xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision c538938fa2cfdc806c6304888e3876729e6939e0)
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/io.h>
45 #include <asm/machdep.h>
46 #include <asm/rtas.h>
47 #include <asm/pmc.h>
48 #include <asm/reg.h>
49 #ifdef CONFIG_PMAC_BACKLIGHT
50 #include <asm/backlight.h>
51 #endif
52 #ifdef CONFIG_PPC64
53 #include <asm/firmware.h>
54 #include <asm/processor.h>
55 #include <asm/tm.h>
56 #endif
57 #include <asm/kexec.h>
58 #include <asm/ppc-opcode.h>
59 #include <asm/rio.h>
60 #include <asm/fadump.h>
61 #include <asm/switch_to.h>
62 #include <asm/tm.h>
63 #include <asm/debug.h>
64 #include <asm/asm-prototypes.h>
65 #include <asm/hmi.h>
66 #include <sysdev/fsl_pci.h>
67 #include <asm/kprobes.h>
68 #include <asm/stacktrace.h>
69 #include <asm/nmi.h>
70 
71 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
72 int (*__debugger)(struct pt_regs *regs) __read_mostly;
73 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
74 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
75 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
79 
80 EXPORT_SYMBOL(__debugger);
81 EXPORT_SYMBOL(__debugger_ipi);
82 EXPORT_SYMBOL(__debugger_bpt);
83 EXPORT_SYMBOL(__debugger_sstep);
84 EXPORT_SYMBOL(__debugger_iabr_match);
85 EXPORT_SYMBOL(__debugger_break_match);
86 EXPORT_SYMBOL(__debugger_fault_handler);
87 #endif
88 
89 /* Transactional Memory trap debug */
90 #ifdef TM_DEBUG_SW
91 #define TM_DEBUG(x...) printk(KERN_INFO x)
92 #else
93 #define TM_DEBUG(x...) do { } while(0)
94 #endif
95 
96 static const char *signame(int signr)
97 {
98 	switch (signr) {
99 	case SIGBUS:	return "bus error";
100 	case SIGFPE:	return "floating point exception";
101 	case SIGILL:	return "illegal instruction";
102 	case SIGSEGV:	return "segfault";
103 	case SIGTRAP:	return "unhandled trap";
104 	}
105 
106 	return "unknown signal";
107 }
108 
109 /*
110  * Trap & Exception support
111  */
112 
113 #ifdef CONFIG_PMAC_BACKLIGHT
114 static void pmac_backlight_unblank(void)
115 {
116 	mutex_lock(&pmac_backlight_mutex);
117 	if (pmac_backlight) {
118 		struct backlight_properties *props;
119 
120 		props = &pmac_backlight->props;
121 		props->brightness = props->max_brightness;
122 		props->power = FB_BLANK_UNBLANK;
123 		backlight_update_status(pmac_backlight);
124 	}
125 	mutex_unlock(&pmac_backlight_mutex);
126 }
127 #else
128 static inline void pmac_backlight_unblank(void) { }
129 #endif
130 
131 /*
132  * If oops/die is expected to crash the machine, return true here.
133  *
134  * This should not be expected to be 100% accurate, there may be
135  * notifiers registered or other unexpected conditions that may bring
136  * down the kernel. Or if the current process in the kernel is holding
137  * locks or has other critical state, the kernel may become effectively
138  * unusable anyway.
139  */
140 bool die_will_crash(void)
141 {
142 	if (should_fadump_crash())
143 		return true;
144 	if (kexec_should_crash(current))
145 		return true;
146 	if (in_interrupt() || panic_on_oops ||
147 			!current->pid || is_global_init(current))
148 		return true;
149 
150 	return false;
151 }
152 
153 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
154 static int die_owner = -1;
155 static unsigned int die_nest_count;
156 static int die_counter;
157 
158 extern void panic_flush_kmsg_start(void)
159 {
160 	/*
161 	 * These are mostly taken from kernel/panic.c, but tries to do
162 	 * relatively minimal work. Don't use delay functions (TB may
163 	 * be broken), don't crash dump (need to set a firmware log),
164 	 * don't run notifiers. We do want to get some information to
165 	 * Linux console.
166 	 */
167 	console_verbose();
168 	bust_spinlocks(1);
169 }
170 
171 extern void panic_flush_kmsg_end(void)
172 {
173 	printk_safe_flush_on_panic();
174 	kmsg_dump(KMSG_DUMP_PANIC);
175 	bust_spinlocks(0);
176 	debug_locks_off();
177 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
178 }
179 
180 static unsigned long oops_begin(struct pt_regs *regs)
181 {
182 	int cpu;
183 	unsigned long flags;
184 
185 	oops_enter();
186 
187 	/* racy, but better than risking deadlock. */
188 	raw_local_irq_save(flags);
189 	cpu = smp_processor_id();
190 	if (!arch_spin_trylock(&die_lock)) {
191 		if (cpu == die_owner)
192 			/* nested oops. should stop eventually */;
193 		else
194 			arch_spin_lock(&die_lock);
195 	}
196 	die_nest_count++;
197 	die_owner = cpu;
198 	console_verbose();
199 	bust_spinlocks(1);
200 	if (machine_is(powermac))
201 		pmac_backlight_unblank();
202 	return flags;
203 }
204 NOKPROBE_SYMBOL(oops_begin);
205 
206 static void oops_end(unsigned long flags, struct pt_regs *regs,
207 			       int signr)
208 {
209 	bust_spinlocks(0);
210 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
211 	die_nest_count--;
212 	oops_exit();
213 	printk("\n");
214 	if (!die_nest_count) {
215 		/* Nest count reaches zero, release the lock. */
216 		die_owner = -1;
217 		arch_spin_unlock(&die_lock);
218 	}
219 	raw_local_irq_restore(flags);
220 
221 	/*
222 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
223 	 */
224 	if (TRAP(regs) == 0x100)
225 		return;
226 
227 	crash_fadump(regs, "die oops");
228 
229 	if (kexec_should_crash(current))
230 		crash_kexec(regs);
231 
232 	if (!signr)
233 		return;
234 
235 	/*
236 	 * While our oops output is serialised by a spinlock, output
237 	 * from panic() called below can race and corrupt it. If we
238 	 * know we are going to panic, delay for 1 second so we have a
239 	 * chance to get clean backtraces from all CPUs that are oopsing.
240 	 */
241 	if (in_interrupt() || panic_on_oops || !current->pid ||
242 	    is_global_init(current)) {
243 		mdelay(MSEC_PER_SEC);
244 	}
245 
246 	if (panic_on_oops)
247 		panic("Fatal exception");
248 	do_exit(signr);
249 }
250 NOKPROBE_SYMBOL(oops_end);
251 
252 static char *get_mmu_str(void)
253 {
254 	if (early_radix_enabled())
255 		return " MMU=Radix";
256 	if (early_mmu_has_feature(MMU_FTR_HPTE_TABLE))
257 		return " MMU=Hash";
258 	return "";
259 }
260 
261 static int __die(const char *str, struct pt_regs *regs, long err)
262 {
263 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
264 
265 	printk("%s PAGE_SIZE=%luK%s%s%s%s%s%s %s\n",
266 	       IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE",
267 	       PAGE_SIZE / 1024, get_mmu_str(),
268 	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
269 	       IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
270 	       IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "",
271 	       debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
272 	       IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "",
273 	       ppc_md.name ? ppc_md.name : "");
274 
275 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
276 		return 1;
277 
278 	print_modules();
279 	show_regs(regs);
280 
281 	return 0;
282 }
283 NOKPROBE_SYMBOL(__die);
284 
285 void die(const char *str, struct pt_regs *regs, long err)
286 {
287 	unsigned long flags;
288 
289 	/*
290 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
291 	 */
292 	if (TRAP(regs) != 0x100) {
293 		if (debugger(regs))
294 			return;
295 	}
296 
297 	flags = oops_begin(regs);
298 	if (__die(str, regs, err))
299 		err = 0;
300 	oops_end(flags, regs, err);
301 }
302 NOKPROBE_SYMBOL(die);
303 
304 void user_single_step_report(struct pt_regs *regs)
305 {
306 	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip);
307 }
308 
309 static void show_signal_msg(int signr, struct pt_regs *regs, int code,
310 			    unsigned long addr)
311 {
312 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
313 				      DEFAULT_RATELIMIT_BURST);
314 
315 	if (!show_unhandled_signals)
316 		return;
317 
318 	if (!unhandled_signal(current, signr))
319 		return;
320 
321 	if (!__ratelimit(&rs))
322 		return;
323 
324 	pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x",
325 		current->comm, current->pid, signame(signr), signr,
326 		addr, regs->nip, regs->link, code);
327 
328 	print_vma_addr(KERN_CONT " in ", regs->nip);
329 
330 	pr_cont("\n");
331 
332 	show_user_instructions(regs);
333 }
334 
335 static bool exception_common(int signr, struct pt_regs *regs, int code,
336 			      unsigned long addr)
337 {
338 	if (!user_mode(regs)) {
339 		die("Exception in kernel mode", regs, signr);
340 		return false;
341 	}
342 
343 	show_signal_msg(signr, regs, code, addr);
344 
345 	if (arch_irqs_disabled() && !arch_irq_disabled_regs(regs))
346 		local_irq_enable();
347 
348 	current->thread.trap_nr = code;
349 
350 	return true;
351 }
352 
353 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key)
354 {
355 	if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr))
356 		return;
357 
358 	force_sig_pkuerr((void __user *) addr, key);
359 }
360 
361 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
362 {
363 	if (!exception_common(signr, regs, code, addr))
364 		return;
365 
366 	force_sig_fault(signr, code, (void __user *)addr);
367 }
368 
369 /*
370  * The interrupt architecture has a quirk in that the HV interrupts excluding
371  * the NMIs (0x100 and 0x200) do not clear MSR[RI] at entry. The first thing
372  * that an interrupt handler must do is save off a GPR into a scratch register,
373  * and all interrupts on POWERNV (HV=1) use the HSPRG1 register as scratch.
374  * Therefore an NMI can clobber an HV interrupt's live HSPRG1 without noticing
375  * that it is non-reentrant, which leads to random data corruption.
376  *
377  * The solution is for NMI interrupts in HV mode to check if they originated
378  * from these critical HV interrupt regions. If so, then mark them not
379  * recoverable.
380  *
381  * An alternative would be for HV NMIs to use SPRG for scratch to avoid the
382  * HSPRG1 clobber, however this would cause guest SPRG to be clobbered. Linux
383  * guests should always have MSR[RI]=0 when its scratch SPRG is in use, so
384  * that would work. However any other guest OS that may have the SPRG live
385  * and MSR[RI]=1 could encounter silent corruption.
386  *
387  * Builds that do not support KVM could take this second option to increase
388  * the recoverability of NMIs.
389  */
390 void hv_nmi_check_nonrecoverable(struct pt_regs *regs)
391 {
392 #ifdef CONFIG_PPC_POWERNV
393 	unsigned long kbase = (unsigned long)_stext;
394 	unsigned long nip = regs->nip;
395 
396 	if (!(regs->msr & MSR_RI))
397 		return;
398 	if (!(regs->msr & MSR_HV))
399 		return;
400 	if (regs->msr & MSR_PR)
401 		return;
402 
403 	/*
404 	 * Now test if the interrupt has hit a range that may be using
405 	 * HSPRG1 without having RI=0 (i.e., an HSRR interrupt). The
406 	 * problem ranges all run un-relocated. Test real and virt modes
407 	 * at the same time by droping the high bit of the nip (virt mode
408 	 * entry points still have the +0x4000 offset).
409 	 */
410 	nip &= ~0xc000000000000000ULL;
411 	if ((nip >= 0x500 && nip < 0x600) || (nip >= 0x4500 && nip < 0x4600))
412 		goto nonrecoverable;
413 	if ((nip >= 0x980 && nip < 0xa00) || (nip >= 0x4980 && nip < 0x4a00))
414 		goto nonrecoverable;
415 	if ((nip >= 0xe00 && nip < 0xec0) || (nip >= 0x4e00 && nip < 0x4ec0))
416 		goto nonrecoverable;
417 	if ((nip >= 0xf80 && nip < 0xfa0) || (nip >= 0x4f80 && nip < 0x4fa0))
418 		goto nonrecoverable;
419 
420 	/* Trampoline code runs un-relocated so subtract kbase. */
421 	if (nip >= (unsigned long)(start_real_trampolines - kbase) &&
422 			nip < (unsigned long)(end_real_trampolines - kbase))
423 		goto nonrecoverable;
424 	if (nip >= (unsigned long)(start_virt_trampolines - kbase) &&
425 			nip < (unsigned long)(end_virt_trampolines - kbase))
426 		goto nonrecoverable;
427 	return;
428 
429 nonrecoverable:
430 	regs->msr &= ~MSR_RI;
431 #endif
432 }
433 
434 void system_reset_exception(struct pt_regs *regs)
435 {
436 	unsigned long hsrr0, hsrr1;
437 	bool saved_hsrrs = false;
438 	u8 ftrace_enabled = this_cpu_get_ftrace_enabled();
439 
440 	this_cpu_set_ftrace_enabled(0);
441 
442 	nmi_enter();
443 
444 	/*
445 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
446 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
447 	 * OPAL), so save them here and restore them before returning.
448 	 *
449 	 * Machine checks don't need to save HSRRs, as the real mode handler
450 	 * is careful to avoid them, and the regular handler is not delivered
451 	 * as an NMI.
452 	 */
453 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
454 		hsrr0 = mfspr(SPRN_HSRR0);
455 		hsrr1 = mfspr(SPRN_HSRR1);
456 		saved_hsrrs = true;
457 	}
458 
459 	hv_nmi_check_nonrecoverable(regs);
460 
461 	__this_cpu_inc(irq_stat.sreset_irqs);
462 
463 	/* See if any machine dependent calls */
464 	if (ppc_md.system_reset_exception) {
465 		if (ppc_md.system_reset_exception(regs))
466 			goto out;
467 	}
468 
469 	if (debugger(regs))
470 		goto out;
471 
472 	kmsg_dump(KMSG_DUMP_OOPS);
473 	/*
474 	 * A system reset is a request to dump, so we always send
475 	 * it through the crashdump code (if fadump or kdump are
476 	 * registered).
477 	 */
478 	crash_fadump(regs, "System Reset");
479 
480 	crash_kexec(regs);
481 
482 	/*
483 	 * We aren't the primary crash CPU. We need to send it
484 	 * to a holding pattern to avoid it ending up in the panic
485 	 * code.
486 	 */
487 	crash_kexec_secondary(regs);
488 
489 	/*
490 	 * No debugger or crash dump registered, print logs then
491 	 * panic.
492 	 */
493 	die("System Reset", regs, SIGABRT);
494 
495 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
496 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
497 	nmi_panic(regs, "System Reset");
498 
499 out:
500 #ifdef CONFIG_PPC_BOOK3S_64
501 	BUG_ON(get_paca()->in_nmi == 0);
502 	if (get_paca()->in_nmi > 1)
503 		die("Unrecoverable nested System Reset", regs, SIGABRT);
504 #endif
505 	/* Must die if the interrupt is not recoverable */
506 	if (!(regs->msr & MSR_RI))
507 		die("Unrecoverable System Reset", regs, SIGABRT);
508 
509 	if (saved_hsrrs) {
510 		mtspr(SPRN_HSRR0, hsrr0);
511 		mtspr(SPRN_HSRR1, hsrr1);
512 	}
513 
514 	nmi_exit();
515 
516 	this_cpu_set_ftrace_enabled(ftrace_enabled);
517 
518 	/* What should we do here? We could issue a shutdown or hard reset. */
519 }
520 NOKPROBE_SYMBOL(system_reset_exception);
521 
522 /*
523  * I/O accesses can cause machine checks on powermacs.
524  * Check if the NIP corresponds to the address of a sync
525  * instruction for which there is an entry in the exception
526  * table.
527  *  -- paulus.
528  */
529 static inline int check_io_access(struct pt_regs *regs)
530 {
531 #ifdef CONFIG_PPC32
532 	unsigned long msr = regs->msr;
533 	const struct exception_table_entry *entry;
534 	unsigned int *nip = (unsigned int *)regs->nip;
535 
536 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
537 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
538 		/*
539 		 * Check that it's a sync instruction, or somewhere
540 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
541 		 * As the address is in the exception table
542 		 * we should be able to read the instr there.
543 		 * For the debug message, we look at the preceding
544 		 * load or store.
545 		 */
546 		if (*nip == PPC_INST_NOP)
547 			nip -= 2;
548 		else if (*nip == PPC_INST_ISYNC)
549 			--nip;
550 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
551 			unsigned int rb;
552 
553 			--nip;
554 			rb = (*nip >> 11) & 0x1f;
555 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
556 			       (*nip & 0x100)? "OUT to": "IN from",
557 			       regs->gpr[rb] - _IO_BASE, nip);
558 			regs->msr |= MSR_RI;
559 			regs->nip = extable_fixup(entry);
560 			return 1;
561 		}
562 	}
563 #endif /* CONFIG_PPC32 */
564 	return 0;
565 }
566 
567 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
568 /* On 4xx, the reason for the machine check or program exception
569    is in the ESR. */
570 #define get_reason(regs)	((regs)->dsisr)
571 #define REASON_FP		ESR_FP
572 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
573 #define REASON_PRIVILEGED	ESR_PPR
574 #define REASON_TRAP		ESR_PTR
575 #define REASON_PREFIXED		0
576 #define REASON_BOUNDARY		0
577 
578 /* single-step stuff */
579 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
580 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
581 #define clear_br_trace(regs)	do {} while(0)
582 #else
583 /* On non-4xx, the reason for the machine check or program
584    exception is in the MSR. */
585 #define get_reason(regs)	((regs)->msr)
586 #define REASON_TM		SRR1_PROGTM
587 #define REASON_FP		SRR1_PROGFPE
588 #define REASON_ILLEGAL		SRR1_PROGILL
589 #define REASON_PRIVILEGED	SRR1_PROGPRIV
590 #define REASON_TRAP		SRR1_PROGTRAP
591 #define REASON_PREFIXED		SRR1_PREFIXED
592 #define REASON_BOUNDARY		SRR1_BOUNDARY
593 
594 #define single_stepping(regs)	((regs)->msr & MSR_SE)
595 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
596 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
597 #endif
598 
599 #define inst_length(reason)	(((reason) & REASON_PREFIXED) ? 8 : 4)
600 
601 #if defined(CONFIG_E500)
602 int machine_check_e500mc(struct pt_regs *regs)
603 {
604 	unsigned long mcsr = mfspr(SPRN_MCSR);
605 	unsigned long pvr = mfspr(SPRN_PVR);
606 	unsigned long reason = mcsr;
607 	int recoverable = 1;
608 
609 	if (reason & MCSR_LD) {
610 		recoverable = fsl_rio_mcheck_exception(regs);
611 		if (recoverable == 1)
612 			goto silent_out;
613 	}
614 
615 	printk("Machine check in kernel mode.\n");
616 	printk("Caused by (from MCSR=%lx): ", reason);
617 
618 	if (reason & MCSR_MCP)
619 		pr_cont("Machine Check Signal\n");
620 
621 	if (reason & MCSR_ICPERR) {
622 		pr_cont("Instruction Cache Parity Error\n");
623 
624 		/*
625 		 * This is recoverable by invalidating the i-cache.
626 		 */
627 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
628 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
629 			;
630 
631 		/*
632 		 * This will generally be accompanied by an instruction
633 		 * fetch error report -- only treat MCSR_IF as fatal
634 		 * if it wasn't due to an L1 parity error.
635 		 */
636 		reason &= ~MCSR_IF;
637 	}
638 
639 	if (reason & MCSR_DCPERR_MC) {
640 		pr_cont("Data Cache Parity Error\n");
641 
642 		/*
643 		 * In write shadow mode we auto-recover from the error, but it
644 		 * may still get logged and cause a machine check.  We should
645 		 * only treat the non-write shadow case as non-recoverable.
646 		 */
647 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
648 		 * is not implemented but L1 data cache always runs in write
649 		 * shadow mode. Hence on data cache parity errors HW will
650 		 * automatically invalidate the L1 Data Cache.
651 		 */
652 		if (PVR_VER(pvr) != PVR_VER_E6500) {
653 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
654 				recoverable = 0;
655 		}
656 	}
657 
658 	if (reason & MCSR_L2MMU_MHIT) {
659 		pr_cont("Hit on multiple TLB entries\n");
660 		recoverable = 0;
661 	}
662 
663 	if (reason & MCSR_NMI)
664 		pr_cont("Non-maskable interrupt\n");
665 
666 	if (reason & MCSR_IF) {
667 		pr_cont("Instruction Fetch Error Report\n");
668 		recoverable = 0;
669 	}
670 
671 	if (reason & MCSR_LD) {
672 		pr_cont("Load Error Report\n");
673 		recoverable = 0;
674 	}
675 
676 	if (reason & MCSR_ST) {
677 		pr_cont("Store Error Report\n");
678 		recoverable = 0;
679 	}
680 
681 	if (reason & MCSR_LDG) {
682 		pr_cont("Guarded Load Error Report\n");
683 		recoverable = 0;
684 	}
685 
686 	if (reason & MCSR_TLBSYNC)
687 		pr_cont("Simultaneous tlbsync operations\n");
688 
689 	if (reason & MCSR_BSL2_ERR) {
690 		pr_cont("Level 2 Cache Error\n");
691 		recoverable = 0;
692 	}
693 
694 	if (reason & MCSR_MAV) {
695 		u64 addr;
696 
697 		addr = mfspr(SPRN_MCAR);
698 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
699 
700 		pr_cont("Machine Check %s Address: %#llx\n",
701 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
702 	}
703 
704 silent_out:
705 	mtspr(SPRN_MCSR, mcsr);
706 	return mfspr(SPRN_MCSR) == 0 && recoverable;
707 }
708 
709 int machine_check_e500(struct pt_regs *regs)
710 {
711 	unsigned long reason = mfspr(SPRN_MCSR);
712 
713 	if (reason & MCSR_BUS_RBERR) {
714 		if (fsl_rio_mcheck_exception(regs))
715 			return 1;
716 		if (fsl_pci_mcheck_exception(regs))
717 			return 1;
718 	}
719 
720 	printk("Machine check in kernel mode.\n");
721 	printk("Caused by (from MCSR=%lx): ", reason);
722 
723 	if (reason & MCSR_MCP)
724 		pr_cont("Machine Check Signal\n");
725 	if (reason & MCSR_ICPERR)
726 		pr_cont("Instruction Cache Parity Error\n");
727 	if (reason & MCSR_DCP_PERR)
728 		pr_cont("Data Cache Push Parity Error\n");
729 	if (reason & MCSR_DCPERR)
730 		pr_cont("Data Cache Parity Error\n");
731 	if (reason & MCSR_BUS_IAERR)
732 		pr_cont("Bus - Instruction Address Error\n");
733 	if (reason & MCSR_BUS_RAERR)
734 		pr_cont("Bus - Read Address Error\n");
735 	if (reason & MCSR_BUS_WAERR)
736 		pr_cont("Bus - Write Address Error\n");
737 	if (reason & MCSR_BUS_IBERR)
738 		pr_cont("Bus - Instruction Data Error\n");
739 	if (reason & MCSR_BUS_RBERR)
740 		pr_cont("Bus - Read Data Bus Error\n");
741 	if (reason & MCSR_BUS_WBERR)
742 		pr_cont("Bus - Write Data Bus Error\n");
743 	if (reason & MCSR_BUS_IPERR)
744 		pr_cont("Bus - Instruction Parity Error\n");
745 	if (reason & MCSR_BUS_RPERR)
746 		pr_cont("Bus - Read Parity Error\n");
747 
748 	return 0;
749 }
750 
751 int machine_check_generic(struct pt_regs *regs)
752 {
753 	return 0;
754 }
755 #elif defined(CONFIG_PPC32)
756 int machine_check_generic(struct pt_regs *regs)
757 {
758 	unsigned long reason = regs->msr;
759 
760 	printk("Machine check in kernel mode.\n");
761 	printk("Caused by (from SRR1=%lx): ", reason);
762 	switch (reason & 0x601F0000) {
763 	case 0x80000:
764 		pr_cont("Machine check signal\n");
765 		break;
766 	case 0x40000:
767 	case 0x140000:	/* 7450 MSS error and TEA */
768 		pr_cont("Transfer error ack signal\n");
769 		break;
770 	case 0x20000:
771 		pr_cont("Data parity error signal\n");
772 		break;
773 	case 0x10000:
774 		pr_cont("Address parity error signal\n");
775 		break;
776 	case 0x20000000:
777 		pr_cont("L1 Data Cache error\n");
778 		break;
779 	case 0x40000000:
780 		pr_cont("L1 Instruction Cache error\n");
781 		break;
782 	case 0x00100000:
783 		pr_cont("L2 data cache parity error\n");
784 		break;
785 	default:
786 		pr_cont("Unknown values in msr\n");
787 	}
788 	return 0;
789 }
790 #endif /* everything else */
791 
792 void die_mce(const char *str, struct pt_regs *regs, long err)
793 {
794 	/*
795 	 * The machine check wants to kill the interrupted context, but
796 	 * do_exit() checks for in_interrupt() and panics in that case, so
797 	 * exit the irq/nmi before calling die.
798 	 */
799 	if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64))
800 		nmi_exit();
801 	die(str, regs, err);
802 }
803 NOKPROBE_SYMBOL(die_mce);
804 
805 void machine_check_exception(struct pt_regs *regs)
806 {
807 	int recover = 0;
808 
809 	/*
810 	 * BOOK3S_64 does not call this handler as a non-maskable interrupt
811 	 * (it uses its own early real-mode handler to handle the MCE proper
812 	 * and then raises irq_work to call this handler when interrupts are
813 	 * enabled).
814 	 *
815 	 * This is silly. The BOOK3S_64 should just call a different function
816 	 * rather than expecting semantics to magically change. Something
817 	 * like 'non_nmi_machine_check_exception()', perhaps?
818 	 */
819 	const bool nmi = !IS_ENABLED(CONFIG_PPC_BOOK3S_64);
820 
821 	if (nmi) nmi_enter();
822 
823 	__this_cpu_inc(irq_stat.mce_exceptions);
824 
825 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
826 
827 	/* See if any machine dependent calls. In theory, we would want
828 	 * to call the CPU first, and call the ppc_md. one if the CPU
829 	 * one returns a positive number. However there is existing code
830 	 * that assumes the board gets a first chance, so let's keep it
831 	 * that way for now and fix things later. --BenH.
832 	 */
833 	if (ppc_md.machine_check_exception)
834 		recover = ppc_md.machine_check_exception(regs);
835 	else if (cur_cpu_spec->machine_check)
836 		recover = cur_cpu_spec->machine_check(regs);
837 
838 	if (recover > 0)
839 		goto bail;
840 
841 	if (debugger_fault_handler(regs))
842 		goto bail;
843 
844 	if (check_io_access(regs))
845 		goto bail;
846 
847 	die_mce("Machine check", regs, SIGBUS);
848 
849 bail:
850 	/* Must die if the interrupt is not recoverable */
851 	if (!(regs->msr & MSR_RI))
852 		die_mce("Unrecoverable Machine check", regs, SIGBUS);
853 
854 	if (nmi) nmi_exit();
855 }
856 NOKPROBE_SYMBOL(machine_check_exception);
857 
858 void SMIException(struct pt_regs *regs)
859 {
860 	die("System Management Interrupt", regs, SIGABRT);
861 }
862 
863 #ifdef CONFIG_VSX
864 static void p9_hmi_special_emu(struct pt_regs *regs)
865 {
866 	unsigned int ra, rb, t, i, sel, instr, rc;
867 	const void __user *addr;
868 	u8 vbuf[16] __aligned(16), *vdst;
869 	unsigned long ea, msr, msr_mask;
870 	bool swap;
871 
872 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
873 		return;
874 
875 	/*
876 	 * lxvb16x	opcode: 0x7c0006d8
877 	 * lxvd2x	opcode: 0x7c000698
878 	 * lxvh8x	opcode: 0x7c000658
879 	 * lxvw4x	opcode: 0x7c000618
880 	 */
881 	if ((instr & 0xfc00073e) != 0x7c000618) {
882 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
883 			 " instr=%08x\n",
884 			 smp_processor_id(), current->comm, current->pid,
885 			 regs->nip, instr);
886 		return;
887 	}
888 
889 	/* Grab vector registers into the task struct */
890 	msr = regs->msr; /* Grab msr before we flush the bits */
891 	flush_vsx_to_thread(current);
892 	enable_kernel_altivec();
893 
894 	/*
895 	 * Is userspace running with a different endian (this is rare but
896 	 * not impossible)
897 	 */
898 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
899 
900 	/* Decode the instruction */
901 	ra = (instr >> 16) & 0x1f;
902 	rb = (instr >> 11) & 0x1f;
903 	t = (instr >> 21) & 0x1f;
904 	if (instr & 1)
905 		vdst = (u8 *)&current->thread.vr_state.vr[t];
906 	else
907 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
908 
909 	/* Grab the vector address */
910 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
911 	if (is_32bit_task())
912 		ea &= 0xfffffffful;
913 	addr = (__force const void __user *)ea;
914 
915 	/* Check it */
916 	if (!access_ok(addr, 16)) {
917 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
918 			 " instr=%08x addr=%016lx\n",
919 			 smp_processor_id(), current->comm, current->pid,
920 			 regs->nip, instr, (unsigned long)addr);
921 		return;
922 	}
923 
924 	/* Read the vector */
925 	rc = 0;
926 	if ((unsigned long)addr & 0xfUL)
927 		/* unaligned case */
928 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
929 	else
930 		__get_user_atomic_128_aligned(vbuf, addr, rc);
931 	if (rc) {
932 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
933 			 " instr=%08x addr=%016lx\n",
934 			 smp_processor_id(), current->comm, current->pid,
935 			 regs->nip, instr, (unsigned long)addr);
936 		return;
937 	}
938 
939 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
940 		 " instr=%08x addr=%016lx\n",
941 		 smp_processor_id(), current->comm, current->pid, regs->nip,
942 		 instr, (unsigned long) addr);
943 
944 	/* Grab instruction "selector" */
945 	sel = (instr >> 6) & 3;
946 
947 	/*
948 	 * Check to make sure the facility is actually enabled. This
949 	 * could happen if we get a false positive hit.
950 	 *
951 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
952 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
953 	 */
954 	msr_mask = MSR_VSX;
955 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
956 		msr_mask = MSR_VEC;
957 	if (!(msr & msr_mask)) {
958 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
959 			 " instr=%08x msr:%016lx\n",
960 			 smp_processor_id(), current->comm, current->pid,
961 			 regs->nip, instr, msr);
962 		return;
963 	}
964 
965 	/* Do logging here before we modify sel based on endian */
966 	switch (sel) {
967 	case 0:	/* lxvw4x */
968 		PPC_WARN_EMULATED(lxvw4x, regs);
969 		break;
970 	case 1: /* lxvh8x */
971 		PPC_WARN_EMULATED(lxvh8x, regs);
972 		break;
973 	case 2: /* lxvd2x */
974 		PPC_WARN_EMULATED(lxvd2x, regs);
975 		break;
976 	case 3: /* lxvb16x */
977 		PPC_WARN_EMULATED(lxvb16x, regs);
978 		break;
979 	}
980 
981 #ifdef __LITTLE_ENDIAN__
982 	/*
983 	 * An LE kernel stores the vector in the task struct as an LE
984 	 * byte array (effectively swapping both the components and
985 	 * the content of the components). Those instructions expect
986 	 * the components to remain in ascending address order, so we
987 	 * swap them back.
988 	 *
989 	 * If we are running a BE user space, the expectation is that
990 	 * of a simple memcpy, so forcing the emulation to look like
991 	 * a lxvb16x should do the trick.
992 	 */
993 	if (swap)
994 		sel = 3;
995 
996 	switch (sel) {
997 	case 0:	/* lxvw4x */
998 		for (i = 0; i < 4; i++)
999 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1000 		break;
1001 	case 1: /* lxvh8x */
1002 		for (i = 0; i < 8; i++)
1003 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1004 		break;
1005 	case 2: /* lxvd2x */
1006 		for (i = 0; i < 2; i++)
1007 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1008 		break;
1009 	case 3: /* lxvb16x */
1010 		for (i = 0; i < 16; i++)
1011 			vdst[i] = vbuf[15-i];
1012 		break;
1013 	}
1014 #else /* __LITTLE_ENDIAN__ */
1015 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1016 	if (!swap)
1017 		sel = 3;
1018 
1019 	/* Otherwise, we need to swap the content of the components */
1020 	switch (sel) {
1021 	case 0:	/* lxvw4x */
1022 		for (i = 0; i < 4; i++)
1023 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1024 		break;
1025 	case 1: /* lxvh8x */
1026 		for (i = 0; i < 8; i++)
1027 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1028 		break;
1029 	case 2: /* lxvd2x */
1030 		for (i = 0; i < 2; i++)
1031 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1032 		break;
1033 	case 3: /* lxvb16x */
1034 		memcpy(vdst, vbuf, 16);
1035 		break;
1036 	}
1037 #endif /* !__LITTLE_ENDIAN__ */
1038 
1039 	/* Go to next instruction */
1040 	regs->nip += 4;
1041 }
1042 #endif /* CONFIG_VSX */
1043 
1044 void handle_hmi_exception(struct pt_regs *regs)
1045 {
1046 	struct pt_regs *old_regs;
1047 
1048 	old_regs = set_irq_regs(regs);
1049 	irq_enter();
1050 
1051 #ifdef CONFIG_VSX
1052 	/* Real mode flagged P9 special emu is needed */
1053 	if (local_paca->hmi_p9_special_emu) {
1054 		local_paca->hmi_p9_special_emu = 0;
1055 
1056 		/*
1057 		 * We don't want to take page faults while doing the
1058 		 * emulation, we just replay the instruction if necessary.
1059 		 */
1060 		pagefault_disable();
1061 		p9_hmi_special_emu(regs);
1062 		pagefault_enable();
1063 	}
1064 #endif /* CONFIG_VSX */
1065 
1066 	if (ppc_md.handle_hmi_exception)
1067 		ppc_md.handle_hmi_exception(regs);
1068 
1069 	irq_exit();
1070 	set_irq_regs(old_regs);
1071 }
1072 
1073 void unknown_exception(struct pt_regs *regs)
1074 {
1075 	enum ctx_state prev_state = exception_enter();
1076 
1077 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1078 	       regs->nip, regs->msr, regs->trap);
1079 
1080 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1081 
1082 	exception_exit(prev_state);
1083 }
1084 
1085 void unknown_async_exception(struct pt_regs *regs)
1086 {
1087 	enum ctx_state prev_state = exception_enter();
1088 
1089 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1090 	       regs->nip, regs->msr, regs->trap);
1091 
1092 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1093 
1094 	exception_exit(prev_state);
1095 }
1096 
1097 void instruction_breakpoint_exception(struct pt_regs *regs)
1098 {
1099 	enum ctx_state prev_state = exception_enter();
1100 
1101 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1102 					5, SIGTRAP) == NOTIFY_STOP)
1103 		goto bail;
1104 	if (debugger_iabr_match(regs))
1105 		goto bail;
1106 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1107 
1108 bail:
1109 	exception_exit(prev_state);
1110 }
1111 
1112 void RunModeException(struct pt_regs *regs)
1113 {
1114 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1115 }
1116 
1117 void single_step_exception(struct pt_regs *regs)
1118 {
1119 	enum ctx_state prev_state = exception_enter();
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 		goto bail;
1130 	if (debugger_sstep(regs))
1131 		goto bail;
1132 
1133 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1134 
1135 bail:
1136 	exception_exit(prev_state);
1137 }
1138 NOKPROBE_SYMBOL(single_step_exception);
1139 
1140 /*
1141  * After we have successfully emulated an instruction, we have to
1142  * check if the instruction was being single-stepped, and if so,
1143  * pretend we got a single-step exception.  This was pointed out
1144  * by Kumar Gala.  -- paulus
1145  */
1146 static void emulate_single_step(struct pt_regs *regs)
1147 {
1148 	if (single_stepping(regs))
1149 		single_step_exception(regs);
1150 }
1151 
1152 static inline int __parse_fpscr(unsigned long fpscr)
1153 {
1154 	int ret = FPE_FLTUNK;
1155 
1156 	/* Invalid operation */
1157 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1158 		ret = FPE_FLTINV;
1159 
1160 	/* Overflow */
1161 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1162 		ret = FPE_FLTOVF;
1163 
1164 	/* Underflow */
1165 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1166 		ret = FPE_FLTUND;
1167 
1168 	/* Divide by zero */
1169 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1170 		ret = FPE_FLTDIV;
1171 
1172 	/* Inexact result */
1173 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1174 		ret = FPE_FLTRES;
1175 
1176 	return ret;
1177 }
1178 
1179 static void parse_fpe(struct pt_regs *regs)
1180 {
1181 	int code = 0;
1182 
1183 	flush_fp_to_thread(current);
1184 
1185 #ifdef CONFIG_PPC_FPU_REGS
1186 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1187 #endif
1188 
1189 	_exception(SIGFPE, regs, code, regs->nip);
1190 }
1191 
1192 /*
1193  * Illegal instruction emulation support.  Originally written to
1194  * provide the PVR to user applications using the mfspr rd, PVR.
1195  * Return non-zero if we can't emulate, or -EFAULT if the associated
1196  * memory access caused an access fault.  Return zero on success.
1197  *
1198  * There are a couple of ways to do this, either "decode" the instruction
1199  * or directly match lots of bits.  In this case, matching lots of
1200  * bits is faster and easier.
1201  *
1202  */
1203 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1204 {
1205 	u8 rT = (instword >> 21) & 0x1f;
1206 	u8 rA = (instword >> 16) & 0x1f;
1207 	u8 NB_RB = (instword >> 11) & 0x1f;
1208 	u32 num_bytes;
1209 	unsigned long EA;
1210 	int pos = 0;
1211 
1212 	/* Early out if we are an invalid form of lswx */
1213 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1214 		if ((rT == rA) || (rT == NB_RB))
1215 			return -EINVAL;
1216 
1217 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1218 
1219 	switch (instword & PPC_INST_STRING_MASK) {
1220 		case PPC_INST_LSWX:
1221 		case PPC_INST_STSWX:
1222 			EA += NB_RB;
1223 			num_bytes = regs->xer & 0x7f;
1224 			break;
1225 		case PPC_INST_LSWI:
1226 		case PPC_INST_STSWI:
1227 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1228 			break;
1229 		default:
1230 			return -EINVAL;
1231 	}
1232 
1233 	while (num_bytes != 0)
1234 	{
1235 		u8 val;
1236 		u32 shift = 8 * (3 - (pos & 0x3));
1237 
1238 		/* if process is 32-bit, clear upper 32 bits of EA */
1239 		if ((regs->msr & MSR_64BIT) == 0)
1240 			EA &= 0xFFFFFFFF;
1241 
1242 		switch ((instword & PPC_INST_STRING_MASK)) {
1243 			case PPC_INST_LSWX:
1244 			case PPC_INST_LSWI:
1245 				if (get_user(val, (u8 __user *)EA))
1246 					return -EFAULT;
1247 				/* first time updating this reg,
1248 				 * zero it out */
1249 				if (pos == 0)
1250 					regs->gpr[rT] = 0;
1251 				regs->gpr[rT] |= val << shift;
1252 				break;
1253 			case PPC_INST_STSWI:
1254 			case PPC_INST_STSWX:
1255 				val = regs->gpr[rT] >> shift;
1256 				if (put_user(val, (u8 __user *)EA))
1257 					return -EFAULT;
1258 				break;
1259 		}
1260 		/* move EA to next address */
1261 		EA += 1;
1262 		num_bytes--;
1263 
1264 		/* manage our position within the register */
1265 		if (++pos == 4) {
1266 			pos = 0;
1267 			if (++rT == 32)
1268 				rT = 0;
1269 		}
1270 	}
1271 
1272 	return 0;
1273 }
1274 
1275 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1276 {
1277 	u32 ra,rs;
1278 	unsigned long tmp;
1279 
1280 	ra = (instword >> 16) & 0x1f;
1281 	rs = (instword >> 21) & 0x1f;
1282 
1283 	tmp = regs->gpr[rs];
1284 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1285 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1286 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1287 	regs->gpr[ra] = tmp;
1288 
1289 	return 0;
1290 }
1291 
1292 static int emulate_isel(struct pt_regs *regs, u32 instword)
1293 {
1294 	u8 rT = (instword >> 21) & 0x1f;
1295 	u8 rA = (instword >> 16) & 0x1f;
1296 	u8 rB = (instword >> 11) & 0x1f;
1297 	u8 BC = (instword >> 6) & 0x1f;
1298 	u8 bit;
1299 	unsigned long tmp;
1300 
1301 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1302 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1303 
1304 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1305 
1306 	return 0;
1307 }
1308 
1309 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1310 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1311 {
1312         /* If we're emulating a load/store in an active transaction, we cannot
1313          * emulate it as the kernel operates in transaction suspended context.
1314          * We need to abort the transaction.  This creates a persistent TM
1315          * abort so tell the user what caused it with a new code.
1316 	 */
1317 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1318 		tm_enable();
1319 		tm_abort(cause);
1320 		return true;
1321 	}
1322 	return false;
1323 }
1324 #else
1325 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1326 {
1327 	return false;
1328 }
1329 #endif
1330 
1331 static int emulate_instruction(struct pt_regs *regs)
1332 {
1333 	u32 instword;
1334 	u32 rd;
1335 
1336 	if (!user_mode(regs))
1337 		return -EINVAL;
1338 	CHECK_FULL_REGS(regs);
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 	extern int do_mathemu(struct pt_regs *regs);
1436 
1437 	ret = do_mathemu(regs);
1438 	if (ret >= 0)
1439 		PPC_WARN_EMULATED(math, regs);
1440 
1441 	switch (ret) {
1442 	case 0:
1443 		emulate_single_step(regs);
1444 		return 0;
1445 	case 1: {
1446 			int code = 0;
1447 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1448 			_exception(SIGFPE, regs, code, regs->nip);
1449 			return 0;
1450 		}
1451 	case -EFAULT:
1452 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1453 		return 0;
1454 	}
1455 
1456 	return -1;
1457 }
1458 #else
1459 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1460 #endif
1461 
1462 void program_check_exception(struct pt_regs *regs)
1463 {
1464 	enum ctx_state prev_state = exception_enter();
1465 	unsigned int reason = get_reason(regs);
1466 
1467 	/* We can now get here via a FP Unavailable exception if the core
1468 	 * has no FPU, in that case the reason flags will be 0 */
1469 
1470 	if (reason & REASON_FP) {
1471 		/* IEEE FP exception */
1472 		parse_fpe(regs);
1473 		goto bail;
1474 	}
1475 	if (reason & REASON_TRAP) {
1476 		unsigned long bugaddr;
1477 		/* Debugger is first in line to stop recursive faults in
1478 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1479 		if (debugger_bpt(regs))
1480 			goto bail;
1481 
1482 		if (kprobe_handler(regs))
1483 			goto bail;
1484 
1485 		/* trap exception */
1486 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1487 				== NOTIFY_STOP)
1488 			goto bail;
1489 
1490 		bugaddr = regs->nip;
1491 		/*
1492 		 * Fixup bugaddr for BUG_ON() in real mode
1493 		 */
1494 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1495 			bugaddr += PAGE_OFFSET;
1496 
1497 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1498 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1499 			regs->nip += 4;
1500 			goto bail;
1501 		}
1502 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1503 		goto bail;
1504 	}
1505 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1506 	if (reason & REASON_TM) {
1507 		/* This is a TM "Bad Thing Exception" program check.
1508 		 * This occurs when:
1509 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1510 		 *    transition in TM states.
1511 		 * -  A trechkpt is attempted when transactional.
1512 		 * -  A treclaim is attempted when non transactional.
1513 		 * -  A tend is illegally attempted.
1514 		 * -  writing a TM SPR when transactional.
1515 		 *
1516 		 * If usermode caused this, it's done something illegal and
1517 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1518 		 * operand to distinguish from the instruction just being bad
1519 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1520 		 * illegal /placement/ of a valid instruction.
1521 		 */
1522 		if (user_mode(regs)) {
1523 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1524 			goto bail;
1525 		} else {
1526 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1527 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1528 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1529 			die("Unrecoverable exception", regs, SIGABRT);
1530 		}
1531 	}
1532 #endif
1533 
1534 	/*
1535 	 * If we took the program check in the kernel skip down to sending a
1536 	 * SIGILL. The subsequent cases all relate to emulating instructions
1537 	 * which we should only do for userspace. We also do not want to enable
1538 	 * interrupts for kernel faults because that might lead to further
1539 	 * faults, and loose the context of the original exception.
1540 	 */
1541 	if (!user_mode(regs))
1542 		goto sigill;
1543 
1544 	/* We restore the interrupt state now */
1545 	if (!arch_irq_disabled_regs(regs))
1546 		local_irq_enable();
1547 
1548 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1549 	 * but there seems to be a hardware bug on the 405GP (RevD)
1550 	 * that means ESR is sometimes set incorrectly - either to
1551 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1552 	 * hardware people - not sure if it can happen on any illegal
1553 	 * instruction or only on FP instructions, whether there is a
1554 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1555 	 */
1556 	if (!emulate_math(regs))
1557 		goto bail;
1558 
1559 	/* Try to emulate it if we should. */
1560 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1561 		switch (emulate_instruction(regs)) {
1562 		case 0:
1563 			regs->nip += 4;
1564 			emulate_single_step(regs);
1565 			goto bail;
1566 		case -EFAULT:
1567 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1568 			goto bail;
1569 		}
1570 	}
1571 
1572 sigill:
1573 	if (reason & REASON_PRIVILEGED)
1574 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1575 	else
1576 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1577 
1578 bail:
1579 	exception_exit(prev_state);
1580 }
1581 NOKPROBE_SYMBOL(program_check_exception);
1582 
1583 /*
1584  * This occurs when running in hypervisor mode on POWER6 or later
1585  * and an illegal instruction is encountered.
1586  */
1587 void emulation_assist_interrupt(struct pt_regs *regs)
1588 {
1589 	regs->msr |= REASON_ILLEGAL;
1590 	program_check_exception(regs);
1591 }
1592 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1593 
1594 void alignment_exception(struct pt_regs *regs)
1595 {
1596 	enum ctx_state prev_state = exception_enter();
1597 	int sig, code, fixed = 0;
1598 	unsigned long  reason;
1599 
1600 	/* We restore the interrupt state now */
1601 	if (!arch_irq_disabled_regs(regs))
1602 		local_irq_enable();
1603 
1604 	reason = get_reason(regs);
1605 
1606 	if (reason & REASON_BOUNDARY) {
1607 		sig = SIGBUS;
1608 		code = BUS_ADRALN;
1609 		goto bad;
1610 	}
1611 
1612 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1613 		goto bail;
1614 
1615 	/* we don't implement logging of alignment exceptions */
1616 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1617 		fixed = fix_alignment(regs);
1618 
1619 	if (fixed == 1) {
1620 		/* skip over emulated instruction */
1621 		regs->nip += inst_length(reason);
1622 		emulate_single_step(regs);
1623 		goto bail;
1624 	}
1625 
1626 	/* Operand address was bad */
1627 	if (fixed == -EFAULT) {
1628 		sig = SIGSEGV;
1629 		code = SEGV_ACCERR;
1630 	} else {
1631 		sig = SIGBUS;
1632 		code = BUS_ADRALN;
1633 	}
1634 bad:
1635 	if (user_mode(regs))
1636 		_exception(sig, regs, code, regs->dar);
1637 	else
1638 		bad_page_fault(regs, sig);
1639 
1640 bail:
1641 	exception_exit(prev_state);
1642 }
1643 
1644 void StackOverflow(struct pt_regs *regs)
1645 {
1646 	pr_crit("Kernel stack overflow in process %s[%d], r1=%lx\n",
1647 		current->comm, task_pid_nr(current), regs->gpr[1]);
1648 	debugger(regs);
1649 	show_regs(regs);
1650 	panic("kernel stack overflow");
1651 }
1652 
1653 void stack_overflow_exception(struct pt_regs *regs)
1654 {
1655 	enum ctx_state prev_state = exception_enter();
1656 
1657 	die("Kernel stack overflow", regs, SIGSEGV);
1658 
1659 	exception_exit(prev_state);
1660 }
1661 
1662 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1663 {
1664 	enum ctx_state prev_state = exception_enter();
1665 
1666 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1667 			  "%lx at %lx\n", regs->trap, regs->nip);
1668 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1669 
1670 	exception_exit(prev_state);
1671 }
1672 
1673 void altivec_unavailable_exception(struct pt_regs *regs)
1674 {
1675 	enum ctx_state prev_state = exception_enter();
1676 
1677 	if (user_mode(regs)) {
1678 		/* A user program has executed an altivec instruction,
1679 		   but this kernel doesn't support altivec. */
1680 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1681 		goto bail;
1682 	}
1683 
1684 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1685 			"%lx at %lx\n", regs->trap, regs->nip);
1686 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1687 
1688 bail:
1689 	exception_exit(prev_state);
1690 }
1691 
1692 void vsx_unavailable_exception(struct pt_regs *regs)
1693 {
1694 	if (user_mode(regs)) {
1695 		/* A user program has executed an vsx instruction,
1696 		   but this kernel doesn't support vsx. */
1697 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1698 		return;
1699 	}
1700 
1701 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1702 			"%lx at %lx\n", regs->trap, regs->nip);
1703 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1704 }
1705 
1706 #ifdef CONFIG_PPC64
1707 static void tm_unavailable(struct pt_regs *regs)
1708 {
1709 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1710 	if (user_mode(regs)) {
1711 		current->thread.load_tm++;
1712 		regs->msr |= MSR_TM;
1713 		tm_enable();
1714 		tm_restore_sprs(&current->thread);
1715 		return;
1716 	}
1717 #endif
1718 	pr_emerg("Unrecoverable TM Unavailable Exception "
1719 			"%lx at %lx\n", regs->trap, regs->nip);
1720 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1721 }
1722 
1723 void facility_unavailable_exception(struct pt_regs *regs)
1724 {
1725 	static char *facility_strings[] = {
1726 		[FSCR_FP_LG] = "FPU",
1727 		[FSCR_VECVSX_LG] = "VMX/VSX",
1728 		[FSCR_DSCR_LG] = "DSCR",
1729 		[FSCR_PM_LG] = "PMU SPRs",
1730 		[FSCR_BHRB_LG] = "BHRB",
1731 		[FSCR_TM_LG] = "TM",
1732 		[FSCR_EBB_LG] = "EBB",
1733 		[FSCR_TAR_LG] = "TAR",
1734 		[FSCR_MSGP_LG] = "MSGP",
1735 		[FSCR_SCV_LG] = "SCV",
1736 		[FSCR_PREFIX_LG] = "PREFIX",
1737 	};
1738 	char *facility = "unknown";
1739 	u64 value;
1740 	u32 instword, rd;
1741 	u8 status;
1742 	bool hv;
1743 
1744 	hv = (TRAP(regs) == 0xf80);
1745 	if (hv)
1746 		value = mfspr(SPRN_HFSCR);
1747 	else
1748 		value = mfspr(SPRN_FSCR);
1749 
1750 	status = value >> 56;
1751 	if ((hv || status >= 2) &&
1752 	    (status < ARRAY_SIZE(facility_strings)) &&
1753 	    facility_strings[status])
1754 		facility = facility_strings[status];
1755 
1756 	/* We should not have taken this interrupt in kernel */
1757 	if (!user_mode(regs)) {
1758 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1759 			 facility, status, regs->nip);
1760 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1761 	}
1762 
1763 	/* We restore the interrupt state now */
1764 	if (!arch_irq_disabled_regs(regs))
1765 		local_irq_enable();
1766 
1767 	if (status == FSCR_DSCR_LG) {
1768 		/*
1769 		 * User is accessing the DSCR register using the problem
1770 		 * state only SPR number (0x03) either through a mfspr or
1771 		 * a mtspr instruction. If it is a write attempt through
1772 		 * a mtspr, then we set the inherit bit. This also allows
1773 		 * the user to write or read the register directly in the
1774 		 * future by setting via the FSCR DSCR bit. But in case it
1775 		 * is a read DSCR attempt through a mfspr instruction, we
1776 		 * just emulate the instruction instead. This code path will
1777 		 * always emulate all the mfspr instructions till the user
1778 		 * has attempted at least one mtspr instruction. This way it
1779 		 * preserves the same behaviour when the user is accessing
1780 		 * the DSCR through privilege level only SPR number (0x11)
1781 		 * which is emulated through illegal instruction exception.
1782 		 * We always leave HFSCR DSCR set.
1783 		 */
1784 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1785 			pr_err("Failed to fetch the user instruction\n");
1786 			return;
1787 		}
1788 
1789 		/* Write into DSCR (mtspr 0x03, RS) */
1790 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1791 				== PPC_INST_MTSPR_DSCR_USER) {
1792 			rd = (instword >> 21) & 0x1f;
1793 			current->thread.dscr = regs->gpr[rd];
1794 			current->thread.dscr_inherit = 1;
1795 			current->thread.fscr |= FSCR_DSCR;
1796 			mtspr(SPRN_FSCR, current->thread.fscr);
1797 		}
1798 
1799 		/* Read from DSCR (mfspr RT, 0x03) */
1800 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1801 				== PPC_INST_MFSPR_DSCR_USER) {
1802 			if (emulate_instruction(regs)) {
1803 				pr_err("DSCR based mfspr emulation failed\n");
1804 				return;
1805 			}
1806 			regs->nip += 4;
1807 			emulate_single_step(regs);
1808 		}
1809 		return;
1810 	}
1811 
1812 	if (status == FSCR_TM_LG) {
1813 		/*
1814 		 * If we're here then the hardware is TM aware because it
1815 		 * generated an exception with FSRM_TM set.
1816 		 *
1817 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1818 		 * told us not to do TM, or the kernel is not built with TM
1819 		 * support.
1820 		 *
1821 		 * If both of those things are true, then userspace can spam the
1822 		 * console by triggering the printk() below just by continually
1823 		 * doing tbegin (or any TM instruction). So in that case just
1824 		 * send the process a SIGILL immediately.
1825 		 */
1826 		if (!cpu_has_feature(CPU_FTR_TM))
1827 			goto out;
1828 
1829 		tm_unavailable(regs);
1830 		return;
1831 	}
1832 
1833 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1834 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1835 
1836 out:
1837 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1838 }
1839 #endif
1840 
1841 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1842 
1843 void fp_unavailable_tm(struct pt_regs *regs)
1844 {
1845 	/* Note:  This does not handle any kind of FP laziness. */
1846 
1847 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1848 		 regs->nip, regs->msr);
1849 
1850         /* We can only have got here if the task started using FP after
1851          * beginning the transaction.  So, the transactional regs are just a
1852          * copy of the checkpointed ones.  But, we still need to recheckpoint
1853          * as we're enabling FP for the process; it will return, abort the
1854          * transaction, and probably retry but now with FP enabled.  So the
1855          * checkpointed FP registers need to be loaded.
1856 	 */
1857 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1858 
1859 	/*
1860 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1861 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1862 	 *
1863 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1864 	 * recheckpoint.
1865 	 */
1866 
1867 	/* Enable FP for the task: */
1868 	current->thread.load_fp = 1;
1869 
1870 	/*
1871 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1872 	 */
1873 	tm_recheckpoint(&current->thread);
1874 }
1875 
1876 void altivec_unavailable_tm(struct pt_regs *regs)
1877 {
1878 	/* See the comments in fp_unavailable_tm().  This function operates
1879 	 * the same way.
1880 	 */
1881 
1882 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1883 		 "MSR=%lx\n",
1884 		 regs->nip, regs->msr);
1885 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1886 	current->thread.load_vec = 1;
1887 	tm_recheckpoint(&current->thread);
1888 	current->thread.used_vr = 1;
1889 }
1890 
1891 void vsx_unavailable_tm(struct pt_regs *regs)
1892 {
1893 	/* See the comments in fp_unavailable_tm().  This works similarly,
1894 	 * though we're loading both FP and VEC registers in here.
1895 	 *
1896 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1897 	 * regs.  Either way, set MSR_VSX.
1898 	 */
1899 
1900 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1901 		 "MSR=%lx\n",
1902 		 regs->nip, regs->msr);
1903 
1904 	current->thread.used_vsr = 1;
1905 
1906 	/* This reclaims FP and/or VR regs if they're already enabled */
1907 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1908 
1909 	current->thread.load_vec = 1;
1910 	current->thread.load_fp = 1;
1911 
1912 	tm_recheckpoint(&current->thread);
1913 }
1914 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1915 
1916 static void performance_monitor_exception_nmi(struct pt_regs *regs)
1917 {
1918 	nmi_enter();
1919 
1920 	__this_cpu_inc(irq_stat.pmu_irqs);
1921 
1922 	perf_irq(regs);
1923 
1924 	nmi_exit();
1925 }
1926 
1927 static void performance_monitor_exception_async(struct pt_regs *regs)
1928 {
1929 	irq_enter();
1930 
1931 	__this_cpu_inc(irq_stat.pmu_irqs);
1932 
1933 	perf_irq(regs);
1934 
1935 	irq_exit();
1936 }
1937 
1938 void performance_monitor_exception(struct pt_regs *regs)
1939 {
1940 	/*
1941 	 * On 64-bit, if perf interrupts hit in a local_irq_disable
1942 	 * (soft-masked) region, we consider them as NMIs. This is required to
1943 	 * prevent hash faults on user addresses when reading callchains (and
1944 	 * looks better from an irq tracing perspective).
1945 	 */
1946 	if (IS_ENABLED(CONFIG_PPC64) && unlikely(arch_irq_disabled_regs(regs)))
1947 		performance_monitor_exception_nmi(regs);
1948 	else
1949 		performance_monitor_exception_async(regs);
1950 }
1951 
1952 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1953 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1954 {
1955 	int changed = 0;
1956 	/*
1957 	 * Determine the cause of the debug event, clear the
1958 	 * event flags and send a trap to the handler. Torez
1959 	 */
1960 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1961 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1962 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1963 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1964 #endif
1965 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1966 			     5);
1967 		changed |= 0x01;
1968 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1969 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1970 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1971 			     6);
1972 		changed |= 0x01;
1973 	}  else if (debug_status & DBSR_IAC1) {
1974 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1975 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1976 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1977 			     1);
1978 		changed |= 0x01;
1979 	}  else if (debug_status & DBSR_IAC2) {
1980 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1981 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1982 			     2);
1983 		changed |= 0x01;
1984 	}  else if (debug_status & DBSR_IAC3) {
1985 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1986 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1987 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1988 			     3);
1989 		changed |= 0x01;
1990 	}  else if (debug_status & DBSR_IAC4) {
1991 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1992 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1993 			     4);
1994 		changed |= 0x01;
1995 	}
1996 	/*
1997 	 * At the point this routine was called, the MSR(DE) was turned off.
1998 	 * Check all other debug flags and see if that bit needs to be turned
1999 	 * back on or not.
2000 	 */
2001 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2002 			       current->thread.debug.dbcr1))
2003 		regs->msr |= MSR_DE;
2004 	else
2005 		/* Make sure the IDM flag is off */
2006 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2007 
2008 	if (changed & 0x01)
2009 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
2010 }
2011 
2012 void DebugException(struct pt_regs *regs)
2013 {
2014 	unsigned long debug_status = regs->dsisr;
2015 
2016 	current->thread.debug.dbsr = debug_status;
2017 
2018 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
2019 	 * on server, it stops on the target of the branch. In order to simulate
2020 	 * the server behaviour, we thus restart right away with a single step
2021 	 * instead of stopping here when hitting a BT
2022 	 */
2023 	if (debug_status & DBSR_BT) {
2024 		regs->msr &= ~MSR_DE;
2025 
2026 		/* Disable BT */
2027 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
2028 		/* Clear the BT event */
2029 		mtspr(SPRN_DBSR, DBSR_BT);
2030 
2031 		/* Do the single step trick only when coming from userspace */
2032 		if (user_mode(regs)) {
2033 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
2034 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
2035 			regs->msr |= MSR_DE;
2036 			return;
2037 		}
2038 
2039 		if (kprobe_post_handler(regs))
2040 			return;
2041 
2042 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
2043 			       5, SIGTRAP) == NOTIFY_STOP) {
2044 			return;
2045 		}
2046 		if (debugger_sstep(regs))
2047 			return;
2048 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
2049 		regs->msr &= ~MSR_DE;
2050 
2051 		/* Disable instruction completion */
2052 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
2053 		/* Clear the instruction completion event */
2054 		mtspr(SPRN_DBSR, DBSR_IC);
2055 
2056 		if (kprobe_post_handler(regs))
2057 			return;
2058 
2059 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2060 			       5, SIGTRAP) == NOTIFY_STOP) {
2061 			return;
2062 		}
2063 
2064 		if (debugger_sstep(regs))
2065 			return;
2066 
2067 		if (user_mode(regs)) {
2068 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2069 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2070 					       current->thread.debug.dbcr1))
2071 				regs->msr |= MSR_DE;
2072 			else
2073 				/* Make sure the IDM bit is off */
2074 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2075 		}
2076 
2077 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2078 	} else
2079 		handle_debug(regs, debug_status);
2080 }
2081 NOKPROBE_SYMBOL(DebugException);
2082 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2083 
2084 #ifdef CONFIG_ALTIVEC
2085 void altivec_assist_exception(struct pt_regs *regs)
2086 {
2087 	int err;
2088 
2089 	if (!user_mode(regs)) {
2090 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2091 		       " at %lx\n", regs->nip);
2092 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2093 	}
2094 
2095 	flush_altivec_to_thread(current);
2096 
2097 	PPC_WARN_EMULATED(altivec, regs);
2098 	err = emulate_altivec(regs);
2099 	if (err == 0) {
2100 		regs->nip += 4;		/* skip emulated instruction */
2101 		emulate_single_step(regs);
2102 		return;
2103 	}
2104 
2105 	if (err == -EFAULT) {
2106 		/* got an error reading the instruction */
2107 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2108 	} else {
2109 		/* didn't recognize the instruction */
2110 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2111 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2112 				   "in %s at %lx\n", current->comm, regs->nip);
2113 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2114 	}
2115 }
2116 #endif /* CONFIG_ALTIVEC */
2117 
2118 #ifdef CONFIG_FSL_BOOKE
2119 void CacheLockingException(struct pt_regs *regs)
2120 {
2121 	unsigned long error_code = regs->dsisr;
2122 
2123 	/* We treat cache locking instructions from the user
2124 	 * as priv ops, in the future we could try to do
2125 	 * something smarter
2126 	 */
2127 	if (error_code & (ESR_DLK|ESR_ILK))
2128 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2129 	return;
2130 }
2131 #endif /* CONFIG_FSL_BOOKE */
2132 
2133 #ifdef CONFIG_SPE
2134 void SPEFloatingPointException(struct pt_regs *regs)
2135 {
2136 	extern int do_spe_mathemu(struct pt_regs *regs);
2137 	unsigned long spefscr;
2138 	int fpexc_mode;
2139 	int code = FPE_FLTUNK;
2140 	int err;
2141 
2142 	/* We restore the interrupt state now */
2143 	if (!arch_irq_disabled_regs(regs))
2144 		local_irq_enable();
2145 
2146 	flush_spe_to_thread(current);
2147 
2148 	spefscr = current->thread.spefscr;
2149 	fpexc_mode = current->thread.fpexc_mode;
2150 
2151 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2152 		code = FPE_FLTOVF;
2153 	}
2154 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2155 		code = FPE_FLTUND;
2156 	}
2157 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2158 		code = FPE_FLTDIV;
2159 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2160 		code = FPE_FLTINV;
2161 	}
2162 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2163 		code = FPE_FLTRES;
2164 
2165 	err = do_spe_mathemu(regs);
2166 	if (err == 0) {
2167 		regs->nip += 4;		/* skip emulated instruction */
2168 		emulate_single_step(regs);
2169 		return;
2170 	}
2171 
2172 	if (err == -EFAULT) {
2173 		/* got an error reading the instruction */
2174 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2175 	} else if (err == -EINVAL) {
2176 		/* didn't recognize the instruction */
2177 		printk(KERN_ERR "unrecognized spe instruction "
2178 		       "in %s at %lx\n", current->comm, regs->nip);
2179 	} else {
2180 		_exception(SIGFPE, regs, code, regs->nip);
2181 	}
2182 
2183 	return;
2184 }
2185 
2186 void SPEFloatingPointRoundException(struct pt_regs *regs)
2187 {
2188 	extern int speround_handler(struct pt_regs *regs);
2189 	int err;
2190 
2191 	/* We restore the interrupt state now */
2192 	if (!arch_irq_disabled_regs(regs))
2193 		local_irq_enable();
2194 
2195 	preempt_disable();
2196 	if (regs->msr & MSR_SPE)
2197 		giveup_spe(current);
2198 	preempt_enable();
2199 
2200 	regs->nip -= 4;
2201 	err = speround_handler(regs);
2202 	if (err == 0) {
2203 		regs->nip += 4;		/* skip emulated instruction */
2204 		emulate_single_step(regs);
2205 		return;
2206 	}
2207 
2208 	if (err == -EFAULT) {
2209 		/* got an error reading the instruction */
2210 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2211 	} else if (err == -EINVAL) {
2212 		/* didn't recognize the instruction */
2213 		printk(KERN_ERR "unrecognized spe instruction "
2214 		       "in %s at %lx\n", current->comm, regs->nip);
2215 	} else {
2216 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2217 		return;
2218 	}
2219 }
2220 #endif
2221 
2222 /*
2223  * We enter here if we get an unrecoverable exception, that is, one
2224  * that happened at a point where the RI (recoverable interrupt) bit
2225  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2226  * we therefore lost state by taking this exception.
2227  */
2228 void unrecoverable_exception(struct pt_regs *regs)
2229 {
2230 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2231 		 regs->trap, regs->nip, regs->msr);
2232 	die("Unrecoverable exception", regs, SIGABRT);
2233 }
2234 NOKPROBE_SYMBOL(unrecoverable_exception);
2235 
2236 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2237 /*
2238  * Default handler for a Watchdog exception,
2239  * spins until a reboot occurs
2240  */
2241 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2242 {
2243 	/* Generic WatchdogHandler, implement your own */
2244 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2245 	return;
2246 }
2247 
2248 void WatchdogException(struct pt_regs *regs)
2249 {
2250 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2251 	WatchdogHandler(regs);
2252 }
2253 #endif
2254 
2255 /*
2256  * We enter here if we discover during exception entry that we are
2257  * running in supervisor mode with a userspace value in the stack pointer.
2258  */
2259 void kernel_bad_stack(struct pt_regs *regs)
2260 {
2261 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2262 	       regs->gpr[1], regs->nip);
2263 	die("Bad kernel stack pointer", regs, SIGABRT);
2264 }
2265 NOKPROBE_SYMBOL(kernel_bad_stack);
2266 
2267 void __init trap_init(void)
2268 {
2269 }
2270 
2271 
2272 #ifdef CONFIG_PPC_EMULATED_STATS
2273 
2274 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2275 
2276 struct ppc_emulated ppc_emulated = {
2277 #ifdef CONFIG_ALTIVEC
2278 	WARN_EMULATED_SETUP(altivec),
2279 #endif
2280 	WARN_EMULATED_SETUP(dcba),
2281 	WARN_EMULATED_SETUP(dcbz),
2282 	WARN_EMULATED_SETUP(fp_pair),
2283 	WARN_EMULATED_SETUP(isel),
2284 	WARN_EMULATED_SETUP(mcrxr),
2285 	WARN_EMULATED_SETUP(mfpvr),
2286 	WARN_EMULATED_SETUP(multiple),
2287 	WARN_EMULATED_SETUP(popcntb),
2288 	WARN_EMULATED_SETUP(spe),
2289 	WARN_EMULATED_SETUP(string),
2290 	WARN_EMULATED_SETUP(sync),
2291 	WARN_EMULATED_SETUP(unaligned),
2292 #ifdef CONFIG_MATH_EMULATION
2293 	WARN_EMULATED_SETUP(math),
2294 #endif
2295 #ifdef CONFIG_VSX
2296 	WARN_EMULATED_SETUP(vsx),
2297 #endif
2298 #ifdef CONFIG_PPC64
2299 	WARN_EMULATED_SETUP(mfdscr),
2300 	WARN_EMULATED_SETUP(mtdscr),
2301 	WARN_EMULATED_SETUP(lq_stq),
2302 	WARN_EMULATED_SETUP(lxvw4x),
2303 	WARN_EMULATED_SETUP(lxvh8x),
2304 	WARN_EMULATED_SETUP(lxvd2x),
2305 	WARN_EMULATED_SETUP(lxvb16x),
2306 #endif
2307 };
2308 
2309 u32 ppc_warn_emulated;
2310 
2311 void ppc_warn_emulated_print(const char *type)
2312 {
2313 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2314 			    type);
2315 }
2316 
2317 static int __init ppc_warn_emulated_init(void)
2318 {
2319 	struct dentry *dir;
2320 	unsigned int i;
2321 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2322 
2323 	dir = debugfs_create_dir("emulated_instructions",
2324 				 powerpc_debugfs_root);
2325 
2326 	debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated);
2327 
2328 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++)
2329 		debugfs_create_u32(entries[i].name, 0644, dir,
2330 				   (u32 *)&entries[i].val.counter);
2331 
2332 	return 0;
2333 }
2334 
2335 device_initcall(ppc_warn_emulated_init);
2336 
2337 #endif /* CONFIG_PPC_EMULATED_STATS */
2338