xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision bcda5fd3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Copyright (C) 1995-1996  Gary Thomas (gdt@linuxppc.org)
4  *  Copyright 2007-2010 Freescale Semiconductor, Inc.
5  *
6  *  Modified by Cort Dougan (cort@cs.nmt.edu)
7  *  and Paul Mackerras (paulus@samba.org)
8  */
9 
10 /*
11  * This file handles the architecture-dependent parts of hardware exceptions
12  */
13 
14 #include <linux/errno.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/pkeys.h>
20 #include <linux/stddef.h>
21 #include <linux/unistd.h>
22 #include <linux/ptrace.h>
23 #include <linux/user.h>
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/extable.h>
27 #include <linux/module.h>	/* print_modules */
28 #include <linux/prctl.h>
29 #include <linux/delay.h>
30 #include <linux/kprobes.h>
31 #include <linux/kexec.h>
32 #include <linux/backlight.h>
33 #include <linux/bug.h>
34 #include <linux/kdebug.h>
35 #include <linux/ratelimit.h>
36 #include <linux/context_tracking.h>
37 #include <linux/smp.h>
38 #include <linux/console.h>
39 #include <linux/kmsg_dump.h>
40 
41 #include <asm/emulated_ops.h>
42 #include <linux/uaccess.h>
43 #include <asm/debugfs.h>
44 #include <asm/interrupt.h>
45 #include <asm/io.h>
46 #include <asm/machdep.h>
47 #include <asm/rtas.h>
48 #include <asm/pmc.h>
49 #include <asm/reg.h>
50 #ifdef CONFIG_PMAC_BACKLIGHT
51 #include <asm/backlight.h>
52 #endif
53 #ifdef CONFIG_PPC64
54 #include <asm/firmware.h>
55 #include <asm/processor.h>
56 #endif
57 #include <asm/kexec.h>
58 #include <asm/ppc-opcode.h>
59 #include <asm/rio.h>
60 #include <asm/fadump.h>
61 #include <asm/switch_to.h>
62 #include <asm/tm.h>
63 #include <asm/debug.h>
64 #include <asm/asm-prototypes.h>
65 #include <asm/hmi.h>
66 #include <sysdev/fsl_pci.h>
67 #include <asm/kprobes.h>
68 #include <asm/stacktrace.h>
69 #include <asm/nmi.h>
70 #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 	printk_safe_flush_on_panic();
175 	kmsg_dump(KMSG_DUMP_PANIC);
176 	bust_spinlocks(0);
177 	debug_locks_off();
178 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
179 }
180 
181 static unsigned long oops_begin(struct pt_regs *regs)
182 {
183 	int cpu;
184 	unsigned long flags;
185 
186 	oops_enter();
187 
188 	/* racy, but better than risking deadlock. */
189 	raw_local_irq_save(flags);
190 	cpu = smp_processor_id();
191 	if (!arch_spin_trylock(&die_lock)) {
192 		if (cpu == die_owner)
193 			/* nested oops. should stop eventually */;
194 		else
195 			arch_spin_lock(&die_lock);
196 	}
197 	die_nest_count++;
198 	die_owner = cpu;
199 	console_verbose();
200 	bust_spinlocks(1);
201 	if (machine_is(powermac))
202 		pmac_backlight_unblank();
203 	return flags;
204 }
205 NOKPROBE_SYMBOL(oops_begin);
206 
207 static void oops_end(unsigned long flags, struct pt_regs *regs,
208 			       int signr)
209 {
210 	bust_spinlocks(0);
211 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
212 	die_nest_count--;
213 	oops_exit();
214 	printk("\n");
215 	if (!die_nest_count) {
216 		/* Nest count reaches zero, release the lock. */
217 		die_owner = -1;
218 		arch_spin_unlock(&die_lock);
219 	}
220 	raw_local_irq_restore(flags);
221 
222 	/*
223 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
224 	 */
225 	if (TRAP(regs) == INTERRUPT_SYSTEM_RESET)
226 		return;
227 
228 	crash_fadump(regs, "die oops");
229 
230 	if (kexec_should_crash(current))
231 		crash_kexec(regs);
232 
233 	if (!signr)
234 		return;
235 
236 	/*
237 	 * While our oops output is serialised by a spinlock, output
238 	 * from panic() called below can race and corrupt it. If we
239 	 * know we are going to panic, delay for 1 second so we have a
240 	 * chance to get clean backtraces from all CPUs that are oopsing.
241 	 */
242 	if (in_interrupt() || panic_on_oops || !current->pid ||
243 	    is_global_init(current)) {
244 		mdelay(MSEC_PER_SEC);
245 	}
246 
247 	if (panic_on_oops)
248 		panic("Fatal exception");
249 	do_exit(signr);
250 }
251 NOKPROBE_SYMBOL(oops_end);
252 
253 static char *get_mmu_str(void)
254 {
255 	if (early_radix_enabled())
256 		return " MMU=Radix";
257 	if (early_mmu_has_feature(MMU_FTR_HPTE_TABLE))
258 		return " MMU=Hash";
259 	return "";
260 }
261 
262 static int __die(const char *str, struct pt_regs *regs, long err)
263 {
264 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
265 
266 	printk("%s PAGE_SIZE=%luK%s%s%s%s%s%s %s\n",
267 	       IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE",
268 	       PAGE_SIZE / 1024, get_mmu_str(),
269 	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
270 	       IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
271 	       IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "",
272 	       debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
273 	       IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "",
274 	       ppc_md.name ? ppc_md.name : "");
275 
276 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
277 		return 1;
278 
279 	print_modules();
280 	show_regs(regs);
281 
282 	return 0;
283 }
284 NOKPROBE_SYMBOL(__die);
285 
286 void die(const char *str, struct pt_regs *regs, long err)
287 {
288 	unsigned long flags;
289 
290 	/*
291 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
292 	 */
293 	if (TRAP(regs) != INTERRUPT_SYSTEM_RESET) {
294 		if (debugger(regs))
295 			return;
296 	}
297 
298 	flags = oops_begin(regs);
299 	if (__die(str, regs, err))
300 		err = 0;
301 	oops_end(flags, regs, err);
302 }
303 NOKPROBE_SYMBOL(die);
304 
305 void user_single_step_report(struct pt_regs *regs)
306 {
307 	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip);
308 }
309 
310 static void show_signal_msg(int signr, struct pt_regs *regs, int code,
311 			    unsigned long addr)
312 {
313 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
314 				      DEFAULT_RATELIMIT_BURST);
315 
316 	if (!show_unhandled_signals)
317 		return;
318 
319 	if (!unhandled_signal(current, signr))
320 		return;
321 
322 	if (!__ratelimit(&rs))
323 		return;
324 
325 	pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x",
326 		current->comm, current->pid, signame(signr), signr,
327 		addr, regs->nip, regs->link, code);
328 
329 	print_vma_addr(KERN_CONT " in ", regs->nip);
330 
331 	pr_cont("\n");
332 
333 	show_user_instructions(regs);
334 }
335 
336 static bool exception_common(int signr, struct pt_regs *regs, int code,
337 			      unsigned long addr)
338 {
339 	if (!user_mode(regs)) {
340 		die("Exception in kernel mode", regs, signr);
341 		return false;
342 	}
343 
344 	show_signal_msg(signr, regs, code, addr);
345 
346 	if (arch_irqs_disabled())
347 		interrupt_cond_local_irq_enable(regs);
348 
349 	current->thread.trap_nr = code;
350 
351 	return true;
352 }
353 
354 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key)
355 {
356 	if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr))
357 		return;
358 
359 	force_sig_pkuerr((void __user *) addr, key);
360 }
361 
362 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
363 {
364 	if (!exception_common(signr, regs, code, addr))
365 		return;
366 
367 	force_sig_fault(signr, code, (void __user *)addr);
368 }
369 
370 /*
371  * The interrupt architecture has a quirk in that the HV interrupts excluding
372  * the NMIs (0x100 and 0x200) do not clear MSR[RI] at entry. The first thing
373  * that an interrupt handler must do is save off a GPR into a scratch register,
374  * and all interrupts on POWERNV (HV=1) use the HSPRG1 register as scratch.
375  * Therefore an NMI can clobber an HV interrupt's live HSPRG1 without noticing
376  * that it is non-reentrant, which leads to random data corruption.
377  *
378  * The solution is for NMI interrupts in HV mode to check if they originated
379  * from these critical HV interrupt regions. If so, then mark them not
380  * recoverable.
381  *
382  * An alternative would be for HV NMIs to use SPRG for scratch to avoid the
383  * HSPRG1 clobber, however this would cause guest SPRG to be clobbered. Linux
384  * guests should always have MSR[RI]=0 when its scratch SPRG is in use, so
385  * that would work. However any other guest OS that may have the SPRG live
386  * and MSR[RI]=1 could encounter silent corruption.
387  *
388  * Builds that do not support KVM could take this second option to increase
389  * the recoverability of NMIs.
390  */
391 void hv_nmi_check_nonrecoverable(struct pt_regs *regs)
392 {
393 #ifdef CONFIG_PPC_POWERNV
394 	unsigned long kbase = (unsigned long)_stext;
395 	unsigned long nip = regs->nip;
396 
397 	if (!(regs->msr & MSR_RI))
398 		return;
399 	if (!(regs->msr & MSR_HV))
400 		return;
401 	if (regs->msr & MSR_PR)
402 		return;
403 
404 	/*
405 	 * Now test if the interrupt has hit a range that may be using
406 	 * HSPRG1 without having RI=0 (i.e., an HSRR interrupt). The
407 	 * problem ranges all run un-relocated. Test real and virt modes
408 	 * at the same time by dropping the high bit of the nip (virt mode
409 	 * entry points still have the +0x4000 offset).
410 	 */
411 	nip &= ~0xc000000000000000ULL;
412 	if ((nip >= 0x500 && nip < 0x600) || (nip >= 0x4500 && nip < 0x4600))
413 		goto nonrecoverable;
414 	if ((nip >= 0x980 && nip < 0xa00) || (nip >= 0x4980 && nip < 0x4a00))
415 		goto nonrecoverable;
416 	if ((nip >= 0xe00 && nip < 0xec0) || (nip >= 0x4e00 && nip < 0x4ec0))
417 		goto nonrecoverable;
418 	if ((nip >= 0xf80 && nip < 0xfa0) || (nip >= 0x4f80 && nip < 0x4fa0))
419 		goto nonrecoverable;
420 
421 	/* Trampoline code runs un-relocated so subtract kbase. */
422 	if (nip >= (unsigned long)(start_real_trampolines - kbase) &&
423 			nip < (unsigned long)(end_real_trampolines - kbase))
424 		goto nonrecoverable;
425 	if (nip >= (unsigned long)(start_virt_trampolines - kbase) &&
426 			nip < (unsigned long)(end_virt_trampolines - kbase))
427 		goto nonrecoverable;
428 	return;
429 
430 nonrecoverable:
431 	regs_set_return_msr(regs, regs->msr & ~MSR_RI);
432 #endif
433 }
434 DEFINE_INTERRUPT_HANDLER_NMI(system_reset_exception)
435 {
436 	unsigned long hsrr0, hsrr1;
437 	bool saved_hsrrs = false;
438 
439 	/*
440 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
441 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
442 	 * OPAL), so save them here and restore them before returning.
443 	 *
444 	 * Machine checks don't need to save HSRRs, as the real mode handler
445 	 * is careful to avoid them, and the regular handler is not delivered
446 	 * as an NMI.
447 	 */
448 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
449 		hsrr0 = mfspr(SPRN_HSRR0);
450 		hsrr1 = mfspr(SPRN_HSRR1);
451 		saved_hsrrs = true;
452 	}
453 
454 	hv_nmi_check_nonrecoverable(regs);
455 
456 	__this_cpu_inc(irq_stat.sreset_irqs);
457 
458 	/* See if any machine dependent calls */
459 	if (ppc_md.system_reset_exception) {
460 		if (ppc_md.system_reset_exception(regs))
461 			goto out;
462 	}
463 
464 	if (debugger(regs))
465 		goto out;
466 
467 	kmsg_dump(KMSG_DUMP_OOPS);
468 	/*
469 	 * A system reset is a request to dump, so we always send
470 	 * it through the crashdump code (if fadump or kdump are
471 	 * registered).
472 	 */
473 	crash_fadump(regs, "System Reset");
474 
475 	crash_kexec(regs);
476 
477 	/*
478 	 * We aren't the primary crash CPU. We need to send it
479 	 * to a holding pattern to avoid it ending up in the panic
480 	 * code.
481 	 */
482 	crash_kexec_secondary(regs);
483 
484 	/*
485 	 * No debugger or crash dump registered, print logs then
486 	 * panic.
487 	 */
488 	die("System Reset", regs, SIGABRT);
489 
490 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
491 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
492 	nmi_panic(regs, "System Reset");
493 
494 out:
495 #ifdef CONFIG_PPC_BOOK3S_64
496 	BUG_ON(get_paca()->in_nmi == 0);
497 	if (get_paca()->in_nmi > 1)
498 		die("Unrecoverable nested System Reset", regs, SIGABRT);
499 #endif
500 	/* Must die if the interrupt is not recoverable */
501 	if (!(regs->msr & MSR_RI)) {
502 		/* For the reason explained in die_mce, nmi_exit before die */
503 		nmi_exit();
504 		die("Unrecoverable System Reset", regs, SIGABRT);
505 	}
506 
507 	if (saved_hsrrs) {
508 		mtspr(SPRN_HSRR0, hsrr0);
509 		mtspr(SPRN_HSRR1, hsrr1);
510 	}
511 
512 	/* What should we do here? We could issue a shutdown or hard reset. */
513 
514 	return 0;
515 }
516 
517 /*
518  * I/O accesses can cause machine checks on powermacs.
519  * Check if the NIP corresponds to the address of a sync
520  * instruction for which there is an entry in the exception
521  * table.
522  *  -- paulus.
523  */
524 static inline int check_io_access(struct pt_regs *regs)
525 {
526 #ifdef CONFIG_PPC32
527 	unsigned long msr = regs->msr;
528 	const struct exception_table_entry *entry;
529 	unsigned int *nip = (unsigned int *)regs->nip;
530 
531 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
532 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
533 		/*
534 		 * Check that it's a sync instruction, or somewhere
535 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
536 		 * As the address is in the exception table
537 		 * we should be able to read the instr there.
538 		 * For the debug message, we look at the preceding
539 		 * load or store.
540 		 */
541 		if (*nip == PPC_RAW_NOP())
542 			nip -= 2;
543 		else if (*nip == PPC_RAW_ISYNC())
544 			--nip;
545 		if (*nip == PPC_RAW_SYNC() || get_op(*nip) == OP_TRAP) {
546 			unsigned int rb;
547 
548 			--nip;
549 			rb = (*nip >> 11) & 0x1f;
550 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
551 			       (*nip & 0x100)? "OUT to": "IN from",
552 			       regs->gpr[rb] - _IO_BASE, nip);
553 			regs_set_return_msr(regs, regs->msr | MSR_RI);
554 			regs_set_return_ip(regs, extable_fixup(entry));
555 			return 1;
556 		}
557 	}
558 #endif /* CONFIG_PPC32 */
559 	return 0;
560 }
561 
562 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
563 /* On 4xx, the reason for the machine check or program exception
564    is in the ESR. */
565 #define get_reason(regs)	((regs)->dsisr)
566 #define REASON_FP		ESR_FP
567 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
568 #define REASON_PRIVILEGED	ESR_PPR
569 #define REASON_TRAP		ESR_PTR
570 #define REASON_PREFIXED		0
571 #define REASON_BOUNDARY		0
572 
573 /* single-step stuff */
574 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
575 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
576 #define clear_br_trace(regs)	do {} while(0)
577 #else
578 /* On non-4xx, the reason for the machine check or program
579    exception is in the MSR. */
580 #define get_reason(regs)	((regs)->msr)
581 #define REASON_TM		SRR1_PROGTM
582 #define REASON_FP		SRR1_PROGFPE
583 #define REASON_ILLEGAL		SRR1_PROGILL
584 #define REASON_PRIVILEGED	SRR1_PROGPRIV
585 #define REASON_TRAP		SRR1_PROGTRAP
586 #define REASON_PREFIXED		SRR1_PREFIXED
587 #define REASON_BOUNDARY		SRR1_BOUNDARY
588 
589 #define single_stepping(regs)	((regs)->msr & MSR_SE)
590 #define clear_single_step(regs)	(regs_set_return_msr((regs), (regs)->msr & ~MSR_SE))
591 #define clear_br_trace(regs)	(regs_set_return_msr((regs), (regs)->msr & ~MSR_BE))
592 #endif
593 
594 #define inst_length(reason)	(((reason) & REASON_PREFIXED) ? 8 : 4)
595 
596 #if defined(CONFIG_E500)
597 int machine_check_e500mc(struct pt_regs *regs)
598 {
599 	unsigned long mcsr = mfspr(SPRN_MCSR);
600 	unsigned long pvr = mfspr(SPRN_PVR);
601 	unsigned long reason = mcsr;
602 	int recoverable = 1;
603 
604 	if (reason & MCSR_LD) {
605 		recoverable = fsl_rio_mcheck_exception(regs);
606 		if (recoverable == 1)
607 			goto silent_out;
608 	}
609 
610 	printk("Machine check in kernel mode.\n");
611 	printk("Caused by (from MCSR=%lx): ", reason);
612 
613 	if (reason & MCSR_MCP)
614 		pr_cont("Machine Check Signal\n");
615 
616 	if (reason & MCSR_ICPERR) {
617 		pr_cont("Instruction Cache Parity Error\n");
618 
619 		/*
620 		 * This is recoverable by invalidating the i-cache.
621 		 */
622 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
623 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
624 			;
625 
626 		/*
627 		 * This will generally be accompanied by an instruction
628 		 * fetch error report -- only treat MCSR_IF as fatal
629 		 * if it wasn't due to an L1 parity error.
630 		 */
631 		reason &= ~MCSR_IF;
632 	}
633 
634 	if (reason & MCSR_DCPERR_MC) {
635 		pr_cont("Data Cache Parity Error\n");
636 
637 		/*
638 		 * In write shadow mode we auto-recover from the error, but it
639 		 * may still get logged and cause a machine check.  We should
640 		 * only treat the non-write shadow case as non-recoverable.
641 		 */
642 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
643 		 * is not implemented but L1 data cache always runs in write
644 		 * shadow mode. Hence on data cache parity errors HW will
645 		 * automatically invalidate the L1 Data Cache.
646 		 */
647 		if (PVR_VER(pvr) != PVR_VER_E6500) {
648 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
649 				recoverable = 0;
650 		}
651 	}
652 
653 	if (reason & MCSR_L2MMU_MHIT) {
654 		pr_cont("Hit on multiple TLB entries\n");
655 		recoverable = 0;
656 	}
657 
658 	if (reason & MCSR_NMI)
659 		pr_cont("Non-maskable interrupt\n");
660 
661 	if (reason & MCSR_IF) {
662 		pr_cont("Instruction Fetch Error Report\n");
663 		recoverable = 0;
664 	}
665 
666 	if (reason & MCSR_LD) {
667 		pr_cont("Load Error Report\n");
668 		recoverable = 0;
669 	}
670 
671 	if (reason & MCSR_ST) {
672 		pr_cont("Store Error Report\n");
673 		recoverable = 0;
674 	}
675 
676 	if (reason & MCSR_LDG) {
677 		pr_cont("Guarded Load Error Report\n");
678 		recoverable = 0;
679 	}
680 
681 	if (reason & MCSR_TLBSYNC)
682 		pr_cont("Simultaneous tlbsync operations\n");
683 
684 	if (reason & MCSR_BSL2_ERR) {
685 		pr_cont("Level 2 Cache Error\n");
686 		recoverable = 0;
687 	}
688 
689 	if (reason & MCSR_MAV) {
690 		u64 addr;
691 
692 		addr = mfspr(SPRN_MCAR);
693 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
694 
695 		pr_cont("Machine Check %s Address: %#llx\n",
696 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
697 	}
698 
699 silent_out:
700 	mtspr(SPRN_MCSR, mcsr);
701 	return mfspr(SPRN_MCSR) == 0 && recoverable;
702 }
703 
704 int machine_check_e500(struct pt_regs *regs)
705 {
706 	unsigned long reason = mfspr(SPRN_MCSR);
707 
708 	if (reason & MCSR_BUS_RBERR) {
709 		if (fsl_rio_mcheck_exception(regs))
710 			return 1;
711 		if (fsl_pci_mcheck_exception(regs))
712 			return 1;
713 	}
714 
715 	printk("Machine check in kernel mode.\n");
716 	printk("Caused by (from MCSR=%lx): ", reason);
717 
718 	if (reason & MCSR_MCP)
719 		pr_cont("Machine Check Signal\n");
720 	if (reason & MCSR_ICPERR)
721 		pr_cont("Instruction Cache Parity Error\n");
722 	if (reason & MCSR_DCP_PERR)
723 		pr_cont("Data Cache Push Parity Error\n");
724 	if (reason & MCSR_DCPERR)
725 		pr_cont("Data Cache Parity Error\n");
726 	if (reason & MCSR_BUS_IAERR)
727 		pr_cont("Bus - Instruction Address Error\n");
728 	if (reason & MCSR_BUS_RAERR)
729 		pr_cont("Bus - Read Address Error\n");
730 	if (reason & MCSR_BUS_WAERR)
731 		pr_cont("Bus - Write Address Error\n");
732 	if (reason & MCSR_BUS_IBERR)
733 		pr_cont("Bus - Instruction Data Error\n");
734 	if (reason & MCSR_BUS_RBERR)
735 		pr_cont("Bus - Read Data Bus Error\n");
736 	if (reason & MCSR_BUS_WBERR)
737 		pr_cont("Bus - Write Data Bus Error\n");
738 	if (reason & MCSR_BUS_IPERR)
739 		pr_cont("Bus - Instruction Parity Error\n");
740 	if (reason & MCSR_BUS_RPERR)
741 		pr_cont("Bus - Read Parity Error\n");
742 
743 	return 0;
744 }
745 
746 int machine_check_generic(struct pt_regs *regs)
747 {
748 	return 0;
749 }
750 #elif defined(CONFIG_PPC32)
751 int machine_check_generic(struct pt_regs *regs)
752 {
753 	unsigned long reason = regs->msr;
754 
755 	printk("Machine check in kernel mode.\n");
756 	printk("Caused by (from SRR1=%lx): ", reason);
757 	switch (reason & 0x601F0000) {
758 	case 0x80000:
759 		pr_cont("Machine check signal\n");
760 		break;
761 	case 0x40000:
762 	case 0x140000:	/* 7450 MSS error and TEA */
763 		pr_cont("Transfer error ack signal\n");
764 		break;
765 	case 0x20000:
766 		pr_cont("Data parity error signal\n");
767 		break;
768 	case 0x10000:
769 		pr_cont("Address parity error signal\n");
770 		break;
771 	case 0x20000000:
772 		pr_cont("L1 Data Cache error\n");
773 		break;
774 	case 0x40000000:
775 		pr_cont("L1 Instruction Cache error\n");
776 		break;
777 	case 0x00100000:
778 		pr_cont("L2 data cache parity error\n");
779 		break;
780 	default:
781 		pr_cont("Unknown values in msr\n");
782 	}
783 	return 0;
784 }
785 #endif /* everything else */
786 
787 void die_mce(const char *str, struct pt_regs *regs, long err)
788 {
789 	/*
790 	 * The machine check wants to kill the interrupted context, but
791 	 * do_exit() checks for in_interrupt() and panics in that case, so
792 	 * exit the irq/nmi before calling die.
793 	 */
794 	if (IS_ENABLED(CONFIG_PPC_BOOK3S_64))
795 		irq_exit();
796 	else
797 		nmi_exit();
798 	die(str, regs, err);
799 }
800 
801 /*
802  * BOOK3S_64 does not call this handler as a non-maskable interrupt
803  * (it uses its own early real-mode handler to handle the MCE proper
804  * and then raises irq_work to call this handler when interrupts are
805  * enabled).
806  */
807 #ifdef CONFIG_PPC_BOOK3S_64
808 DEFINE_INTERRUPT_HANDLER_ASYNC(machine_check_exception)
809 #else
810 DEFINE_INTERRUPT_HANDLER_NMI(machine_check_exception)
811 #endif
812 {
813 	int recover = 0;
814 
815 	__this_cpu_inc(irq_stat.mce_exceptions);
816 
817 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
818 
819 	/* See if any machine dependent calls. In theory, we would want
820 	 * to call the CPU first, and call the ppc_md. one if the CPU
821 	 * one returns a positive number. However there is existing code
822 	 * that assumes the board gets a first chance, so let's keep it
823 	 * that way for now and fix things later. --BenH.
824 	 */
825 	if (ppc_md.machine_check_exception)
826 		recover = ppc_md.machine_check_exception(regs);
827 	else if (cur_cpu_spec->machine_check)
828 		recover = cur_cpu_spec->machine_check(regs);
829 
830 	if (recover > 0)
831 		goto bail;
832 
833 	if (debugger_fault_handler(regs))
834 		goto bail;
835 
836 	if (check_io_access(regs))
837 		goto bail;
838 
839 	die_mce("Machine check", regs, SIGBUS);
840 
841 bail:
842 	/* Must die if the interrupt is not recoverable */
843 	if (!(regs->msr & MSR_RI))
844 		die_mce("Unrecoverable Machine check", regs, SIGBUS);
845 
846 #ifdef CONFIG_PPC_BOOK3S_64
847 	return;
848 #else
849 	return 0;
850 #endif
851 }
852 
853 DEFINE_INTERRUPT_HANDLER(SMIException) /* async? */
854 {
855 	die("System Management Interrupt", regs, SIGABRT);
856 }
857 
858 #ifdef CONFIG_VSX
859 static void p9_hmi_special_emu(struct pt_regs *regs)
860 {
861 	unsigned int ra, rb, t, i, sel, instr, rc;
862 	const void __user *addr;
863 	u8 vbuf[16] __aligned(16), *vdst;
864 	unsigned long ea, msr, msr_mask;
865 	bool swap;
866 
867 	if (__get_user(instr, (unsigned int __user *)regs->nip))
868 		return;
869 
870 	/*
871 	 * lxvb16x	opcode: 0x7c0006d8
872 	 * lxvd2x	opcode: 0x7c000698
873 	 * lxvh8x	opcode: 0x7c000658
874 	 * lxvw4x	opcode: 0x7c000618
875 	 */
876 	if ((instr & 0xfc00073e) != 0x7c000618) {
877 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
878 			 " instr=%08x\n",
879 			 smp_processor_id(), current->comm, current->pid,
880 			 regs->nip, instr);
881 		return;
882 	}
883 
884 	/* Grab vector registers into the task struct */
885 	msr = regs->msr; /* Grab msr before we flush the bits */
886 	flush_vsx_to_thread(current);
887 	enable_kernel_altivec();
888 
889 	/*
890 	 * Is userspace running with a different endian (this is rare but
891 	 * not impossible)
892 	 */
893 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
894 
895 	/* Decode the instruction */
896 	ra = (instr >> 16) & 0x1f;
897 	rb = (instr >> 11) & 0x1f;
898 	t = (instr >> 21) & 0x1f;
899 	if (instr & 1)
900 		vdst = (u8 *)&current->thread.vr_state.vr[t];
901 	else
902 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
903 
904 	/* Grab the vector address */
905 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
906 	if (is_32bit_task())
907 		ea &= 0xfffffffful;
908 	addr = (__force const void __user *)ea;
909 
910 	/* Check it */
911 	if (!access_ok(addr, 16)) {
912 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
913 			 " instr=%08x addr=%016lx\n",
914 			 smp_processor_id(), current->comm, current->pid,
915 			 regs->nip, instr, (unsigned long)addr);
916 		return;
917 	}
918 
919 	/* Read the vector */
920 	rc = 0;
921 	if ((unsigned long)addr & 0xfUL)
922 		/* unaligned case */
923 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
924 	else
925 		__get_user_atomic_128_aligned(vbuf, addr, rc);
926 	if (rc) {
927 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
928 			 " instr=%08x addr=%016lx\n",
929 			 smp_processor_id(), current->comm, current->pid,
930 			 regs->nip, instr, (unsigned long)addr);
931 		return;
932 	}
933 
934 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
935 		 " instr=%08x addr=%016lx\n",
936 		 smp_processor_id(), current->comm, current->pid, regs->nip,
937 		 instr, (unsigned long) addr);
938 
939 	/* Grab instruction "selector" */
940 	sel = (instr >> 6) & 3;
941 
942 	/*
943 	 * Check to make sure the facility is actually enabled. This
944 	 * could happen if we get a false positive hit.
945 	 *
946 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
947 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
948 	 */
949 	msr_mask = MSR_VSX;
950 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
951 		msr_mask = MSR_VEC;
952 	if (!(msr & msr_mask)) {
953 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
954 			 " instr=%08x msr:%016lx\n",
955 			 smp_processor_id(), current->comm, current->pid,
956 			 regs->nip, instr, msr);
957 		return;
958 	}
959 
960 	/* Do logging here before we modify sel based on endian */
961 	switch (sel) {
962 	case 0:	/* lxvw4x */
963 		PPC_WARN_EMULATED(lxvw4x, regs);
964 		break;
965 	case 1: /* lxvh8x */
966 		PPC_WARN_EMULATED(lxvh8x, regs);
967 		break;
968 	case 2: /* lxvd2x */
969 		PPC_WARN_EMULATED(lxvd2x, regs);
970 		break;
971 	case 3: /* lxvb16x */
972 		PPC_WARN_EMULATED(lxvb16x, regs);
973 		break;
974 	}
975 
976 #ifdef __LITTLE_ENDIAN__
977 	/*
978 	 * An LE kernel stores the vector in the task struct as an LE
979 	 * byte array (effectively swapping both the components and
980 	 * the content of the components). Those instructions expect
981 	 * the components to remain in ascending address order, so we
982 	 * swap them back.
983 	 *
984 	 * If we are running a BE user space, the expectation is that
985 	 * of a simple memcpy, so forcing the emulation to look like
986 	 * a lxvb16x should do the trick.
987 	 */
988 	if (swap)
989 		sel = 3;
990 
991 	switch (sel) {
992 	case 0:	/* lxvw4x */
993 		for (i = 0; i < 4; i++)
994 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
995 		break;
996 	case 1: /* lxvh8x */
997 		for (i = 0; i < 8; i++)
998 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
999 		break;
1000 	case 2: /* lxvd2x */
1001 		for (i = 0; i < 2; i++)
1002 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1003 		break;
1004 	case 3: /* lxvb16x */
1005 		for (i = 0; i < 16; i++)
1006 			vdst[i] = vbuf[15-i];
1007 		break;
1008 	}
1009 #else /* __LITTLE_ENDIAN__ */
1010 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1011 	if (!swap)
1012 		sel = 3;
1013 
1014 	/* Otherwise, we need to swap the content of the components */
1015 	switch (sel) {
1016 	case 0:	/* lxvw4x */
1017 		for (i = 0; i < 4; i++)
1018 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1019 		break;
1020 	case 1: /* lxvh8x */
1021 		for (i = 0; i < 8; i++)
1022 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1023 		break;
1024 	case 2: /* lxvd2x */
1025 		for (i = 0; i < 2; i++)
1026 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1027 		break;
1028 	case 3: /* lxvb16x */
1029 		memcpy(vdst, vbuf, 16);
1030 		break;
1031 	}
1032 #endif /* !__LITTLE_ENDIAN__ */
1033 
1034 	/* Go to next instruction */
1035 	regs_add_return_ip(regs, 4);
1036 }
1037 #endif /* CONFIG_VSX */
1038 
1039 DEFINE_INTERRUPT_HANDLER_ASYNC(handle_hmi_exception)
1040 {
1041 	struct pt_regs *old_regs;
1042 
1043 	old_regs = set_irq_regs(regs);
1044 
1045 #ifdef CONFIG_VSX
1046 	/* Real mode flagged P9 special emu is needed */
1047 	if (local_paca->hmi_p9_special_emu) {
1048 		local_paca->hmi_p9_special_emu = 0;
1049 
1050 		/*
1051 		 * We don't want to take page faults while doing the
1052 		 * emulation, we just replay the instruction if necessary.
1053 		 */
1054 		pagefault_disable();
1055 		p9_hmi_special_emu(regs);
1056 		pagefault_enable();
1057 	}
1058 #endif /* CONFIG_VSX */
1059 
1060 	if (ppc_md.handle_hmi_exception)
1061 		ppc_md.handle_hmi_exception(regs);
1062 
1063 	set_irq_regs(old_regs);
1064 }
1065 
1066 DEFINE_INTERRUPT_HANDLER(unknown_exception)
1067 {
1068 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1069 	       regs->nip, regs->msr, regs->trap);
1070 
1071 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1072 }
1073 
1074 DEFINE_INTERRUPT_HANDLER_ASYNC(unknown_async_exception)
1075 {
1076 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1077 	       regs->nip, regs->msr, regs->trap);
1078 
1079 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1080 }
1081 
1082 DEFINE_INTERRUPT_HANDLER_NMI(unknown_nmi_exception)
1083 {
1084 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1085 	       regs->nip, regs->msr, regs->trap);
1086 
1087 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1088 
1089 	return 0;
1090 }
1091 
1092 DEFINE_INTERRUPT_HANDLER(instruction_breakpoint_exception)
1093 {
1094 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1095 					5, SIGTRAP) == NOTIFY_STOP)
1096 		return;
1097 	if (debugger_iabr_match(regs))
1098 		return;
1099 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1100 }
1101 
1102 DEFINE_INTERRUPT_HANDLER(RunModeException)
1103 {
1104 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1105 }
1106 
1107 static void __single_step_exception(struct pt_regs *regs)
1108 {
1109 	clear_single_step(regs);
1110 	clear_br_trace(regs);
1111 
1112 	if (kprobe_post_handler(regs))
1113 		return;
1114 
1115 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1116 					5, SIGTRAP) == NOTIFY_STOP)
1117 		return;
1118 	if (debugger_sstep(regs))
1119 		return;
1120 
1121 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1122 }
1123 
1124 DEFINE_INTERRUPT_HANDLER(single_step_exception)
1125 {
1126 	__single_step_exception(regs);
1127 }
1128 
1129 /*
1130  * After we have successfully emulated an instruction, we have to
1131  * check if the instruction was being single-stepped, and if so,
1132  * pretend we got a single-step exception.  This was pointed out
1133  * by Kumar Gala.  -- paulus
1134  */
1135 static void emulate_single_step(struct pt_regs *regs)
1136 {
1137 	if (single_stepping(regs))
1138 		__single_step_exception(regs);
1139 }
1140 
1141 static inline int __parse_fpscr(unsigned long fpscr)
1142 {
1143 	int ret = FPE_FLTUNK;
1144 
1145 	/* Invalid operation */
1146 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1147 		ret = FPE_FLTINV;
1148 
1149 	/* Overflow */
1150 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1151 		ret = FPE_FLTOVF;
1152 
1153 	/* Underflow */
1154 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1155 		ret = FPE_FLTUND;
1156 
1157 	/* Divide by zero */
1158 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1159 		ret = FPE_FLTDIV;
1160 
1161 	/* Inexact result */
1162 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1163 		ret = FPE_FLTRES;
1164 
1165 	return ret;
1166 }
1167 
1168 static void parse_fpe(struct pt_regs *regs)
1169 {
1170 	int code = 0;
1171 
1172 	flush_fp_to_thread(current);
1173 
1174 #ifdef CONFIG_PPC_FPU_REGS
1175 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1176 #endif
1177 
1178 	_exception(SIGFPE, regs, code, regs->nip);
1179 }
1180 
1181 /*
1182  * Illegal instruction emulation support.  Originally written to
1183  * provide the PVR to user applications using the mfspr rd, PVR.
1184  * Return non-zero if we can't emulate, or -EFAULT if the associated
1185  * memory access caused an access fault.  Return zero on success.
1186  *
1187  * There are a couple of ways to do this, either "decode" the instruction
1188  * or directly match lots of bits.  In this case, matching lots of
1189  * bits is faster and easier.
1190  *
1191  */
1192 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1193 {
1194 	u8 rT = (instword >> 21) & 0x1f;
1195 	u8 rA = (instword >> 16) & 0x1f;
1196 	u8 NB_RB = (instword >> 11) & 0x1f;
1197 	u32 num_bytes;
1198 	unsigned long EA;
1199 	int pos = 0;
1200 
1201 	/* Early out if we are an invalid form of lswx */
1202 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1203 		if ((rT == rA) || (rT == NB_RB))
1204 			return -EINVAL;
1205 
1206 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1207 
1208 	switch (instword & PPC_INST_STRING_MASK) {
1209 		case PPC_INST_LSWX:
1210 		case PPC_INST_STSWX:
1211 			EA += NB_RB;
1212 			num_bytes = regs->xer & 0x7f;
1213 			break;
1214 		case PPC_INST_LSWI:
1215 		case PPC_INST_STSWI:
1216 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1217 			break;
1218 		default:
1219 			return -EINVAL;
1220 	}
1221 
1222 	while (num_bytes != 0)
1223 	{
1224 		u8 val;
1225 		u32 shift = 8 * (3 - (pos & 0x3));
1226 
1227 		/* if process is 32-bit, clear upper 32 bits of EA */
1228 		if ((regs->msr & MSR_64BIT) == 0)
1229 			EA &= 0xFFFFFFFF;
1230 
1231 		switch ((instword & PPC_INST_STRING_MASK)) {
1232 			case PPC_INST_LSWX:
1233 			case PPC_INST_LSWI:
1234 				if (get_user(val, (u8 __user *)EA))
1235 					return -EFAULT;
1236 				/* first time updating this reg,
1237 				 * zero it out */
1238 				if (pos == 0)
1239 					regs->gpr[rT] = 0;
1240 				regs->gpr[rT] |= val << shift;
1241 				break;
1242 			case PPC_INST_STSWI:
1243 			case PPC_INST_STSWX:
1244 				val = regs->gpr[rT] >> shift;
1245 				if (put_user(val, (u8 __user *)EA))
1246 					return -EFAULT;
1247 				break;
1248 		}
1249 		/* move EA to next address */
1250 		EA += 1;
1251 		num_bytes--;
1252 
1253 		/* manage our position within the register */
1254 		if (++pos == 4) {
1255 			pos = 0;
1256 			if (++rT == 32)
1257 				rT = 0;
1258 		}
1259 	}
1260 
1261 	return 0;
1262 }
1263 
1264 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1265 {
1266 	u32 ra,rs;
1267 	unsigned long tmp;
1268 
1269 	ra = (instword >> 16) & 0x1f;
1270 	rs = (instword >> 21) & 0x1f;
1271 
1272 	tmp = regs->gpr[rs];
1273 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1274 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1275 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1276 	regs->gpr[ra] = tmp;
1277 
1278 	return 0;
1279 }
1280 
1281 static int emulate_isel(struct pt_regs *regs, u32 instword)
1282 {
1283 	u8 rT = (instword >> 21) & 0x1f;
1284 	u8 rA = (instword >> 16) & 0x1f;
1285 	u8 rB = (instword >> 11) & 0x1f;
1286 	u8 BC = (instword >> 6) & 0x1f;
1287 	u8 bit;
1288 	unsigned long tmp;
1289 
1290 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1291 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1292 
1293 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1294 
1295 	return 0;
1296 }
1297 
1298 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1299 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1300 {
1301         /* If we're emulating a load/store in an active transaction, we cannot
1302          * emulate it as the kernel operates in transaction suspended context.
1303          * We need to abort the transaction.  This creates a persistent TM
1304          * abort so tell the user what caused it with a new code.
1305 	 */
1306 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1307 		tm_enable();
1308 		tm_abort(cause);
1309 		return true;
1310 	}
1311 	return false;
1312 }
1313 #else
1314 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1315 {
1316 	return false;
1317 }
1318 #endif
1319 
1320 static int emulate_instruction(struct pt_regs *regs)
1321 {
1322 	u32 instword;
1323 	u32 rd;
1324 
1325 	if (!user_mode(regs))
1326 		return -EINVAL;
1327 
1328 	if (get_user(instword, (u32 __user *)(regs->nip)))
1329 		return -EFAULT;
1330 
1331 	/* Emulate the mfspr rD, PVR. */
1332 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1333 		PPC_WARN_EMULATED(mfpvr, regs);
1334 		rd = (instword >> 21) & 0x1f;
1335 		regs->gpr[rd] = mfspr(SPRN_PVR);
1336 		return 0;
1337 	}
1338 
1339 	/* Emulating the dcba insn is just a no-op.  */
1340 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1341 		PPC_WARN_EMULATED(dcba, regs);
1342 		return 0;
1343 	}
1344 
1345 	/* Emulate the mcrxr insn.  */
1346 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1347 		int shift = (instword >> 21) & 0x1c;
1348 		unsigned long msk = 0xf0000000UL >> shift;
1349 
1350 		PPC_WARN_EMULATED(mcrxr, regs);
1351 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1352 		regs->xer &= ~0xf0000000UL;
1353 		return 0;
1354 	}
1355 
1356 	/* Emulate load/store string insn. */
1357 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1358 		if (tm_abort_check(regs,
1359 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1360 			return -EINVAL;
1361 		PPC_WARN_EMULATED(string, regs);
1362 		return emulate_string_inst(regs, instword);
1363 	}
1364 
1365 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1366 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1367 		PPC_WARN_EMULATED(popcntb, regs);
1368 		return emulate_popcntb_inst(regs, instword);
1369 	}
1370 
1371 	/* Emulate isel (Integer Select) instruction */
1372 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1373 		PPC_WARN_EMULATED(isel, regs);
1374 		return emulate_isel(regs, instword);
1375 	}
1376 
1377 	/* Emulate sync instruction variants */
1378 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1379 		PPC_WARN_EMULATED(sync, regs);
1380 		asm volatile("sync");
1381 		return 0;
1382 	}
1383 
1384 #ifdef CONFIG_PPC64
1385 	/* Emulate the mfspr rD, DSCR. */
1386 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1387 		PPC_INST_MFSPR_DSCR_USER) ||
1388 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1389 		PPC_INST_MFSPR_DSCR)) &&
1390 			cpu_has_feature(CPU_FTR_DSCR)) {
1391 		PPC_WARN_EMULATED(mfdscr, regs);
1392 		rd = (instword >> 21) & 0x1f;
1393 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1394 		return 0;
1395 	}
1396 	/* Emulate the mtspr DSCR, rD. */
1397 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1398 		PPC_INST_MTSPR_DSCR_USER) ||
1399 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1400 		PPC_INST_MTSPR_DSCR)) &&
1401 			cpu_has_feature(CPU_FTR_DSCR)) {
1402 		PPC_WARN_EMULATED(mtdscr, regs);
1403 		rd = (instword >> 21) & 0x1f;
1404 		current->thread.dscr = regs->gpr[rd];
1405 		current->thread.dscr_inherit = 1;
1406 		mtspr(SPRN_DSCR, current->thread.dscr);
1407 		return 0;
1408 	}
1409 #endif
1410 
1411 	return -EINVAL;
1412 }
1413 
1414 int is_valid_bugaddr(unsigned long addr)
1415 {
1416 	return is_kernel_addr(addr);
1417 }
1418 
1419 #ifdef CONFIG_MATH_EMULATION
1420 static int emulate_math(struct pt_regs *regs)
1421 {
1422 	int ret;
1423 
1424 	ret = do_mathemu(regs);
1425 	if (ret >= 0)
1426 		PPC_WARN_EMULATED(math, regs);
1427 
1428 	switch (ret) {
1429 	case 0:
1430 		emulate_single_step(regs);
1431 		return 0;
1432 	case 1: {
1433 			int code = 0;
1434 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1435 			_exception(SIGFPE, regs, code, regs->nip);
1436 			return 0;
1437 		}
1438 	case -EFAULT:
1439 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1440 		return 0;
1441 	}
1442 
1443 	return -1;
1444 }
1445 #else
1446 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1447 #endif
1448 
1449 static void do_program_check(struct pt_regs *regs)
1450 {
1451 	unsigned int reason = get_reason(regs);
1452 
1453 	/* We can now get here via a FP Unavailable exception if the core
1454 	 * has no FPU, in that case the reason flags will be 0 */
1455 
1456 	if (reason & REASON_FP) {
1457 		/* IEEE FP exception */
1458 		parse_fpe(regs);
1459 		return;
1460 	}
1461 	if (reason & REASON_TRAP) {
1462 		unsigned long bugaddr;
1463 		/* Debugger is first in line to stop recursive faults in
1464 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1465 		if (debugger_bpt(regs))
1466 			return;
1467 
1468 		if (kprobe_handler(regs))
1469 			return;
1470 
1471 		/* trap exception */
1472 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1473 				== NOTIFY_STOP)
1474 			return;
1475 
1476 		bugaddr = regs->nip;
1477 		/*
1478 		 * Fixup bugaddr for BUG_ON() in real mode
1479 		 */
1480 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1481 			bugaddr += PAGE_OFFSET;
1482 
1483 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1484 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1485 			regs_add_return_ip(regs, 4);
1486 			return;
1487 		}
1488 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1489 		return;
1490 	}
1491 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1492 	if (reason & REASON_TM) {
1493 		/* This is a TM "Bad Thing Exception" program check.
1494 		 * This occurs when:
1495 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1496 		 *    transition in TM states.
1497 		 * -  A trechkpt is attempted when transactional.
1498 		 * -  A treclaim is attempted when non transactional.
1499 		 * -  A tend is illegally attempted.
1500 		 * -  writing a TM SPR when transactional.
1501 		 *
1502 		 * If usermode caused this, it's done something illegal and
1503 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1504 		 * operand to distinguish from the instruction just being bad
1505 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1506 		 * illegal /placement/ of a valid instruction.
1507 		 */
1508 		if (user_mode(regs)) {
1509 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1510 			return;
1511 		} else {
1512 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1513 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1514 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1515 			die("Unrecoverable exception", regs, SIGABRT);
1516 		}
1517 	}
1518 #endif
1519 
1520 	/*
1521 	 * If we took the program check in the kernel skip down to sending a
1522 	 * SIGILL. The subsequent cases all relate to emulating instructions
1523 	 * which we should only do for userspace. We also do not want to enable
1524 	 * interrupts for kernel faults because that might lead to further
1525 	 * faults, and loose the context of the original exception.
1526 	 */
1527 	if (!user_mode(regs))
1528 		goto sigill;
1529 
1530 	interrupt_cond_local_irq_enable(regs);
1531 
1532 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1533 	 * but there seems to be a hardware bug on the 405GP (RevD)
1534 	 * that means ESR is sometimes set incorrectly - either to
1535 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1536 	 * hardware people - not sure if it can happen on any illegal
1537 	 * instruction or only on FP instructions, whether there is a
1538 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1539 	 */
1540 	if (!emulate_math(regs))
1541 		return;
1542 
1543 	/* Try to emulate it if we should. */
1544 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1545 		switch (emulate_instruction(regs)) {
1546 		case 0:
1547 			regs_add_return_ip(regs, 4);
1548 			emulate_single_step(regs);
1549 			return;
1550 		case -EFAULT:
1551 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1552 			return;
1553 		}
1554 	}
1555 
1556 sigill:
1557 	if (reason & REASON_PRIVILEGED)
1558 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1559 	else
1560 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1561 
1562 }
1563 
1564 DEFINE_INTERRUPT_HANDLER(program_check_exception)
1565 {
1566 	do_program_check(regs);
1567 }
1568 
1569 /*
1570  * This occurs when running in hypervisor mode on POWER6 or later
1571  * and an illegal instruction is encountered.
1572  */
1573 DEFINE_INTERRUPT_HANDLER(emulation_assist_interrupt)
1574 {
1575 	regs_set_return_msr(regs, regs->msr | REASON_ILLEGAL);
1576 	do_program_check(regs);
1577 }
1578 
1579 DEFINE_INTERRUPT_HANDLER(alignment_exception)
1580 {
1581 	int sig, code, fixed = 0;
1582 	unsigned long  reason;
1583 
1584 	interrupt_cond_local_irq_enable(regs);
1585 
1586 	reason = get_reason(regs);
1587 	if (reason & REASON_BOUNDARY) {
1588 		sig = SIGBUS;
1589 		code = BUS_ADRALN;
1590 		goto bad;
1591 	}
1592 
1593 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1594 		return;
1595 
1596 	/* we don't implement logging of alignment exceptions */
1597 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1598 		fixed = fix_alignment(regs);
1599 
1600 	if (fixed == 1) {
1601 		/* skip over emulated instruction */
1602 		regs_add_return_ip(regs, inst_length(reason));
1603 		emulate_single_step(regs);
1604 		return;
1605 	}
1606 
1607 	/* Operand address was bad */
1608 	if (fixed == -EFAULT) {
1609 		sig = SIGSEGV;
1610 		code = SEGV_ACCERR;
1611 	} else {
1612 		sig = SIGBUS;
1613 		code = BUS_ADRALN;
1614 	}
1615 bad:
1616 	if (user_mode(regs))
1617 		_exception(sig, regs, code, regs->dar);
1618 	else
1619 		bad_page_fault(regs, sig);
1620 }
1621 
1622 DEFINE_INTERRUPT_HANDLER(stack_overflow_exception)
1623 {
1624 	die("Kernel stack overflow", regs, SIGSEGV);
1625 }
1626 
1627 DEFINE_INTERRUPT_HANDLER(kernel_fp_unavailable_exception)
1628 {
1629 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1630 			  "%lx at %lx\n", regs->trap, regs->nip);
1631 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1632 }
1633 
1634 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_exception)
1635 {
1636 	if (user_mode(regs)) {
1637 		/* A user program has executed an altivec instruction,
1638 		   but this kernel doesn't support altivec. */
1639 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1640 		return;
1641 	}
1642 
1643 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1644 			"%lx at %lx\n", regs->trap, regs->nip);
1645 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1646 }
1647 
1648 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_exception)
1649 {
1650 	if (user_mode(regs)) {
1651 		/* A user program has executed an vsx instruction,
1652 		   but this kernel doesn't support vsx. */
1653 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1654 		return;
1655 	}
1656 
1657 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1658 			"%lx at %lx\n", regs->trap, regs->nip);
1659 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1660 }
1661 
1662 #ifdef CONFIG_PPC64
1663 static void tm_unavailable(struct pt_regs *regs)
1664 {
1665 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1666 	if (user_mode(regs)) {
1667 		current->thread.load_tm++;
1668 		regs_set_return_msr(regs, regs->msr | MSR_TM);
1669 		tm_enable();
1670 		tm_restore_sprs(&current->thread);
1671 		return;
1672 	}
1673 #endif
1674 	pr_emerg("Unrecoverable TM Unavailable Exception "
1675 			"%lx at %lx\n", regs->trap, regs->nip);
1676 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1677 }
1678 
1679 DEFINE_INTERRUPT_HANDLER(facility_unavailable_exception)
1680 {
1681 	static char *facility_strings[] = {
1682 		[FSCR_FP_LG] = "FPU",
1683 		[FSCR_VECVSX_LG] = "VMX/VSX",
1684 		[FSCR_DSCR_LG] = "DSCR",
1685 		[FSCR_PM_LG] = "PMU SPRs",
1686 		[FSCR_BHRB_LG] = "BHRB",
1687 		[FSCR_TM_LG] = "TM",
1688 		[FSCR_EBB_LG] = "EBB",
1689 		[FSCR_TAR_LG] = "TAR",
1690 		[FSCR_MSGP_LG] = "MSGP",
1691 		[FSCR_SCV_LG] = "SCV",
1692 		[FSCR_PREFIX_LG] = "PREFIX",
1693 	};
1694 	char *facility = "unknown";
1695 	u64 value;
1696 	u32 instword, rd;
1697 	u8 status;
1698 	bool hv;
1699 
1700 	hv = (TRAP(regs) == INTERRUPT_H_FAC_UNAVAIL);
1701 	if (hv)
1702 		value = mfspr(SPRN_HFSCR);
1703 	else
1704 		value = mfspr(SPRN_FSCR);
1705 
1706 	status = value >> 56;
1707 	if ((hv || status >= 2) &&
1708 	    (status < ARRAY_SIZE(facility_strings)) &&
1709 	    facility_strings[status])
1710 		facility = facility_strings[status];
1711 
1712 	/* We should not have taken this interrupt in kernel */
1713 	if (!user_mode(regs)) {
1714 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1715 			 facility, status, regs->nip);
1716 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1717 	}
1718 
1719 	interrupt_cond_local_irq_enable(regs);
1720 
1721 	if (status == FSCR_DSCR_LG) {
1722 		/*
1723 		 * User is accessing the DSCR register using the problem
1724 		 * state only SPR number (0x03) either through a mfspr or
1725 		 * a mtspr instruction. If it is a write attempt through
1726 		 * a mtspr, then we set the inherit bit. This also allows
1727 		 * the user to write or read the register directly in the
1728 		 * future by setting via the FSCR DSCR bit. But in case it
1729 		 * is a read DSCR attempt through a mfspr instruction, we
1730 		 * just emulate the instruction instead. This code path will
1731 		 * always emulate all the mfspr instructions till the user
1732 		 * has attempted at least one mtspr instruction. This way it
1733 		 * preserves the same behaviour when the user is accessing
1734 		 * the DSCR through privilege level only SPR number (0x11)
1735 		 * which is emulated through illegal instruction exception.
1736 		 * We always leave HFSCR DSCR set.
1737 		 */
1738 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1739 			pr_err("Failed to fetch the user instruction\n");
1740 			return;
1741 		}
1742 
1743 		/* Write into DSCR (mtspr 0x03, RS) */
1744 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1745 				== PPC_INST_MTSPR_DSCR_USER) {
1746 			rd = (instword >> 21) & 0x1f;
1747 			current->thread.dscr = regs->gpr[rd];
1748 			current->thread.dscr_inherit = 1;
1749 			current->thread.fscr |= FSCR_DSCR;
1750 			mtspr(SPRN_FSCR, current->thread.fscr);
1751 		}
1752 
1753 		/* Read from DSCR (mfspr RT, 0x03) */
1754 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1755 				== PPC_INST_MFSPR_DSCR_USER) {
1756 			if (emulate_instruction(regs)) {
1757 				pr_err("DSCR based mfspr emulation failed\n");
1758 				return;
1759 			}
1760 			regs_add_return_ip(regs, 4);
1761 			emulate_single_step(regs);
1762 		}
1763 		return;
1764 	}
1765 
1766 	if (status == FSCR_TM_LG) {
1767 		/*
1768 		 * If we're here then the hardware is TM aware because it
1769 		 * generated an exception with FSRM_TM set.
1770 		 *
1771 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1772 		 * told us not to do TM, or the kernel is not built with TM
1773 		 * support.
1774 		 *
1775 		 * If both of those things are true, then userspace can spam the
1776 		 * console by triggering the printk() below just by continually
1777 		 * doing tbegin (or any TM instruction). So in that case just
1778 		 * send the process a SIGILL immediately.
1779 		 */
1780 		if (!cpu_has_feature(CPU_FTR_TM))
1781 			goto out;
1782 
1783 		tm_unavailable(regs);
1784 		return;
1785 	}
1786 
1787 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1788 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1789 
1790 out:
1791 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1792 }
1793 #endif
1794 
1795 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1796 
1797 DEFINE_INTERRUPT_HANDLER(fp_unavailable_tm)
1798 {
1799 	/* Note:  This does not handle any kind of FP laziness. */
1800 
1801 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1802 		 regs->nip, regs->msr);
1803 
1804         /* We can only have got here if the task started using FP after
1805          * beginning the transaction.  So, the transactional regs are just a
1806          * copy of the checkpointed ones.  But, we still need to recheckpoint
1807          * as we're enabling FP for the process; it will return, abort the
1808          * transaction, and probably retry but now with FP enabled.  So the
1809          * checkpointed FP registers need to be loaded.
1810 	 */
1811 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1812 
1813 	/*
1814 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1815 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1816 	 *
1817 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1818 	 * recheckpoint.
1819 	 */
1820 
1821 	/* Enable FP for the task: */
1822 	current->thread.load_fp = 1;
1823 
1824 	/*
1825 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1826 	 */
1827 	tm_recheckpoint(&current->thread);
1828 }
1829 
1830 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_tm)
1831 {
1832 	/* See the comments in fp_unavailable_tm().  This function operates
1833 	 * the same way.
1834 	 */
1835 
1836 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1837 		 "MSR=%lx\n",
1838 		 regs->nip, regs->msr);
1839 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1840 	current->thread.load_vec = 1;
1841 	tm_recheckpoint(&current->thread);
1842 	current->thread.used_vr = 1;
1843 }
1844 
1845 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_tm)
1846 {
1847 	/* See the comments in fp_unavailable_tm().  This works similarly,
1848 	 * though we're loading both FP and VEC registers in here.
1849 	 *
1850 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1851 	 * regs.  Either way, set MSR_VSX.
1852 	 */
1853 
1854 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1855 		 "MSR=%lx\n",
1856 		 regs->nip, regs->msr);
1857 
1858 	current->thread.used_vsr = 1;
1859 
1860 	/* This reclaims FP and/or VR regs if they're already enabled */
1861 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1862 
1863 	current->thread.load_vec = 1;
1864 	current->thread.load_fp = 1;
1865 
1866 	tm_recheckpoint(&current->thread);
1867 }
1868 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1869 
1870 #ifdef CONFIG_PPC64
1871 DECLARE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi);
1872 DEFINE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi)
1873 {
1874 	__this_cpu_inc(irq_stat.pmu_irqs);
1875 
1876 	perf_irq(regs);
1877 
1878 	return 0;
1879 }
1880 #endif
1881 
1882 DECLARE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async);
1883 DEFINE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async)
1884 {
1885 	__this_cpu_inc(irq_stat.pmu_irqs);
1886 
1887 	perf_irq(regs);
1888 }
1889 
1890 DEFINE_INTERRUPT_HANDLER_RAW(performance_monitor_exception)
1891 {
1892 	/*
1893 	 * On 64-bit, if perf interrupts hit in a local_irq_disable
1894 	 * (soft-masked) region, we consider them as NMIs. This is required to
1895 	 * prevent hash faults on user addresses when reading callchains (and
1896 	 * looks better from an irq tracing perspective).
1897 	 */
1898 	if (IS_ENABLED(CONFIG_PPC64) && unlikely(arch_irq_disabled_regs(regs)))
1899 		performance_monitor_exception_nmi(regs);
1900 	else
1901 		performance_monitor_exception_async(regs);
1902 
1903 	return 0;
1904 }
1905 
1906 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1907 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1908 {
1909 	int changed = 0;
1910 	/*
1911 	 * Determine the cause of the debug event, clear the
1912 	 * event flags and send a trap to the handler. Torez
1913 	 */
1914 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1915 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1916 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1917 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1918 #endif
1919 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1920 			     5);
1921 		changed |= 0x01;
1922 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1923 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1924 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1925 			     6);
1926 		changed |= 0x01;
1927 	}  else if (debug_status & DBSR_IAC1) {
1928 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1929 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1930 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1931 			     1);
1932 		changed |= 0x01;
1933 	}  else if (debug_status & DBSR_IAC2) {
1934 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1935 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1936 			     2);
1937 		changed |= 0x01;
1938 	}  else if (debug_status & DBSR_IAC3) {
1939 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1940 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1941 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1942 			     3);
1943 		changed |= 0x01;
1944 	}  else if (debug_status & DBSR_IAC4) {
1945 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1946 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1947 			     4);
1948 		changed |= 0x01;
1949 	}
1950 	/*
1951 	 * At the point this routine was called, the MSR(DE) was turned off.
1952 	 * Check all other debug flags and see if that bit needs to be turned
1953 	 * back on or not.
1954 	 */
1955 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1956 			       current->thread.debug.dbcr1))
1957 		regs_set_return_msr(regs, regs->msr | MSR_DE);
1958 	else
1959 		/* Make sure the IDM flag is off */
1960 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1961 
1962 	if (changed & 0x01)
1963 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1964 }
1965 
1966 DEFINE_INTERRUPT_HANDLER(DebugException)
1967 {
1968 	unsigned long debug_status = regs->dsisr;
1969 
1970 	current->thread.debug.dbsr = debug_status;
1971 
1972 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1973 	 * on server, it stops on the target of the branch. In order to simulate
1974 	 * the server behaviour, we thus restart right away with a single step
1975 	 * instead of stopping here when hitting a BT
1976 	 */
1977 	if (debug_status & DBSR_BT) {
1978 		regs_set_return_msr(regs, regs->msr & ~MSR_DE);
1979 
1980 		/* Disable BT */
1981 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1982 		/* Clear the BT event */
1983 		mtspr(SPRN_DBSR, DBSR_BT);
1984 
1985 		/* Do the single step trick only when coming from userspace */
1986 		if (user_mode(regs)) {
1987 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1988 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1989 			regs_set_return_msr(regs, regs->msr | MSR_DE);
1990 			return;
1991 		}
1992 
1993 		if (kprobe_post_handler(regs))
1994 			return;
1995 
1996 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1997 			       5, SIGTRAP) == NOTIFY_STOP) {
1998 			return;
1999 		}
2000 		if (debugger_sstep(regs))
2001 			return;
2002 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
2003 		regs_set_return_msr(regs, regs->msr & ~MSR_DE);
2004 
2005 		/* Disable instruction completion */
2006 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
2007 		/* Clear the instruction completion event */
2008 		mtspr(SPRN_DBSR, DBSR_IC);
2009 
2010 		if (kprobe_post_handler(regs))
2011 			return;
2012 
2013 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2014 			       5, SIGTRAP) == NOTIFY_STOP) {
2015 			return;
2016 		}
2017 
2018 		if (debugger_sstep(regs))
2019 			return;
2020 
2021 		if (user_mode(regs)) {
2022 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2023 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2024 					       current->thread.debug.dbcr1))
2025 				regs_set_return_msr(regs, regs->msr | MSR_DE);
2026 			else
2027 				/* Make sure the IDM bit is off */
2028 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2029 		}
2030 
2031 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2032 	} else
2033 		handle_debug(regs, debug_status);
2034 }
2035 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2036 
2037 #ifdef CONFIG_ALTIVEC
2038 DEFINE_INTERRUPT_HANDLER(altivec_assist_exception)
2039 {
2040 	int err;
2041 
2042 	if (!user_mode(regs)) {
2043 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2044 		       " at %lx\n", regs->nip);
2045 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2046 	}
2047 
2048 	flush_altivec_to_thread(current);
2049 
2050 	PPC_WARN_EMULATED(altivec, regs);
2051 	err = emulate_altivec(regs);
2052 	if (err == 0) {
2053 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2054 		emulate_single_step(regs);
2055 		return;
2056 	}
2057 
2058 	if (err == -EFAULT) {
2059 		/* got an error reading the instruction */
2060 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2061 	} else {
2062 		/* didn't recognize the instruction */
2063 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2064 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2065 				   "in %s at %lx\n", current->comm, regs->nip);
2066 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2067 	}
2068 }
2069 #endif /* CONFIG_ALTIVEC */
2070 
2071 #ifdef CONFIG_FSL_BOOKE
2072 DEFINE_INTERRUPT_HANDLER(CacheLockingException)
2073 {
2074 	unsigned long error_code = regs->dsisr;
2075 
2076 	/* We treat cache locking instructions from the user
2077 	 * as priv ops, in the future we could try to do
2078 	 * something smarter
2079 	 */
2080 	if (error_code & (ESR_DLK|ESR_ILK))
2081 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2082 	return;
2083 }
2084 #endif /* CONFIG_FSL_BOOKE */
2085 
2086 #ifdef CONFIG_SPE
2087 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointException)
2088 {
2089 	extern int do_spe_mathemu(struct pt_regs *regs);
2090 	unsigned long spefscr;
2091 	int fpexc_mode;
2092 	int code = FPE_FLTUNK;
2093 	int err;
2094 
2095 	interrupt_cond_local_irq_enable(regs);
2096 
2097 	flush_spe_to_thread(current);
2098 
2099 	spefscr = current->thread.spefscr;
2100 	fpexc_mode = current->thread.fpexc_mode;
2101 
2102 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2103 		code = FPE_FLTOVF;
2104 	}
2105 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2106 		code = FPE_FLTUND;
2107 	}
2108 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2109 		code = FPE_FLTDIV;
2110 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2111 		code = FPE_FLTINV;
2112 	}
2113 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2114 		code = FPE_FLTRES;
2115 
2116 	err = do_spe_mathemu(regs);
2117 	if (err == 0) {
2118 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2119 		emulate_single_step(regs);
2120 		return;
2121 	}
2122 
2123 	if (err == -EFAULT) {
2124 		/* got an error reading the instruction */
2125 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2126 	} else if (err == -EINVAL) {
2127 		/* didn't recognize the instruction */
2128 		printk(KERN_ERR "unrecognized spe instruction "
2129 		       "in %s at %lx\n", current->comm, regs->nip);
2130 	} else {
2131 		_exception(SIGFPE, regs, code, regs->nip);
2132 	}
2133 
2134 	return;
2135 }
2136 
2137 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointRoundException)
2138 {
2139 	extern int speround_handler(struct pt_regs *regs);
2140 	int err;
2141 
2142 	interrupt_cond_local_irq_enable(regs);
2143 
2144 	preempt_disable();
2145 	if (regs->msr & MSR_SPE)
2146 		giveup_spe(current);
2147 	preempt_enable();
2148 
2149 	regs_add_return_ip(regs, -4);
2150 	err = speround_handler(regs);
2151 	if (err == 0) {
2152 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2153 		emulate_single_step(regs);
2154 		return;
2155 	}
2156 
2157 	if (err == -EFAULT) {
2158 		/* got an error reading the instruction */
2159 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2160 	} else if (err == -EINVAL) {
2161 		/* didn't recognize the instruction */
2162 		printk(KERN_ERR "unrecognized spe instruction "
2163 		       "in %s at %lx\n", current->comm, regs->nip);
2164 	} else {
2165 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2166 		return;
2167 	}
2168 }
2169 #endif
2170 
2171 /*
2172  * We enter here if we get an unrecoverable exception, that is, one
2173  * that happened at a point where the RI (recoverable interrupt) bit
2174  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2175  * we therefore lost state by taking this exception.
2176  */
2177 void __noreturn unrecoverable_exception(struct pt_regs *regs)
2178 {
2179 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2180 		 regs->trap, regs->nip, regs->msr);
2181 	die("Unrecoverable exception", regs, SIGABRT);
2182 	/* die() should not return */
2183 	for (;;)
2184 		;
2185 }
2186 
2187 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2188 /*
2189  * Default handler for a Watchdog exception,
2190  * spins until a reboot occurs
2191  */
2192 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2193 {
2194 	/* Generic WatchdogHandler, implement your own */
2195 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2196 	return;
2197 }
2198 
2199 DEFINE_INTERRUPT_HANDLER_NMI(WatchdogException)
2200 {
2201 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2202 	WatchdogHandler(regs);
2203 	return 0;
2204 }
2205 #endif
2206 
2207 /*
2208  * We enter here if we discover during exception entry that we are
2209  * running in supervisor mode with a userspace value in the stack pointer.
2210  */
2211 DEFINE_INTERRUPT_HANDLER(kernel_bad_stack)
2212 {
2213 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2214 	       regs->gpr[1], regs->nip);
2215 	die("Bad kernel stack pointer", regs, SIGABRT);
2216 }
2217 
2218 void __init trap_init(void)
2219 {
2220 }
2221 
2222 
2223 #ifdef CONFIG_PPC_EMULATED_STATS
2224 
2225 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2226 
2227 struct ppc_emulated ppc_emulated = {
2228 #ifdef CONFIG_ALTIVEC
2229 	WARN_EMULATED_SETUP(altivec),
2230 #endif
2231 	WARN_EMULATED_SETUP(dcba),
2232 	WARN_EMULATED_SETUP(dcbz),
2233 	WARN_EMULATED_SETUP(fp_pair),
2234 	WARN_EMULATED_SETUP(isel),
2235 	WARN_EMULATED_SETUP(mcrxr),
2236 	WARN_EMULATED_SETUP(mfpvr),
2237 	WARN_EMULATED_SETUP(multiple),
2238 	WARN_EMULATED_SETUP(popcntb),
2239 	WARN_EMULATED_SETUP(spe),
2240 	WARN_EMULATED_SETUP(string),
2241 	WARN_EMULATED_SETUP(sync),
2242 	WARN_EMULATED_SETUP(unaligned),
2243 #ifdef CONFIG_MATH_EMULATION
2244 	WARN_EMULATED_SETUP(math),
2245 #endif
2246 #ifdef CONFIG_VSX
2247 	WARN_EMULATED_SETUP(vsx),
2248 #endif
2249 #ifdef CONFIG_PPC64
2250 	WARN_EMULATED_SETUP(mfdscr),
2251 	WARN_EMULATED_SETUP(mtdscr),
2252 	WARN_EMULATED_SETUP(lq_stq),
2253 	WARN_EMULATED_SETUP(lxvw4x),
2254 	WARN_EMULATED_SETUP(lxvh8x),
2255 	WARN_EMULATED_SETUP(lxvd2x),
2256 	WARN_EMULATED_SETUP(lxvb16x),
2257 #endif
2258 };
2259 
2260 u32 ppc_warn_emulated;
2261 
2262 void ppc_warn_emulated_print(const char *type)
2263 {
2264 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2265 			    type);
2266 }
2267 
2268 static int __init ppc_warn_emulated_init(void)
2269 {
2270 	struct dentry *dir;
2271 	unsigned int i;
2272 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2273 
2274 	dir = debugfs_create_dir("emulated_instructions",
2275 				 powerpc_debugfs_root);
2276 
2277 	debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated);
2278 
2279 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++)
2280 		debugfs_create_u32(entries[i].name, 0644, dir,
2281 				   (u32 *)&entries[i].val.counter);
2282 
2283 	return 0;
2284 }
2285 
2286 device_initcall(ppc_warn_emulated_init);
2287 
2288 #endif /* CONFIG_PPC_EMULATED_STATS */
2289