 0x1f;
557 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
558 			       (*nip & 0x100)? "OUT to": "IN from",
559 			       regs->gpr[rb] - _IO_BASE, nip);
560 			regs_set_recoverable(regs);
561 			regs_set_return_ip(regs, extable_fixup(entry));
562 			return 1;
563 		}
564 	}
565 #endif /* CONFIG_PPC32 */
566 	return 0;
567 }
568 
569 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
570 /* On 4xx, the reason for the machine check or program exception
571    is in the ESR. */
572 #define get_reason(regs)	((regs)->esr)
573 #define REASON_FP		ESR_FP
574 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
575 #define REASON_PRIVILEGED	ESR_PPR
576 #define REASON_TRAP		ESR_PTR
577 #define REASON_PREFIXED		0
578 #define REASON_BOUNDARY		0
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 #define REASON_PREFIXED		SRR1_PREFIXED
594 #define REASON_BOUNDARY		SRR1_BOUNDARY
595 
596 #define single_stepping(regs)	((regs)->msr & MSR_SE)
597 #define clear_single_step(regs)	(regs_set_return_msr((regs), (regs)->msr & ~MSR_SE))
598 #define clear_br_trace(regs)	(regs_set_return_msr((regs), (regs)->msr & ~MSR_BE))
599 #endif
600 
601 #define inst_length(reason)	(((reason) & REASON_PREFIXED) ? 8 : 4)
602 
603 #if defined(CONFIG_E500)
604 int machine_check_e500mc(struct pt_regs *regs)
605 {
606 	unsigned long mcsr = mfspr(SPRN_MCSR);
607 	unsigned long pvr = mfspr(SPRN_PVR);
608 	unsigned long reason = mcsr;
609 	int recoverable = 1;
610 
611 	if (reason & MCSR_LD) {
612 		recoverable = fsl_rio_mcheck_exception(regs);
613 		if (recoverable == 1)
614 			goto silent_out;
615 	}
616 
617 	printk("Machine check in kernel mode.\n");
618 	printk("Caused by (from MCSR=%lx): ", reason);
619 
620 	if (reason & MCSR_MCP)
621 		pr_cont("Machine Check Signal\n");
622 
623 	if (reason & MCSR_ICPERR) {
624 		pr_cont("Instruction Cache Parity Error\n");
625 
626 		/*
627 		 * This is recoverable by invalidating the i-cache.
628 		 */
629 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
630 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
631 			;
632 
633 		/*
634 		 * This will generally be accompanied by an instruction
635 		 * fetch error report -- only treat MCSR_IF as fatal
636 		 * if it wasn't due to an L1 parity error.
637 		 */
638 		reason &= ~MCSR_IF;
639 	}
640 
641 	if (reason & MCSR_DCPERR_MC) {
642 		pr_cont("Data Cache Parity Error\n");
643 
644 		/*
645 		 * In write shadow mode we auto-recover from the error, but it
646 		 * may still get logged and cause a machine check.  We should
647 		 * only treat the non-write shadow case as non-recoverable.
648 		 */
649 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
650 		 * is not implemented but L1 data cache always runs in write
651 		 * shadow mode. Hence on data cache parity errors HW will
652 		 * automatically invalidate the L1 Data Cache.
653 		 */
654 		if (PVR_VER(pvr) != PVR_VER_E6500) {
655 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
656 				recoverable = 0;
657 		}
658 	}
659 
660 	if (reason & MCSR_L2MMU_MHIT) {
661 		pr_cont("Hit on multiple TLB entries\n");
662 		recoverable = 0;
663 	}
664 
665 	if (reason & MCSR_NMI)
666 		pr_cont("Non-maskable interrupt\n");
667 
668 	if (reason & MCSR_IF) {
669 		pr_cont("Instruction Fetch Error Report\n");
670 		recoverable = 0;
671 	}
672 
673 	if (reason & MCSR_LD) {
674 		pr_cont("Load Error Report\n");
675 		recoverable = 0;
676 	}
677 
678 	if (reason & MCSR_ST) {
679 		pr_cont("Store Error Report\n");
680 		recoverable = 0;
681 	}
682 
683 	if (reason & MCSR_LDG) {
684 		pr_cont("Guarded Load Error Report\n");
685 		recoverable = 0;
686 	}
687 
688 	if (reason & MCSR_TLBSYNC)
689 		pr_cont("Simultaneous tlbsync operations\n");
690 
691 	if (reason & MCSR_BSL2_ERR) {
692 		pr_cont("Level 2 Cache Error\n");
693 		recoverable = 0;
694 	}
695 
696 	if (reason & MCSR_MAV) {
697 		u64 addr;
698 
699 		addr = mfspr(SPRN_MCAR);
700 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
701 
702 		pr_cont("Machine Check %s Address: %#llx\n",
703 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
704 	}
705 
706 silent_out:
707 	mtspr(SPRN_MCSR, mcsr);
708 	return mfspr(SPRN_MCSR) == 0 && recoverable;
709 }
710 
711 int machine_check_e500(struct pt_regs *regs)
712 {
713 	unsigned long reason = mfspr(SPRN_MCSR);
714 
715 	if (reason & MCSR_BUS_RBERR) {
716 		if (fsl_rio_mcheck_exception(regs))
717 			return 1;
718 		if (fsl_pci_mcheck_exception(regs))
719 			return 1;
720 	}
721 
722 	printk("Machine check in kernel mode.\n");
723 	printk("Caused by (from MCSR=%lx): ", reason);
724 
725 	if (reason & MCSR_MCP)
726 		pr_cont("Machine Check Signal\n");
727 	if (reason & MCSR_ICPERR)
728 		pr_cont("Instruction Cache Parity Error\n");
729 	if (reason & MCSR_DCP_PERR)
730 		pr_cont("Data Cache Push Parity Error\n");
731 	if (reason & MCSR_DCPERR)
732 		pr_cont("Data Cache Parity Error\n");
733 	if (reason & MCSR_BUS_IAERR)
734 		pr_cont("Bus - Instruction Address Error\n");
735 	if (reason & MCSR_BUS_RAERR)
736 		pr_cont("Bus - Read Address Error\n");
737 	if (reason & MCSR_BUS_WAERR)
738 		pr_cont("Bus - Write Address Error\n");
739 	if (reason & MCSR_BUS_IBERR)
740 		pr_cont("Bus - Instruction Data Error\n");
741 	if (reason & MCSR_BUS_RBERR)
742 		pr_cont("Bus - Read Data Bus Error\n");
743 	if (reason & MCSR_BUS_WBERR)
744 		pr_cont("Bus - Write Data Bus Error\n");
745 	if (reason & MCSR_BUS_IPERR)
746 		pr_cont("Bus - Instruction Parity Error\n");
747 	if (reason & MCSR_BUS_RPERR)
748 		pr_cont("Bus - Read Parity Error\n");
749 
750 	return 0;
751 }
752 
753 int machine_check_generic(struct pt_regs *regs)
754 {
755 	return 0;
756 }
757 #elif defined(CONFIG_PPC32)
758 int machine_check_generic(struct pt_regs *regs)
759 {
760 	unsigned long reason = regs->msr;
761 
762 	printk("Machine check in kernel mode.\n");
763 	printk("Caused by (from SRR1=%lx): ", reason);
764 	switch (reason & 0x601F0000) {
765 	case 0x80000:
766 		pr_cont("Machine check signal\n");
767 		break;
768 	case 0x40000:
769 	case 0x140000:	/* 7450 MSS error and TEA */
770 		pr_cont("Transfer error ack signal\n");
771 		break;
772 	case 0x20000:
773 		pr_cont("Data parity error signal\n");
774 		break;
775 	case 0x10000:
776 		pr_cont("Address parity error signal\n");
777 		break;
778 	case 0x20000000:
779 		pr_cont("L1 Data Cache error\n");
780 		break;
781 	case 0x40000000:
782 		pr_cont("L1 Instruction Cache error\n");
783 		break;
784 	case 0x00100000:
785 		pr_cont("L2 data cache parity error\n");
786 		break;
787 	default:
788 		pr_cont("Unknown values in msr\n");
789 	}
790 	return 0;
791 }
792 #endif /* everything else */
793 
794 void die_mce(const char *str, struct pt_regs *regs, long err)
795 {
796 	/*
797 	 * The machine check wants to kill the interrupted context,
798 	 * but make_task_dead() checks for in_interrupt() and panics
799 	 * in that case, so exit the irq/nmi before calling die.
800 	 */
801 	if (in_nmi())
802 		nmi_exit();
803 	else
804 		irq_exit();
805 	die(str, regs, err);
806 }
807 
808 /*
809  * BOOK3S_64 does not usually call this handler as a non-maskable interrupt
810  * (it uses its own early real-mode handler to handle the MCE proper
811  * and then raises irq_work to call this handler when interrupts are
812  * enabled). The only time when this is not true is if the early handler
813  * is unrecoverable, then it does call this directly to try to get a
814  * message out.
815  */
816 static void __machine_check_exception(struct pt_regs *regs)
817 {
818 	int recover = 0;
819 
820 	__this_cpu_inc(irq_stat.mce_exceptions);
821 
822 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
823 
824 	/* See if any machine dependent calls. In theory, we would want
825 	 * to call the CPU first, and call the ppc_md. one if the CPU
826 	 * one returns a positive number. However there is existing code
827 	 * that assumes the board gets a first chance, so let's keep it
828 	 * that way for now and fix things later. --BenH.
829 	 */
830 	if (ppc_md.machine_check_exception)
831 		recover = ppc_md.machine_check_exception(regs);
832 	else if (cur_cpu_spec->machine_check)
833 		recover = cur_cpu_spec->machine_check(regs);
834 
835 	if (recover > 0)
836 		goto bail;
837 
838 	if (debugger_fault_handler(regs))
839 		goto bail;
840 
841 	if (check_io_access(regs))
842 		goto bail;
843 
844 	die_mce("Machine check", regs, SIGBUS);
845 
846 bail:
847 	/* Must die if the interrupt is not recoverable */
848 	if (regs_is_unrecoverable(regs))
849 		die_mce("Unrecoverable Machine check", regs, SIGBUS);
850 }
851 
852 #ifdef CONFIG_PPC_BOOK3S_64
853 DEFINE_INTERRUPT_HANDLER_ASYNC(machine_check_exception_async)
854 {
855 	__machine_check_exception(regs);
856 }
857 #endif
858 DEFINE_INTERRUPT_HANDLER_NMI(machine_check_exception)
859 {
860 	__machine_check_exception(regs);
861 
862 	return 0;
863 }
864 
865 DEFINE_INTERRUPT_HANDLER(SMIException) /* async? */
866 {
867 	die("System Management Interrupt", regs, SIGABRT);
868 }
869 
870 #ifdef CONFIG_VSX
871 static void p9_hmi_special_emu(struct pt_regs *regs)
872 {
873 	unsigned int ra, rb, t, i, sel, instr, rc;
874 	const void __user *addr;
875 	u8 vbuf[16] __aligned(16), *vdst;
876 	unsigned long ea, msr, msr_mask;
877 	bool swap;
878 
879 	if (__get_user(instr, (unsigned int __user *)regs->nip))
880 		return;
881 
882 	/*
883 	 * lxvb16x	opcode: 0x7c0006d8
884 	 * lxvd2x	opcode: 0x7c000698
885 	 * lxvh8x	opcode: 0x7c000658
886 	 * lxvw4x	opcode: 0x7c000618
887 	 */
888 	if ((instr & 0xfc00073e) != 0x7c000618) {
889 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
890 			 " instr=%08x\n",
891 			 smp_processor_id(), current->comm, current->pid,
892 			 regs->nip, instr);
893 		return;
894 	}
895 
896 	/* Grab vector registers into the task struct */
897 	msr = regs->msr; /* Grab msr before we flush the bits */
898 	flush_vsx_to_thread(current);
899 	enable_kernel_altivec();
900 
901 	/*
902 	 * Is userspace running with a different endian (this is rare but
903 	 * not impossible)
904 	 */
905 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
906 
907 	/* Decode the instruction */
908 	ra = (instr >> 16) & 0x1f;
909 	rb = (instr >> 11) & 0x1f;
910 	t = (instr >> 21) & 0x1f;
911 	if (instr & 1)
912 		vdst = (u8 *)&current->thread.vr_state.vr[t];
913 	else
914 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
915 
916 	/* Grab the vector address */
917 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
918 	if (is_32bit_task())
919 		ea &= 0xfffffffful;
920 	addr = (__force const void __user *)ea;
921 
922 	/* Check it */
923 	if (!access_ok(addr, 16)) {
924 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
925 			 " instr=%08x addr=%016lx\n",
926 			 smp_processor_id(), current->comm, current->pid,
927 			 regs->nip, instr, (unsigned long)addr);
928 		return;
929 	}
930 
931 	/* Read the vector */
932 	rc = 0;
933 	if ((unsigned long)addr & 0xfUL)
934 		/* unaligned case */
935 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
936 	else
937 		__get_user_atomic_128_aligned(vbuf, addr, rc);
938 	if (rc) {
939 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
940 			 " instr=%08x addr=%016lx\n",
941 			 smp_processor_id(), current->comm, current->pid,
942 			 regs->nip, instr, (unsigned long)addr);
943 		return;
944 	}
945 
946 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
947 		 " instr=%08x addr=%016lx\n",
948 		 smp_processor_id(), current->comm, current->pid, regs->nip,
949 		 instr, (unsigned long) addr);
950 
951 	/* Grab instruction "selector" */
952 	sel = (instr >> 6) & 3;
953 
954 	/*
955 	 * Check to make sure the facility is actually enabled. This
956 	 * could happen if we get a false positive hit.
957 	 *
958 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
959 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
960 	 */
961 	msr_mask = MSR_VSX;
962 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
963 		msr_mask = MSR_VEC;
964 	if (!(msr & msr_mask)) {
965 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
966 			 " instr=%08x msr:%016lx\n",
967 			 smp_processor_id(), current->comm, current->pid,
968 			 regs->nip, instr, msr);
969 		return;
970 	}
971 
972 	/* Do logging here before we modify sel based on endian */
973 	switch (sel) {
974 	case 0:	/* lxvw4x */
975 		PPC_WARN_EMULATED(lxvw4x, regs);
976 		break;
977 	case 1: /* lxvh8x */
978 		PPC_WARN_EMULATED(lxvh8x, regs);
979 		break;
980 	case 2: /* lxvd2x */
981 		PPC_WARN_EMULATED(lxvd2x, regs);
982 		break;
983 	case 3: /* lxvb16x */
984 		PPC_WARN_EMULATED(lxvb16x, regs);
985 		break;
986 	}
987 
988 #ifdef __LITTLE_ENDIAN__
989 	/*
990 	 * An LE kernel stores the vector in the task struct as an LE
991 	 * byte array (effectively swapping both the components and
992 	 * the content of the components). Those instructions expect
993 	 * the components to remain in ascending address order, so we
994 	 * swap them back.
995 	 *
996 	 * If we are running a BE user space, the expectation is that
997 	 * of a simple memcpy, so forcing the emulation to look like
998 	 * a lxvb16x should do the trick.
999 	 */
1000 	if (swap)
1001 		sel = 3;
1002 
1003 	switch (sel) {
1004 	case 0:	/* lxvw4x */
1005 		for (i = 0; i < 4; i++)
1006 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1007 		break;
1008 	case 1: /* lxvh8x */
1009 		for (i = 0; i < 8; i++)
1010 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1011 		break;
1012 	case 2: /* lxvd2x */
1013 		for (i = 0; i < 2; i++)
1014 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1015 		break;
1016 	case 3: /* lxvb16x */
1017 		for (i = 0; i < 16; i++)
1018 			vdst[i] = vbuf[15-i];
1019 		break;
1020 	}
1021 #else /* __LITTLE_ENDIAN__ */
1022 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1023 	if (!swap)
1024 		sel = 3;
1025 
1026 	/* Otherwise, we need to swap the content of the components */
1027 	switch (sel) {
1028 	case 0:	/* lxvw4x */
1029 		for (i = 0; i < 4; i++)
1030 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1031 		break;
1032 	case 1: /* lxvh8x */
1033 		for (i = 0; i < 8; i++)
1034 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1035 		break;
1036 	case 2: /* lxvd2x */
1037 		for (i = 0; i < 2; i++)
1038 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1039 		break;
1040 	case 3: /* lxvb16x */
1041 		memcpy(vdst, vbuf, 16);
1042 		break;
1043 	}
1044 #endif /* !__LITTLE_ENDIAN__ */
1045 
1046 	/* Go to next instruction */
1047 	regs_add_return_ip(regs, 4);
1048 }
1049 #endif /* CONFIG_VSX */
1050 
1051 DEFINE_INTERRUPT_HANDLER_ASYNC(handle_hmi_exception)
1052 {
1053 	struct pt_regs *old_regs;
1054 
1055 	old_regs = set_irq_regs(regs);
1056 
1057 #ifdef CONFIG_VSX
1058 	/* Real mode flagged P9 special emu is needed */
1059 	if (local_paca->hmi_p9_special_emu) {
1060 		local_paca->hmi_p9_special_emu = 0;
1061 
1062 		/*
1063 		 * We don't want to take page faults while doing the
1064 		 * emulation, we just replay the instruction if necessary.
1065 		 */
1066 		pagefault_disable();
1067 		p9_hmi_special_emu(regs);
1068 		pagefault_enable();
1069 	}
1070 #endif /* CONFIG_VSX */
1071 
1072 	if (ppc_md.handle_hmi_exception)
1073 		ppc_md.handle_hmi_exception(regs);
1074 
1075 	set_irq_regs(old_regs);
1076 }
1077 
1078 DEFINE_INTERRUPT_HANDLER(unknown_exception)
1079 {
1080 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1081 	       regs->nip, regs->msr, regs->trap);
1082 
1083 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1084 }
1085 
1086 DEFINE_INTERRUPT_HANDLER_ASYNC(unknown_async_exception)
1087 {
1088 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1089 	       regs->nip, regs->msr, regs->trap);
1090 
1091 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1092 }
1093 
1094 DEFINE_INTERRUPT_HANDLER_NMI(unknown_nmi_exception)
1095 {
1096 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1097 	       regs->nip, regs->msr, regs->trap);
1098 
1099 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1100 
1101 	return 0;
1102 }
1103 
1104 DEFINE_INTERRUPT_HANDLER(instruction_breakpoint_exception)
1105 {
1106 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1107 					5, SIGTRAP) == NOTIFY_STOP)
1108 		return;
1109 	if (debugger_iabr_match(regs))
1110 		return;
1111 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1112 }
1113 
1114 DEFINE_INTERRUPT_HANDLER(RunModeException)
1115 {
1116 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1117 }
1118 
1119 static void __single_step_exception(struct pt_regs *regs)
1120 {
1121 	clear_single_step(regs);
1122 	clear_br_trace(regs);
1123 
1124 	if (kprobe_post_handler(regs))
1125 		return;
1126 
1127 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1128 					5, SIGTRAP) == NOTIFY_STOP)
1129 		return;
1130 	if (debugger_sstep(regs))
1131 		return;
1132 
1133 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1134 }
1135 
1136 DEFINE_INTERRUPT_HANDLER(single_step_exception)
1137 {
1138 	__single_step_exception(regs);
1139 }
1140 
1141 /*
1142  * After we have successfully emulated an instruction, we have to
1143  * check if the instruction was being single-stepped, and if so,
1144  * pretend we got a single-step exception.  This was pointed out
1145  * by Kumar Gala.  -- paulus
1146  */
1147 static void emulate_single_step(struct pt_regs *regs)
1148 {
1149 	if (single_stepping(regs))
1150 		__single_step_exception(regs);
1151 }
1152 
1153 static inline int __parse_fpscr(unsigned long fpscr)
1154 {
1155 	int ret = FPE_FLTUNK;
1156 
1157 	/* Invalid operation */
1158 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1159 		ret = FPE_FLTINV;
1160 
1161 	/* Overflow */
1162 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1163 		ret = FPE_FLTOVF;
1164 
1165 	/* Underflow */
1166 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1167 		ret = FPE_FLTUND;
1168 
1169 	/* Divide by zero */
1170 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1171 		ret = FPE_FLTDIV;
1172 
1173 	/* Inexact result */
1174 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1175 		ret = FPE_FLTRES;
1176 
1177 	return ret;
1178 }
1179 
1180 static void parse_fpe(struct pt_regs *regs)
1181 {
1182 	int code = 0;
1183 
1184 	flush_fp_to_thread(current);
1185 
1186 #ifdef CONFIG_PPC_FPU_REGS
1187 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1188 #endif
1189 
1190 	_exception(SIGFPE, regs, code, regs->nip);
1191 }
1192 
1193 /*
1194  * Illegal instruction emulation support.  Originally written to
1195  * provide the PVR to user applications using the mfspr rd, PVR.
1196  * Return non-zero if we can't emulate, or -EFAULT if the associated
1197  * memory access caused an access fault.  Return zero on success.
1198  *
1199  * There are a couple of ways to do this, either "decode" the instruction
1200  * or directly match lots of bits.  In this case, matching lots of
1201  * bits is faster and easier.
1202  *
1203  */
1204 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1205 {
1206 	u8 rT = (instword >> 21) & 0x1f;
1207 	u8 rA = (instword >> 16) & 0x1f;
1208 	u8 NB_RB = (instword >> 11) & 0x1f;
1209 	u32 num_bytes;
1210 	unsigned long EA;
1211 	int pos = 0;
1212 
1213 	/* Early out if we are an invalid form of lswx */
1214 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1215 		if ((rT == rA) || (rT == NB_RB))
1216 			return -EINVAL;
1217 
1218 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1219 
1220 	switch (instword & PPC_INST_STRING_MASK) {
1221 		case PPC_INST_LSWX:
1222 		case PPC_INST_STSWX:
1223 			EA += NB_RB;
1224 			num_bytes = regs->xer & 0x7f;
1225 			break;
1226 		case PPC_INST_LSWI:
1227 		case PPC_INST_STSWI:
1228 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1229 			break;
1230 		default:
1231 			return -EINVAL;
1232 	}
1233 
1234 	while (num_bytes != 0)
1235 	{
1236 		u8 val;
1237 		u32 shift = 8 * (3 - (pos & 0x3));
1238 
1239 		/* if process is 32-bit, clear upper 32 bits of EA */
1240 		if ((regs->msr & MSR_64BIT) == 0)
1241 			EA &= 0xFFFFFFFF;
1242 
1243 		switch ((instword & PPC_INST_STRING_MASK)) {
1244 			case PPC_INST_LSWX:
1245 			case PPC_INST_LSWI:
1246 				if (get_user(val, (u8 __user *)EA))
1247 					return -EFAULT;
1248 				/* first time updating this reg,
1249 				 * zero it out */
1250 				if (pos == 0)
1251 					regs->gpr[rT] = 0;
1252 				regs->gpr[rT] |= val << shift;
1253 				break;
1254 			case PPC_INST_STSWI:
1255 			case PPC_INST_STSWX:
1256 				val = regs->gpr[rT] >> shift;
1257 				if (put_user(val, (u8 __user *)EA))
1258 					return -EFAULT;
1259 				break;
1260 		}
1261 		/* move EA to next address */
1262 		EA += 1;
1263 		num_bytes--;
1264 
1265 		/* manage our position within the register */
1266 		if (++pos == 4) {
1267 			pos = 0;
1268 			if (++rT == 32)
1269 				rT = 0;
1270 		}
1271 	}
1272 
1273 	return 0;
1274 }
1275 
1276 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1277 {
1278 	u32 ra,rs;
1279 	unsigned long tmp;
1280 
1281 	ra = (instword >> 16) & 0x1f;
1282 	rs = (instword >> 21) & 0x1f;
1283 
1284 	tmp = regs->gpr[rs];
1285 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1286 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1287 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1288 	regs->gpr[ra] = tmp;
1289 
1290 	return 0;
1291 }
1292 
1293 static int emulate_isel(struct pt_regs *regs, u32 instword)
1294 {
1295 	u8 rT = (instword >> 21) & 0x1f;
1296 	u8 rA = (instword >> 16) & 0x1f;
1297 	u8 rB = (instword >> 11) & 0x1f;
1298 	u8 BC = (instword >> 6) & 0x1f;
1299 	u8 bit;
1300 	unsigned long tmp;
1301 
1302 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1303 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1304 
1305 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1306 
1307 	return 0;
1308 }
1309 
1310 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1311 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1312 {
1313         /* If we're emulating a load/store in an active transaction, we cannot
1314          * emulate it as the kernel operates in transaction suspended context.
1315          * We need to abort the transaction.  This creates a persistent TM
1316          * abort so tell the user what caused it with a new code.
1317 	 */
1318 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1319 		tm_enable();
1320 		tm_abort(cause);
1321 		return true;
1322 	}
1323 	return false;
1324 }
1325 #else
1326 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1327 {
1328 	return false;
1329 }
1330 #endif
1331 
1332 static int emulate_instruction(struct pt_regs *regs)
1333 {
1334 	u32 instword;
1335 	u32 rd;
1336 
1337 	if (!user_mode(regs))
1338 		return -EINVAL;
1339 
1340 	if (get_user(instword, (u32 __user *)(regs->nip)))
1341 		return -EFAULT;
1342 
1343 	/* Emulate the mfspr rD, PVR. */
1344 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1345 		PPC_WARN_EMULATED(mfpvr, regs);
1346 		rd = (instword >> 21) & 0x1f;
1347 		regs->gpr[rd] = mfspr(SPRN_PVR);
1348 		return 0;
1349 	}
1350 
1351 	/* Emulating the dcba insn is just a no-op.  */
1352 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1353 		PPC_WARN_EMULATED(dcba, regs);
1354 		return 0;
1355 	}
1356 
1357 	/* Emulate the mcrxr insn.  */
1358 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1359 		int shift = (instword >> 21) & 0x1c;
1360 		unsigned long msk = 0xf0000000UL >> shift;
1361 
1362 		PPC_WARN_EMULATED(mcrxr, regs);
1363 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1364 		regs->xer &= ~0xf0000000UL;
1365 		return 0;
1366 	}
1367 
1368 	/* Emulate load/store string insn. */
1369 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1370 		if (tm_abort_check(regs,
1371 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1372 			return -EINVAL;
1373 		PPC_WARN_EMULATED(string, regs);
1374 		return emulate_string_inst(regs, instword);
1375 	}
1376 
1377 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1378 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1379 		PPC_WARN_EMULATED(popcntb, regs);
1380 		return emulate_popcntb_inst(regs, instword);
1381 	}
1382 
1383 	/* Emulate isel (Integer Select) instruction */
1384 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1385 		PPC_WARN_EMULATED(isel, regs);
1386 		return emulate_isel(regs, instword);
1387 	}
1388 
1389 	/* Emulate sync instruction variants */
1390 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1391 		PPC_WARN_EMULATED(sync, regs);
1392 		asm volatile("sync");
1393 		return 0;
1394 	}
1395 
1396 #ifdef CONFIG_PPC64
1397 	/* Emulate the mfspr rD, DSCR. */
1398 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1399 		PPC_INST_MFSPR_DSCR_USER) ||
1400 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1401 		PPC_INST_MFSPR_DSCR)) &&
1402 			cpu_has_feature(CPU_FTR_DSCR)) {
1403 		PPC_WARN_EMULATED(mfdscr, regs);
1404 		rd = (instword >> 21) & 0x1f;
1405 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1406 		return 0;
1407 	}
1408 	/* Emulate the mtspr DSCR, rD. */
1409 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1410 		PPC_INST_MTSPR_DSCR_USER) ||
1411 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1412 		PPC_INST_MTSPR_DSCR)) &&
1413 			cpu_has_feature(CPU_FTR_DSCR)) {
1414 		PPC_WARN_EMULATED(mtdscr, regs);
1415 		rd = (instword >> 21) & 0x1f;
1416 		current->thread.dscr = regs->gpr[rd];
1417 		current->thread.dscr_inherit = 1;
1418 		mtspr(SPRN_DSCR, current->thread.dscr);
1419 		return 0;
1420 	}
1421 #endif
1422 
1423 	return -EINVAL;
1424 }
1425 
1426 int is_valid_bugaddr(unsigned long addr)
1427 {
1428 	return is_kernel_addr(addr);
1429 }
1430 
1431 #ifdef CONFIG_MATH_EMULATION
1432 static int emulate_math(struct pt_regs *regs)
1433 {
1434 	int ret;
1435 
1436 	ret = do_mathemu(regs);
1437 	if (ret >= 0)
1438 		PPC_WARN_EMULATED(math, regs);
1439 
1440 	switch (ret) {
1441 	case 0:
1442 		emulate_single_step(regs);
1443 		return 0;
1444 	case 1: {
1445 			int code = 0;
1446 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1447 			_exception(SIGFPE, regs, code, regs->nip);
1448 			return 0;
1449 		}
1450 	case -EFAULT:
1451 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1452 		return 0;
1453 	}
1454 
1455 	return -1;
1456 }
1457 #else
1458 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1459 #endif
1460 
1461 static void do_program_check(struct pt_regs *regs)
1462 {
1463 	unsigned int reason = get_reason(regs);
1464 
1465 	/* We can now get here via a FP Unavailable exception if the core
1466 	 * has no FPU, in that case the reason flags will be 0 */
1467 
1468 	if (reason & REASON_FP) {
1469 		/* IEEE FP exception */
1470 		parse_fpe(regs);
1471 		return;
1472 	}
1473 	if (reason & REASON_TRAP) {
1474 		unsigned long bugaddr;
1475 		/* Debugger is first in line to stop recursive faults in
1476 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1477 		if (debugger_bpt(regs))
1478 			return;
1479 
1480 		if (kprobe_handler(regs))
1481 			return;
1482 
1483 		/* trap exception */
1484 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1485 				== NOTIFY_STOP)
1486 			return;
1487 
1488 		bugaddr = regs->nip;
1489 		/*
1490 		 * Fixup bugaddr for BUG_ON() in real mode
1491 		 */
1492 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1493 			bugaddr += PAGE_OFFSET;
1494 
1495 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1496 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1497 			const struct exception_table_entry *entry;
1498 
1499 			entry = search_exception_tables(bugaddr);
1500 			if (entry) {
1501 				regs_set_return_ip(regs, extable_fixup(entry) + regs->nip - bugaddr);
1502 				return;
1503 			}
1504 		}
1505 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1506 		return;
1507 	}
1508 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1509 	if (reason & REASON_TM) {
1510 		/* This is a TM "Bad Thing Exception" program check.
1511 		 * This occurs when:
1512 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1513 		 *    transition in TM states.
1514 		 * -  A trechkpt is attempted when transactional.
1515 		 * -  A treclaim is attempted when non transactional.
1516 		 * -  A tend is illegally attempted.
1517 		 * -  writing a TM SPR when transactional.
1518 		 *
1519 		 * If usermode caused this, it's done something illegal and
1520 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1521 		 * operand to distinguish from the instruction just being bad
1522 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1523 		 * illegal /placement/ of a valid instruction.
1524 		 */
1525 		if (user_mode(regs)) {
1526 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1527 			return;
1528 		} else {
1529 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1530 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1531 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1532 			die("Unrecoverable exception", regs, SIGABRT);
1533 		}
1534 	}
1535 #endif
1536 
1537 	/*
1538 	 * If we took the program check in the kernel skip down to sending a
1539 	 * SIGILL. The subsequent cases all relate to emulating instructions
1540 	 * which we should only do for userspace. We also do not want to enable
1541 	 * interrupts for kernel faults because that might lead to further
1542 	 * faults, and loose the context of the original exception.
1543 	 */
1544 	if (!user_mode(regs))
1545 		goto sigill;
1546 
1547 	interrupt_cond_local_irq_enable(regs);
1548 
1549 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1550 	 * but there seems to be a hardware bug on the 405GP (RevD)
1551 	 * that means ESR is sometimes set incorrectly - either to
1552 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1553 	 * hardware people - not sure if it can happen on any illegal
1554 	 * instruction or only on FP instructions, whether there is a
1555 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1556 	 */
1557 	if (!emulate_math(regs))
1558 		return;
1559 
1560 	/* Try to emulate it if we should. */
1561 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1562 		switch (emulate_instruction(regs)) {
1563 		case 0:
1564 			regs_add_return_ip(regs, 4);
1565 			emulate_single_step(regs);
1566 			return;
1567 		case -EFAULT:
1568 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1569 			return;
1570 		}
1571 	}
1572 
1573 sigill:
1574 	if (reason & REASON_PRIVILEGED)
1575 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1576 	else
1577 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1578 
1579 }
1580 
1581 DEFINE_INTERRUPT_HANDLER(program_check_exception)
1582 {
1583 	do_program_check(regs);
1584 }
1585 
1586 /*
1587  * This occurs when running in hypervisor mode on POWER6 or later
1588  * and an illegal instruction is encountered.
1589  */
1590 DEFINE_INTERRUPT_HANDLER(emulation_assist_interrupt)
1591 {
1592 	regs_set_return_msr(regs, regs->msr | REASON_ILLEGAL);
1593 	do_program_check(regs);
1594 }
1595 
1596 DEFINE_INTERRUPT_HANDLER(alignment_exception)
1597 {
1598 	int sig, code, fixed = 0;
1599 	unsigned long  reason;
1600 
1601 	interrupt_cond_local_irq_enable(regs);
1602 
1603 	reason = get_reason(regs);
1604 	if (reason & REASON_BOUNDARY) {
1605 		sig = SIGBUS;
1606 		code = BUS_ADRALN;
1607 		goto bad;
1608 	}
1609 
1610 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1611 		return;
1612 
1613 	/* we don't implement logging of alignment exceptions */
1614 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1615 		fixed = fix_alignment(regs);
1616 
1617 	if (fixed == 1) {
1618 		/* skip over emulated instruction */
1619 		regs_add_return_ip(regs, inst_length(reason));
1620 		emulate_single_step(regs);
1621 		return;
1622 	}
1623 
1624 	/* Operand address was bad */
1625 	if (fixed == -EFAULT) {
1626 		sig = SIGSEGV;
1627 		code = SEGV_ACCERR;
1628 	} else {
1629 		sig = SIGBUS;
1630 		code = BUS_ADRALN;
1631 	}
1632 bad:
1633 	if (user_mode(regs))
1634 		_exception(sig, regs, code, regs->dar);
1635 	else
1636 		bad_page_fault(regs, sig);
1637 }
1638 
1639 DEFINE_INTERRUPT_HANDLER(stack_overflow_exception)
1640 {
1641 	die("Kernel stack overflow", regs, SIGSEGV);
1642 }
1643 
1644 DEFINE_INTERRUPT_HANDLER(kernel_fp_unavailable_exception)
1645 {
1646 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1647 			  "%lx at %lx\n", regs->trap, regs->nip);
1648 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1649 }
1650 
1651 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_exception)
1652 {
1653 	if (user_mode(regs)) {
1654 		/* A user program has executed an altivec instruction,
1655 		   but this kernel doesn't support altivec. */
1656 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1657 		return;
1658 	}
1659 
1660 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1661 			"%lx at %lx\n", regs->trap, regs->nip);
1662 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1663 }
1664 
1665 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_exception)
1666 {
1667 	if (user_mode(regs)) {
1668 		/* A user program has executed an vsx instruction,
1669 		   but this kernel doesn't support vsx. */
1670 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1671 		return;
1672 	}
1673 
1674 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1675 			"%lx at %lx\n", regs->trap, regs->nip);
1676 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1677 }
1678 
1679 #ifdef CONFIG_PPC_BOOK3S_64
1680 static void tm_unavailable(struct pt_regs *regs)
1681 {
1682 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1683 	if (user_mode(regs)) {
1684 		current->thread.load_tm++;
1685 		regs_set_return_msr(regs, regs->msr | MSR_TM);
1686 		tm_enable();
1687 		tm_restore_sprs(&current->thread);
1688 		return;
1689 	}
1690 #endif
1691 	pr_emerg("Unrecoverable TM Unavailable Exception "
1692 			"%lx at %lx\n", regs->trap, regs->nip);
1693 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1694 }
1695 
1696 DEFINE_INTERRUPT_HANDLER(facility_unavailable_exception)
1697 {
1698 	static char *facility_strings[] = {
1699 		[FSCR_FP_LG] = "FPU",
1700 		[FSCR_VECVSX_LG] = "VMX/VSX",
1701 		[FSCR_DSCR_LG] = "DSCR",
1702 		[FSCR_PM_LG] = "PMU SPRs",
1703 		[FSCR_BHRB_LG] = "BHRB",
1704 		[FSCR_TM_LG] = "TM",
1705 		[FSCR_EBB_LG] = "EBB",
1706 		[FSCR_TAR_LG] = "TAR",
1707 		[FSCR_MSGP_LG] = "MSGP",
1708 		[FSCR_SCV_LG] = "SCV",
1709 		[FSCR_PREFIX_LG] = "PREFIX",
1710 	};
1711 	char *facility = "unknown";
1712 	u64 value;
1713 	u32 instword, rd;
1714 	u8 status;
1715 	bool hv;
1716 
1717 	hv = (TRAP(regs) == INTERRUPT_H_FAC_UNAVAIL);
1718 	if (hv)
1719 		value = mfspr(SPRN_HFSCR);
1720 	else
1721 		value = mfspr(SPRN_FSCR);
1722 
1723 	status = value >> 56;
1724 	if ((hv || status >= 2) &&
1725 	    (status < ARRAY_SIZE(facility_strings)) &&
1726 	    facility_strings[status])
1727 		facility = facility_strings[status];
1728 
1729 	/* We should not have taken this interrupt in kernel */
1730 	if (!user_mode(regs)) {
1731 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1732 			 facility, status, regs->nip);
1733 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1734 	}
1735 
1736 	interrupt_cond_local_irq_enable(regs);
1737 
1738 	if (status == FSCR_DSCR_LG) {
1739 		/*
1740 		 * User is accessing the DSCR register using the problem
1741 		 * state only SPR number (0x03) either through a mfspr or
1742 		 * a mtspr instruction. If it is a write attempt through
1743 		 * a mtspr, then we set the inherit bit. This also allows
1744 		 * the user to write or read the register directly in the
1745 		 * future by setting via the FSCR DSCR bit. But in case it
1746 		 * is a read DSCR attempt through a mfspr instruction, we
1747 		 * just emulate the instruction instead. This code path will
1748 		 * always emulate all the mfspr instructions till the user
1749 		 * has attempted at least one mtspr instruction. This way it
1750 		 * preserves the same behaviour when the user is accessing
1751 		 * the DSCR through privilege level only SPR number (0x11)
1752 		 * which is emulated through illegal instruction exception.
1753 		 * We always leave HFSCR DSCR set.
1754 		 */
1755 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1756 			pr_err("Failed to fetch the user instruction\n");
1757 			return;
1758 		}
1759 
1760 		/* Write into DSCR (mtspr 0x03, RS) */
1761 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1762 				== PPC_INST_MTSPR_DSCR_USER) {
1763 			rd = (instword >> 21) & 0x1f;
1764 			current->thread.dscr = regs->gpr[rd];
1765 			current->thread.dscr_inherit = 1;
1766 			current->thread.fscr |= FSCR_DSCR;
1767 			mtspr(SPRN_FSCR, current->thread.fscr);
1768 		}
1769 
1770 		/* Read from DSCR (mfspr RT, 0x03) */
1771 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1772 				== PPC_INST_MFSPR_DSCR_USER) {
1773 			if (emulate_instruction(regs)) {
1774 				pr_err("DSCR based mfspr emulation failed\n");
1775 				return;
1776 			}
1777 			regs_add_return_ip(regs, 4);
1778 			emulate_single_step(regs);
1779 		}
1780 		return;
1781 	}
1782 
1783 	if (status == FSCR_TM_LG) {
1784 		/*
1785 		 * If we're here then the hardware is TM aware because it
1786 		 * generated an exception with FSRM_TM set.
1787 		 *
1788 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1789 		 * told us not to do TM, or the kernel is not built with TM
1790 		 * support.
1791 		 *
1792 		 * If both of those things are true, then userspace can spam the
1793 		 * console by triggering the printk() below just by continually
1794 		 * doing tbegin (or any TM instruction). So in that case just
1795 		 * send the process a SIGILL immediately.
1796 		 */
1797 		if (!cpu_has_feature(CPU_FTR_TM))
1798 			goto out;
1799 
1800 		tm_unavailable(regs);
1801 		return;
1802 	}
1803 
1804 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1805 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1806 
1807 out:
1808 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1809 }
1810 #endif
1811 
1812 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1813 
1814 DEFINE_INTERRUPT_HANDLER(fp_unavailable_tm)
1815 {
1816 	/* Note:  This does not handle any kind of FP laziness. */
1817 
1818 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1819 		 regs->nip, regs->msr);
1820 
1821         /* We can only have got here if the task started using FP after
1822          * beginning the transaction.  So, the transactional regs are just a
1823          * copy of the checkpointed ones.  But, we still need to recheckpoint
1824          * as we're enabling FP for the process; it will return, abort the
1825          * transaction, and probably retry but now with FP enabled.  So the
1826          * checkpointed FP registers need to be loaded.
1827 	 */
1828 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1829 
1830 	/*
1831 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1832 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1833 	 *
1834 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1835 	 * recheckpoint.
1836 	 */
1837 
1838 	/* Enable FP for the task: */
1839 	current->thread.load_fp = 1;
1840 
1841 	/*
1842 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1843 	 */
1844 	tm_recheckpoint(&current->thread);
1845 }
1846 
1847 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_tm)
1848 {
1849 	/* See the comments in fp_unavailable_tm().  This function operates
1850 	 * the same way.
1851 	 */
1852 
1853 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1854 		 "MSR=%lx\n",
1855 		 regs->nip, regs->msr);
1856 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1857 	current->thread.load_vec = 1;
1858 	tm_recheckpoint(&current->thread);
1859 	current->thread.used_vr = 1;
1860 }
1861 
1862 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_tm)
1863 {
1864 	/* See the comments in fp_unavailable_tm().  This works similarly,
1865 	 * though we're loading both FP and VEC registers in here.
1866 	 *
1867 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1868 	 * regs.  Either way, set MSR_VSX.
1869 	 */
1870 
1871 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1872 		 "MSR=%lx\n",
1873 		 regs->nip, regs->msr);
1874 
1875 	current->thread.used_vsr = 1;
1876 
1877 	/* This reclaims FP and/or VR regs if they're already enabled */
1878 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1879 
1880 	current->thread.load_vec = 1;
1881 	current->thread.load_fp = 1;
1882 
1883 	tm_recheckpoint(&current->thread);
1884 }
1885 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1886 
1887 #ifdef CONFIG_PPC64
1888 DECLARE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi);
1889 DEFINE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi)
1890 {
1891 	__this_cpu_inc(irq_stat.pmu_irqs);
1892 
1893 	perf_irq(regs);
1894 
1895 	return 0;
1896 }
1897 #endif
1898 
1899 DECLARE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async);
1900 DEFINE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async)
1901 {
1902 	__this_cpu_inc(irq_stat.pmu_irqs);
1903 
1904 	perf_irq(regs);
1905 }
1906 
1907 DEFINE_INTERRUPT_HANDLER_RAW(performance_monitor_exception)
1908 {
1909 	/*
1910 	 * On 64-bit, if perf interrupts hit in a local_irq_disable
1911 	 * (soft-masked) region, we consider them as NMIs. This is required to
1912 	 * prevent hash faults on user addresses when reading callchains (and
1913 	 * looks better from an irq tracing perspective).
1914 	 */
1915 	if (IS_ENABLED(CONFIG_PPC64) && unlikely(arch_irq_disabled_regs(regs)))
1916 		performance_monitor_exception_nmi(regs);
1917 	else
1918 		performance_monitor_exception_async(regs);
1919 
1920 	return 0;
1921 }
1922 
1923 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1924 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1925 {
1926 	int changed = 0;
1927 	/*
1928 	 * Determine the cause of the debug event, clear the
1929 	 * event flags and send a trap to the handler. Torez
1930 	 */
1931 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1932 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1933 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1934 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1935 #endif
1936 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1937 			     5);
1938 		changed |= 0x01;
1939 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1940 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1941 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1942 			     6);
1943 		changed |= 0x01;
1944 	}  else if (debug_status & DBSR_IAC1) {
1945 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1946 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1947 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1948 			     1);
1949 		changed |= 0x01;
1950 	}  else if (debug_status & DBSR_IAC2) {
1951 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1952 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1953 			     2);
1954 		changed |= 0x01;
1955 	}  else if (debug_status & DBSR_IAC3) {
1956 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1957 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1958 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1959 			     3);
1960 		changed |= 0x01;
1961 	}  else if (debug_status & DBSR_IAC4) {
1962 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1963 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1964 			     4);
1965 		changed |= 0x01;
1966 	}
1967 	/*
1968 	 * At the point this routine was called, the MSR(DE) was turned off.
1969 	 * Check all other debug flags and see if that bit needs to be turned
1970 	 * back on or not.
1971 	 */
1972 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1973 			       current->thread.debug.dbcr1))
1974 		regs_set_return_msr(regs, regs->msr | MSR_DE);
1975 	else
1976 		/* Make sure the IDM flag is off */
1977 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1978 
1979 	if (changed & 0x01)
1980 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1981 }
1982 
1983 DEFINE_INTERRUPT_HANDLER(DebugException)
1984 {
1985 	unsigned long debug_status = regs->dsisr;
1986 
1987 	current->thread.debug.dbsr = debug_status;
1988 
1989 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1990 	 * on server, it stops on the target of the branch. In order to simulate
1991 	 * the server behaviour, we thus restart right away with a single step
1992 	 * instead of stopping here when hitting a BT
1993 	 */
1994 	if (debug_status & DBSR_BT) {
1995 		regs_set_return_msr(regs, regs->msr & ~MSR_DE);
1996 
1997 		/* Disable BT */
1998 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1999 		/* Clear the BT event */
2000 		mtspr(SPRN_DBSR, DBSR_BT);
2001 
2002 		/* Do the single step trick only when coming from userspace */
2003 		if (user_mode(regs)) {
2004 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
2005 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
2006 			regs_set_return_msr(regs, regs->msr | MSR_DE);
2007 			return;
2008 		}
2009 
2010 		if (kprobe_post_handler(regs))
2011 			return;
2012 
2013 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
2014 			       5, SIGTRAP) == NOTIFY_STOP) {
2015 			return;
2016 		}
2017 		if (debugger_sstep(regs))
2018 			return;
2019 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
2020 		regs_set_return_msr(regs, regs->msr & ~MSR_DE);
2021 
2022 		/* Disable instruction completion */
2023 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
2024 		/* Clear the instruction completion event */
2025 		mtspr(SPRN_DBSR, DBSR_IC);
2026 
2027 		if (kprobe_post_handler(regs))
2028 			return;
2029 
2030 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2031 			       5, SIGTRAP) == NOTIFY_STOP) {
2032 			return;
2033 		}
2034 
2035 		if (debugger_sstep(regs))
2036 			return;
2037 
2038 		if (user_mode(regs)) {
2039 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2040 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2041 					       current->thread.debug.dbcr1))
2042 				regs_set_return_msr(regs, regs->msr | MSR_DE);
2043 			else
2044 				/* Make sure the IDM bit is off */
2045 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2046 		}
2047 
2048 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2049 	} else
2050 		handle_debug(regs, debug_status);
2051 }
2052 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2053 
2054 #ifdef CONFIG_ALTIVEC
2055 DEFINE_INTERRUPT_HANDLER(altivec_assist_exception)
2056 {
2057 	int err;
2058 
2059 	if (!user_mode(regs)) {
2060 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2061 		       " at %lx\n", regs->nip);
2062 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2063 	}
2064 
2065 	flush_altivec_to_thread(current);
2066 
2067 	PPC_WARN_EMULATED(altivec, regs);
2068 	err = emulate_altivec(regs);
2069 	if (err == 0) {
2070 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2071 		emulate_single_step(regs);
2072 		return;
2073 	}
2074 
2075 	if (err == -EFAULT) {
2076 		/* got an error reading the instruction */
2077 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2078 	} else {
2079 		/* didn't recognize the instruction */
2080 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2081 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2082 				   "in %s at %lx\n", current->comm, regs->nip);
2083 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2084 	}
2085 }
2086 #endif /* CONFIG_ALTIVEC */
2087 
2088 #ifdef CONFIG_FSL_BOOKE
2089 DEFINE_INTERRUPT_HANDLER(CacheLockingException)
2090 {
2091 	unsigned long error_code = regs->dsisr;
2092 
2093 	/* We treat cache locking instructions from the user
2094 	 * as priv ops, in the future we could try to do
2095 	 * something smarter
2096 	 */
2097 	if (error_code & (ESR_DLK|ESR_ILK))
2098 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2099 	return;
2100 }
2101 #endif /* CONFIG_FSL_BOOKE */
2102 
2103 #ifdef CONFIG_SPE
2104 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointException)
2105 {
2106 	extern int do_spe_mathemu(struct pt_regs *regs);
2107 	unsigned long spefscr;
2108 	int fpexc_mode;
2109 	int code = FPE_FLTUNK;
2110 	int err;
2111 
2112 	interrupt_cond_local_irq_enable(regs);
2113 
2114 	flush_spe_to_thread(current);
2115 
2116 	spefscr = current->thread.spefscr;
2117 	fpexc_mode = current->thread.fpexc_mode;
2118 
2119 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2120 		code = FPE_FLTOVF;
2121 	}
2122 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2123 		code = FPE_FLTUND;
2124 	}
2125 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2126 		code = FPE_FLTDIV;
2127 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2128 		code = FPE_FLTINV;
2129 	}
2130 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2131 		code = FPE_FLTRES;
2132 
2133 	err = do_spe_mathemu(regs);
2134 	if (err == 0) {
2135 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2136 		emulate_single_step(regs);
2137 		return;
2138 	}
2139 
2140 	if (err == -EFAULT) {
2141 		/* got an error reading the instruction */
2142 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2143 	} else if (err == -EINVAL) {
2144 		/* didn't recognize the instruction */
2145 		printk(KERN_ERR "unrecognized spe instruction "
2146 		       "in %s at %lx\n", current->comm, regs->nip);
2147 	} else {
2148 		_exception(SIGFPE, regs, code, regs->nip);
2149 	}
2150 
2151 	return;
2152 }
2153 
2154 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointRoundException)
2155 {
2156 	extern int speround_handler(struct pt_regs *regs);
2157 	int err;
2158 
2159 	interrupt_cond_local_irq_enable(regs);
2160 
2161 	preempt_disable();
2162 	if (regs->msr & MSR_SPE)
2163 		giveup_spe(current);
2164 	preempt_enable();
2165 
2166 	regs_add_return_ip(regs, -4);
2167 	err = speround_handler(regs);
2168 	if (err == 0) {
2169 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2170 		emulate_single_step(regs);
2171 		return;
2172 	}
2173 
2174 	if (err == -EFAULT) {
2175 		/* got an error reading the instruction */
2176 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2177 	} else if (err == -EINVAL) {
2178 		/* didn't recognize the instruction */
2179 		printk(KERN_ERR "unrecognized spe instruction "
2180 		       "in %s at %lx\n", current->comm, regs->nip);
2181 	} else {
2182 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2183 		return;
2184 	}
2185 }
2186 #endif
2187 
2188 /*
2189  * We enter here if we get an unrecoverable exception, that is, one
2190  * that happened at a point where the RI (recoverable interrupt) bit
2191  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2192  * we therefore lost state by taking this exception.
2193  */
2194 void __noreturn unrecoverable_exception(struct pt_regs *regs)
2195 {
2196 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2197 		 regs->trap, regs->nip, regs->msr);
2198 	die("Unrecoverable exception", regs, SIGABRT);
2199 	/* die() should not return */
2200 	for (;;)
2201 		;
2202 }
2203 
2204 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2205 /*
2206  * Default handler for a Watchdog exception,
2207  * spins until a reboot occurs
2208  */
2209 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2210 {
2211 	/* Generic WatchdogHandler, implement your own */
2212 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2213 	return;
2214 }
2215 
2216 DEFINE_INTERRUPT_HANDLER_NMI(WatchdogException)
2217 {
2218 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2219 	WatchdogHandler(regs);
2220 	return 0;
2221 }
2222 #endif
2223 
2224 /*
2225  * We enter here if we discover during exception entry that we are
2226  * running in supervisor mode with a userspace value in the stack pointer.
2227  */
2228 DEFINE_INTERRUPT_HANDLER(kernel_bad_stack)
2229 {
2230 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2231 	       regs->gpr[1], regs->nip);
2232 	die("Bad kernel stack pointer", regs, SIGABRT);
2233 }
2234 
2235 #ifdef CONFIG_PPC_EMULATED_STATS
2236 
2237 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2238 
2239 struct ppc_emulated ppc_emulated = {
2240 #ifdef CONFIG_ALTIVEC
2241 	WARN_EMULATED_SETUP(altivec),
2242 #endif
2243 	WARN_EMULATED_SETUP(dcba),
2244 	WARN_EMULATED_SETUP(dcbz),
2245 	WARN_EMULATED_SETUP(fp_pair),
2246 	WARN_EMULATED_SETUP(isel),
2247 	WARN_EMULATED_SETUP(mcrxr),
2248 	WARN_EMULATED_SETUP(mfpvr),
2249 	WARN_EMULATED_SETUP(multiple),
2250 	WARN_EMULATED_SETUP(popcntb),
2251 	WARN_EMULATED_SETUP(spe),
2252 	WARN_EMULATED_SETUP(string),
2253 	WARN_EMULATED_SETUP(sync),
2254 	WARN_EMULATED_SETUP(unaligned),
2255 #ifdef CONFIG_MATH_EMULATION
2256 	WARN_EMULATED_SETUP(math),
2257 #endif
2258 #ifdef CONFIG_VSX
2259 	WARN_EMULATED_SETUP(vsx),
2260 #endif
2261 #ifdef CONFIG_PPC64
2262 	WARN_EMULATED_SETUP(mfdscr),
2263 	WARN_EMULATED_SETUP(mtdscr),
2264 	WARN_EMULATED_SETUP(lq_stq),
2265 	WARN_EMULATED_SETUP(lxvw4x),
2266 	WARN_EMULATED_SETUP(lxvh8x),
2267 	WARN_EMULATED_SETUP(lxvd2x),
2268 	WARN_EMULATED_SETUP(lxvb16x),
2269 #endif
2270 };
2271 
2272 u32 ppc_warn_emulated;
2273 
2274 void ppc_warn_emulated_print(const char *type)
2275 {
2276 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2277 			    type);
2278 }
2279 
2280 static int __init ppc_warn_emulated_init(void)
2281 {
2282 	struct dentry *dir;
2283 	unsigned int i;
2284 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2285 
2286 	dir = debugfs_create_dir("emulated_instructions",
2287 				 arch_debugfs_dir);
2288 
2289 	debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated);
2290 
2291 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++)
2292 		debugfs_create_u32(entries[i].name, 0644, dir,
2293 				   (u32 *)&entries[i].val.counter);
2294 
2295 	return 0;
2296 }
2297 
2298 device_initcall(ppc_warn_emulated_init);
2299 
2300 #endif /* CONFIG_PPC_EMULATED_STATS */
2301