xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision f5ad1c74)
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 	/*
351 	 * Save all the pkey registers AMR/IAMR/UAMOR. Eg: Core dumps need
352 	 * to capture the content, if the task gets killed.
353 	 */
354 	thread_pkey_regs_save(&current->thread);
355 
356 	return true;
357 }
358 
359 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key)
360 {
361 	if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr))
362 		return;
363 
364 	force_sig_pkuerr((void __user *) addr, key);
365 }
366 
367 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
368 {
369 	if (!exception_common(signr, regs, code, addr))
370 		return;
371 
372 	force_sig_fault(signr, code, (void __user *)addr);
373 }
374 
375 /*
376  * The interrupt architecture has a quirk in that the HV interrupts excluding
377  * the NMIs (0x100 and 0x200) do not clear MSR[RI] at entry. The first thing
378  * that an interrupt handler must do is save off a GPR into a scratch register,
379  * and all interrupts on POWERNV (HV=1) use the HSPRG1 register as scratch.
380  * Therefore an NMI can clobber an HV interrupt's live HSPRG1 without noticing
381  * that it is non-reentrant, which leads to random data corruption.
382  *
383  * The solution is for NMI interrupts in HV mode to check if they originated
384  * from these critical HV interrupt regions. If so, then mark them not
385  * recoverable.
386  *
387  * An alternative would be for HV NMIs to use SPRG for scratch to avoid the
388  * HSPRG1 clobber, however this would cause guest SPRG to be clobbered. Linux
389  * guests should always have MSR[RI]=0 when its scratch SPRG is in use, so
390  * that would work. However any other guest OS that may have the SPRG live
391  * and MSR[RI]=1 could encounter silent corruption.
392  *
393  * Builds that do not support KVM could take this second option to increase
394  * the recoverability of NMIs.
395  */
396 void hv_nmi_check_nonrecoverable(struct pt_regs *regs)
397 {
398 #ifdef CONFIG_PPC_POWERNV
399 	unsigned long kbase = (unsigned long)_stext;
400 	unsigned long nip = regs->nip;
401 
402 	if (!(regs->msr & MSR_RI))
403 		return;
404 	if (!(regs->msr & MSR_HV))
405 		return;
406 	if (regs->msr & MSR_PR)
407 		return;
408 
409 	/*
410 	 * Now test if the interrupt has hit a range that may be using
411 	 * HSPRG1 without having RI=0 (i.e., an HSRR interrupt). The
412 	 * problem ranges all run un-relocated. Test real and virt modes
413 	 * at the same time by droping the high bit of the nip (virt mode
414 	 * entry points still have the +0x4000 offset).
415 	 */
416 	nip &= ~0xc000000000000000ULL;
417 	if ((nip >= 0x500 && nip < 0x600) || (nip >= 0x4500 && nip < 0x4600))
418 		goto nonrecoverable;
419 	if ((nip >= 0x980 && nip < 0xa00) || (nip >= 0x4980 && nip < 0x4a00))
420 		goto nonrecoverable;
421 	if ((nip >= 0xe00 && nip < 0xec0) || (nip >= 0x4e00 && nip < 0x4ec0))
422 		goto nonrecoverable;
423 	if ((nip >= 0xf80 && nip < 0xfa0) || (nip >= 0x4f80 && nip < 0x4fa0))
424 		goto nonrecoverable;
425 
426 	/* Trampoline code runs un-relocated so subtract kbase. */
427 	if (nip >= (unsigned long)(start_real_trampolines - kbase) &&
428 			nip < (unsigned long)(end_real_trampolines - kbase))
429 		goto nonrecoverable;
430 	if (nip >= (unsigned long)(start_virt_trampolines - kbase) &&
431 			nip < (unsigned long)(end_virt_trampolines - kbase))
432 		goto nonrecoverable;
433 	return;
434 
435 nonrecoverable:
436 	regs->msr &= ~MSR_RI;
437 #endif
438 }
439 
440 void system_reset_exception(struct pt_regs *regs)
441 {
442 	unsigned long hsrr0, hsrr1;
443 	bool saved_hsrrs = false;
444 	u8 ftrace_enabled = this_cpu_get_ftrace_enabled();
445 
446 	this_cpu_set_ftrace_enabled(0);
447 
448 	nmi_enter();
449 
450 	/*
451 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
452 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
453 	 * OPAL), so save them here and restore them before returning.
454 	 *
455 	 * Machine checks don't need to save HSRRs, as the real mode handler
456 	 * is careful to avoid them, and the regular handler is not delivered
457 	 * as an NMI.
458 	 */
459 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
460 		hsrr0 = mfspr(SPRN_HSRR0);
461 		hsrr1 = mfspr(SPRN_HSRR1);
462 		saved_hsrrs = true;
463 	}
464 
465 	hv_nmi_check_nonrecoverable(regs);
466 
467 	__this_cpu_inc(irq_stat.sreset_irqs);
468 
469 	/* See if any machine dependent calls */
470 	if (ppc_md.system_reset_exception) {
471 		if (ppc_md.system_reset_exception(regs))
472 			goto out;
473 	}
474 
475 	if (debugger(regs))
476 		goto out;
477 
478 	kmsg_dump(KMSG_DUMP_OOPS);
479 	/*
480 	 * A system reset is a request to dump, so we always send
481 	 * it through the crashdump code (if fadump or kdump are
482 	 * registered).
483 	 */
484 	crash_fadump(regs, "System Reset");
485 
486 	crash_kexec(regs);
487 
488 	/*
489 	 * We aren't the primary crash CPU. We need to send it
490 	 * to a holding pattern to avoid it ending up in the panic
491 	 * code.
492 	 */
493 	crash_kexec_secondary(regs);
494 
495 	/*
496 	 * No debugger or crash dump registered, print logs then
497 	 * panic.
498 	 */
499 	die("System Reset", regs, SIGABRT);
500 
501 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
502 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
503 	nmi_panic(regs, "System Reset");
504 
505 out:
506 #ifdef CONFIG_PPC_BOOK3S_64
507 	BUG_ON(get_paca()->in_nmi == 0);
508 	if (get_paca()->in_nmi > 1)
509 		die("Unrecoverable nested System Reset", regs, SIGABRT);
510 #endif
511 	/* Must die if the interrupt is not recoverable */
512 	if (!(regs->msr & MSR_RI))
513 		die("Unrecoverable System Reset", regs, SIGABRT);
514 
515 	if (saved_hsrrs) {
516 		mtspr(SPRN_HSRR0, hsrr0);
517 		mtspr(SPRN_HSRR1, hsrr1);
518 	}
519 
520 	nmi_exit();
521 
522 	this_cpu_set_ftrace_enabled(ftrace_enabled);
523 
524 	/* What should we do here? We could issue a shutdown or hard reset. */
525 }
526 
527 /*
528  * I/O accesses can cause machine checks on powermacs.
529  * Check if the NIP corresponds to the address of a sync
530  * instruction for which there is an entry in the exception
531  * table.
532  *  -- paulus.
533  */
534 static inline int check_io_access(struct pt_regs *regs)
535 {
536 #ifdef CONFIG_PPC32
537 	unsigned long msr = regs->msr;
538 	const struct exception_table_entry *entry;
539 	unsigned int *nip = (unsigned int *)regs->nip;
540 
541 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
542 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
543 		/*
544 		 * Check that it's a sync instruction, or somewhere
545 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
546 		 * As the address is in the exception table
547 		 * we should be able to read the instr there.
548 		 * For the debug message, we look at the preceding
549 		 * load or store.
550 		 */
551 		if (*nip == PPC_INST_NOP)
552 			nip -= 2;
553 		else if (*nip == PPC_INST_ISYNC)
554 			--nip;
555 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
556 			unsigned int rb;
557 
558 			--nip;
559 			rb = (*nip >> 11) & 0x1f;
560 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
561 			       (*nip & 0x100)? "OUT to": "IN from",
562 			       regs->gpr[rb] - _IO_BASE, nip);
563 			regs->msr |= MSR_RI;
564 			regs->nip = extable_fixup(entry);
565 			return 1;
566 		}
567 	}
568 #endif /* CONFIG_PPC32 */
569 	return 0;
570 }
571 
572 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
573 /* On 4xx, the reason for the machine check or program exception
574    is in the ESR. */
575 #define get_reason(regs)	((regs)->dsisr)
576 #define REASON_FP		ESR_FP
577 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
578 #define REASON_PRIVILEGED	ESR_PPR
579 #define REASON_TRAP		ESR_PTR
580 #define REASON_PREFIXED		0
581 #define REASON_BOUNDARY		0
582 
583 /* single-step stuff */
584 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
585 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
586 #define clear_br_trace(regs)	do {} while(0)
587 #else
588 /* On non-4xx, the reason for the machine check or program
589    exception is in the MSR. */
590 #define get_reason(regs)	((regs)->msr)
591 #define REASON_TM		SRR1_PROGTM
592 #define REASON_FP		SRR1_PROGFPE
593 #define REASON_ILLEGAL		SRR1_PROGILL
594 #define REASON_PRIVILEGED	SRR1_PROGPRIV
595 #define REASON_TRAP		SRR1_PROGTRAP
596 #define REASON_PREFIXED		SRR1_PREFIXED
597 #define REASON_BOUNDARY		SRR1_BOUNDARY
598 
599 #define single_stepping(regs)	((regs)->msr & MSR_SE)
600 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
601 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
602 #endif
603 
604 #define inst_length(reason)	(((reason) & REASON_PREFIXED) ? 8 : 4)
605 
606 #if defined(CONFIG_E500)
607 int machine_check_e500mc(struct pt_regs *regs)
608 {
609 	unsigned long mcsr = mfspr(SPRN_MCSR);
610 	unsigned long pvr = mfspr(SPRN_PVR);
611 	unsigned long reason = mcsr;
612 	int recoverable = 1;
613 
614 	if (reason & MCSR_LD) {
615 		recoverable = fsl_rio_mcheck_exception(regs);
616 		if (recoverable == 1)
617 			goto silent_out;
618 	}
619 
620 	printk("Machine check in kernel mode.\n");
621 	printk("Caused by (from MCSR=%lx): ", reason);
622 
623 	if (reason & MCSR_MCP)
624 		pr_cont("Machine Check Signal\n");
625 
626 	if (reason & MCSR_ICPERR) {
627 		pr_cont("Instruction Cache Parity Error\n");
628 
629 		/*
630 		 * This is recoverable by invalidating the i-cache.
631 		 */
632 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
633 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
634 			;
635 
636 		/*
637 		 * This will generally be accompanied by an instruction
638 		 * fetch error report -- only treat MCSR_IF as fatal
639 		 * if it wasn't due to an L1 parity error.
640 		 */
641 		reason &= ~MCSR_IF;
642 	}
643 
644 	if (reason & MCSR_DCPERR_MC) {
645 		pr_cont("Data Cache Parity Error\n");
646 
647 		/*
648 		 * In write shadow mode we auto-recover from the error, but it
649 		 * may still get logged and cause a machine check.  We should
650 		 * only treat the non-write shadow case as non-recoverable.
651 		 */
652 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
653 		 * is not implemented but L1 data cache always runs in write
654 		 * shadow mode. Hence on data cache parity errors HW will
655 		 * automatically invalidate the L1 Data Cache.
656 		 */
657 		if (PVR_VER(pvr) != PVR_VER_E6500) {
658 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
659 				recoverable = 0;
660 		}
661 	}
662 
663 	if (reason & MCSR_L2MMU_MHIT) {
664 		pr_cont("Hit on multiple TLB entries\n");
665 		recoverable = 0;
666 	}
667 
668 	if (reason & MCSR_NMI)
669 		pr_cont("Non-maskable interrupt\n");
670 
671 	if (reason & MCSR_IF) {
672 		pr_cont("Instruction Fetch Error Report\n");
673 		recoverable = 0;
674 	}
675 
676 	if (reason & MCSR_LD) {
677 		pr_cont("Load Error Report\n");
678 		recoverable = 0;
679 	}
680 
681 	if (reason & MCSR_ST) {
682 		pr_cont("Store Error Report\n");
683 		recoverable = 0;
684 	}
685 
686 	if (reason & MCSR_LDG) {
687 		pr_cont("Guarded Load Error Report\n");
688 		recoverable = 0;
689 	}
690 
691 	if (reason & MCSR_TLBSYNC)
692 		pr_cont("Simultaneous tlbsync operations\n");
693 
694 	if (reason & MCSR_BSL2_ERR) {
695 		pr_cont("Level 2 Cache Error\n");
696 		recoverable = 0;
697 	}
698 
699 	if (reason & MCSR_MAV) {
700 		u64 addr;
701 
702 		addr = mfspr(SPRN_MCAR);
703 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
704 
705 		pr_cont("Machine Check %s Address: %#llx\n",
706 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
707 	}
708 
709 silent_out:
710 	mtspr(SPRN_MCSR, mcsr);
711 	return mfspr(SPRN_MCSR) == 0 && recoverable;
712 }
713 
714 int machine_check_e500(struct pt_regs *regs)
715 {
716 	unsigned long reason = mfspr(SPRN_MCSR);
717 
718 	if (reason & MCSR_BUS_RBERR) {
719 		if (fsl_rio_mcheck_exception(regs))
720 			return 1;
721 		if (fsl_pci_mcheck_exception(regs))
722 			return 1;
723 	}
724 
725 	printk("Machine check in kernel mode.\n");
726 	printk("Caused by (from MCSR=%lx): ", reason);
727 
728 	if (reason & MCSR_MCP)
729 		pr_cont("Machine Check Signal\n");
730 	if (reason & MCSR_ICPERR)
731 		pr_cont("Instruction Cache Parity Error\n");
732 	if (reason & MCSR_DCP_PERR)
733 		pr_cont("Data Cache Push Parity Error\n");
734 	if (reason & MCSR_DCPERR)
735 		pr_cont("Data Cache Parity Error\n");
736 	if (reason & MCSR_BUS_IAERR)
737 		pr_cont("Bus - Instruction Address Error\n");
738 	if (reason & MCSR_BUS_RAERR)
739 		pr_cont("Bus - Read Address Error\n");
740 	if (reason & MCSR_BUS_WAERR)
741 		pr_cont("Bus - Write Address Error\n");
742 	if (reason & MCSR_BUS_IBERR)
743 		pr_cont("Bus - Instruction Data Error\n");
744 	if (reason & MCSR_BUS_RBERR)
745 		pr_cont("Bus - Read Data Bus Error\n");
746 	if (reason & MCSR_BUS_WBERR)
747 		pr_cont("Bus - Write Data Bus Error\n");
748 	if (reason & MCSR_BUS_IPERR)
749 		pr_cont("Bus - Instruction Parity Error\n");
750 	if (reason & MCSR_BUS_RPERR)
751 		pr_cont("Bus - Read Parity Error\n");
752 
753 	return 0;
754 }
755 
756 int machine_check_generic(struct pt_regs *regs)
757 {
758 	return 0;
759 }
760 #elif defined(CONFIG_E200)
761 int machine_check_e200(struct pt_regs *regs)
762 {
763 	unsigned long reason = mfspr(SPRN_MCSR);
764 
765 	printk("Machine check in kernel mode.\n");
766 	printk("Caused by (from MCSR=%lx): ", reason);
767 
768 	if (reason & MCSR_MCP)
769 		pr_cont("Machine Check Signal\n");
770 	if (reason & MCSR_CP_PERR)
771 		pr_cont("Cache Push Parity Error\n");
772 	if (reason & MCSR_CPERR)
773 		pr_cont("Cache Parity Error\n");
774 	if (reason & MCSR_EXCP_ERR)
775 		pr_cont("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
776 	if (reason & MCSR_BUS_IRERR)
777 		pr_cont("Bus - Read Bus Error on instruction fetch\n");
778 	if (reason & MCSR_BUS_DRERR)
779 		pr_cont("Bus - Read Bus Error on data load\n");
780 	if (reason & MCSR_BUS_WRERR)
781 		pr_cont("Bus - Write Bus Error on buffered store or cache line push\n");
782 
783 	return 0;
784 }
785 #elif defined(CONFIG_PPC32)
786 int machine_check_generic(struct pt_regs *regs)
787 {
788 	unsigned long reason = regs->msr;
789 
790 	printk("Machine check in kernel mode.\n");
791 	printk("Caused by (from SRR1=%lx): ", reason);
792 	switch (reason & 0x601F0000) {
793 	case 0x80000:
794 		pr_cont("Machine check signal\n");
795 		break;
796 	case 0x40000:
797 	case 0x140000:	/* 7450 MSS error and TEA */
798 		pr_cont("Transfer error ack signal\n");
799 		break;
800 	case 0x20000:
801 		pr_cont("Data parity error signal\n");
802 		break;
803 	case 0x10000:
804 		pr_cont("Address parity error signal\n");
805 		break;
806 	case 0x20000000:
807 		pr_cont("L1 Data Cache error\n");
808 		break;
809 	case 0x40000000:
810 		pr_cont("L1 Instruction Cache error\n");
811 		break;
812 	case 0x00100000:
813 		pr_cont("L2 data cache parity error\n");
814 		break;
815 	default:
816 		pr_cont("Unknown values in msr\n");
817 	}
818 	return 0;
819 }
820 #endif /* everything else */
821 
822 void machine_check_exception(struct pt_regs *regs)
823 {
824 	int recover = 0;
825 
826 	/*
827 	 * BOOK3S_64 does not call this handler as a non-maskable interrupt
828 	 * (it uses its own early real-mode handler to handle the MCE proper
829 	 * and then raises irq_work to call this handler when interrupts are
830 	 * enabled).
831 	 *
832 	 * This is silly. The BOOK3S_64 should just call a different function
833 	 * rather than expecting semantics to magically change. Something
834 	 * like 'non_nmi_machine_check_exception()', perhaps?
835 	 */
836 	const bool nmi = !IS_ENABLED(CONFIG_PPC_BOOK3S_64);
837 
838 	if (nmi) nmi_enter();
839 
840 	__this_cpu_inc(irq_stat.mce_exceptions);
841 
842 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
843 
844 	/* See if any machine dependent calls. In theory, we would want
845 	 * to call the CPU first, and call the ppc_md. one if the CPU
846 	 * one returns a positive number. However there is existing code
847 	 * that assumes the board gets a first chance, so let's keep it
848 	 * that way for now and fix things later. --BenH.
849 	 */
850 	if (ppc_md.machine_check_exception)
851 		recover = ppc_md.machine_check_exception(regs);
852 	else if (cur_cpu_spec->machine_check)
853 		recover = cur_cpu_spec->machine_check(regs);
854 
855 	if (recover > 0)
856 		goto bail;
857 
858 	if (debugger_fault_handler(regs))
859 		goto bail;
860 
861 	if (check_io_access(regs))
862 		goto bail;
863 
864 	if (nmi) nmi_exit();
865 
866 	die("Machine check", regs, SIGBUS);
867 
868 	/* Must die if the interrupt is not recoverable */
869 	if (!(regs->msr & MSR_RI))
870 		die("Unrecoverable Machine check", regs, SIGBUS);
871 
872 	return;
873 
874 bail:
875 	if (nmi) nmi_exit();
876 }
877 
878 void SMIException(struct pt_regs *regs)
879 {
880 	die("System Management Interrupt", regs, SIGABRT);
881 }
882 
883 #ifdef CONFIG_VSX
884 static void p9_hmi_special_emu(struct pt_regs *regs)
885 {
886 	unsigned int ra, rb, t, i, sel, instr, rc;
887 	const void __user *addr;
888 	u8 vbuf[16] __aligned(16), *vdst;
889 	unsigned long ea, msr, msr_mask;
890 	bool swap;
891 
892 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
893 		return;
894 
895 	/*
896 	 * lxvb16x	opcode: 0x7c0006d8
897 	 * lxvd2x	opcode: 0x7c000698
898 	 * lxvh8x	opcode: 0x7c000658
899 	 * lxvw4x	opcode: 0x7c000618
900 	 */
901 	if ((instr & 0xfc00073e) != 0x7c000618) {
902 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
903 			 " instr=%08x\n",
904 			 smp_processor_id(), current->comm, current->pid,
905 			 regs->nip, instr);
906 		return;
907 	}
908 
909 	/* Grab vector registers into the task struct */
910 	msr = regs->msr; /* Grab msr before we flush the bits */
911 	flush_vsx_to_thread(current);
912 	enable_kernel_altivec();
913 
914 	/*
915 	 * Is userspace running with a different endian (this is rare but
916 	 * not impossible)
917 	 */
918 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
919 
920 	/* Decode the instruction */
921 	ra = (instr >> 16) & 0x1f;
922 	rb = (instr >> 11) & 0x1f;
923 	t = (instr >> 21) & 0x1f;
924 	if (instr & 1)
925 		vdst = (u8 *)&current->thread.vr_state.vr[t];
926 	else
927 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
928 
929 	/* Grab the vector address */
930 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
931 	if (is_32bit_task())
932 		ea &= 0xfffffffful;
933 	addr = (__force const void __user *)ea;
934 
935 	/* Check it */
936 	if (!access_ok(addr, 16)) {
937 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
938 			 " instr=%08x addr=%016lx\n",
939 			 smp_processor_id(), current->comm, current->pid,
940 			 regs->nip, instr, (unsigned long)addr);
941 		return;
942 	}
943 
944 	/* Read the vector */
945 	rc = 0;
946 	if ((unsigned long)addr & 0xfUL)
947 		/* unaligned case */
948 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
949 	else
950 		__get_user_atomic_128_aligned(vbuf, addr, rc);
951 	if (rc) {
952 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
953 			 " instr=%08x addr=%016lx\n",
954 			 smp_processor_id(), current->comm, current->pid,
955 			 regs->nip, instr, (unsigned long)addr);
956 		return;
957 	}
958 
959 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
960 		 " instr=%08x addr=%016lx\n",
961 		 smp_processor_id(), current->comm, current->pid, regs->nip,
962 		 instr, (unsigned long) addr);
963 
964 	/* Grab instruction "selector" */
965 	sel = (instr >> 6) & 3;
966 
967 	/*
968 	 * Check to make sure the facility is actually enabled. This
969 	 * could happen if we get a false positive hit.
970 	 *
971 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
972 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
973 	 */
974 	msr_mask = MSR_VSX;
975 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
976 		msr_mask = MSR_VEC;
977 	if (!(msr & msr_mask)) {
978 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
979 			 " instr=%08x msr:%016lx\n",
980 			 smp_processor_id(), current->comm, current->pid,
981 			 regs->nip, instr, msr);
982 		return;
983 	}
984 
985 	/* Do logging here before we modify sel based on endian */
986 	switch (sel) {
987 	case 0:	/* lxvw4x */
988 		PPC_WARN_EMULATED(lxvw4x, regs);
989 		break;
990 	case 1: /* lxvh8x */
991 		PPC_WARN_EMULATED(lxvh8x, regs);
992 		break;
993 	case 2: /* lxvd2x */
994 		PPC_WARN_EMULATED(lxvd2x, regs);
995 		break;
996 	case 3: /* lxvb16x */
997 		PPC_WARN_EMULATED(lxvb16x, regs);
998 		break;
999 	}
1000 
1001 #ifdef __LITTLE_ENDIAN__
1002 	/*
1003 	 * An LE kernel stores the vector in the task struct as an LE
1004 	 * byte array (effectively swapping both the components and
1005 	 * the content of the components). Those instructions expect
1006 	 * the components to remain in ascending address order, so we
1007 	 * swap them back.
1008 	 *
1009 	 * If we are running a BE user space, the expectation is that
1010 	 * of a simple memcpy, so forcing the emulation to look like
1011 	 * a lxvb16x should do the trick.
1012 	 */
1013 	if (swap)
1014 		sel = 3;
1015 
1016 	switch (sel) {
1017 	case 0:	/* lxvw4x */
1018 		for (i = 0; i < 4; i++)
1019 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1020 		break;
1021 	case 1: /* lxvh8x */
1022 		for (i = 0; i < 8; i++)
1023 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1024 		break;
1025 	case 2: /* lxvd2x */
1026 		for (i = 0; i < 2; i++)
1027 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1028 		break;
1029 	case 3: /* lxvb16x */
1030 		for (i = 0; i < 16; i++)
1031 			vdst[i] = vbuf[15-i];
1032 		break;
1033 	}
1034 #else /* __LITTLE_ENDIAN__ */
1035 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1036 	if (!swap)
1037 		sel = 3;
1038 
1039 	/* Otherwise, we need to swap the content of the components */
1040 	switch (sel) {
1041 	case 0:	/* lxvw4x */
1042 		for (i = 0; i < 4; i++)
1043 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1044 		break;
1045 	case 1: /* lxvh8x */
1046 		for (i = 0; i < 8; i++)
1047 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1048 		break;
1049 	case 2: /* lxvd2x */
1050 		for (i = 0; i < 2; i++)
1051 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1052 		break;
1053 	case 3: /* lxvb16x */
1054 		memcpy(vdst, vbuf, 16);
1055 		break;
1056 	}
1057 #endif /* !__LITTLE_ENDIAN__ */
1058 
1059 	/* Go to next instruction */
1060 	regs->nip += 4;
1061 }
1062 #endif /* CONFIG_VSX */
1063 
1064 void handle_hmi_exception(struct pt_regs *regs)
1065 {
1066 	struct pt_regs *old_regs;
1067 
1068 	old_regs = set_irq_regs(regs);
1069 	irq_enter();
1070 
1071 #ifdef CONFIG_VSX
1072 	/* Real mode flagged P9 special emu is needed */
1073 	if (local_paca->hmi_p9_special_emu) {
1074 		local_paca->hmi_p9_special_emu = 0;
1075 
1076 		/*
1077 		 * We don't want to take page faults while doing the
1078 		 * emulation, we just replay the instruction if necessary.
1079 		 */
1080 		pagefault_disable();
1081 		p9_hmi_special_emu(regs);
1082 		pagefault_enable();
1083 	}
1084 #endif /* CONFIG_VSX */
1085 
1086 	if (ppc_md.handle_hmi_exception)
1087 		ppc_md.handle_hmi_exception(regs);
1088 
1089 	irq_exit();
1090 	set_irq_regs(old_regs);
1091 }
1092 
1093 void unknown_exception(struct pt_regs *regs)
1094 {
1095 	enum ctx_state prev_state = exception_enter();
1096 
1097 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1098 	       regs->nip, regs->msr, regs->trap);
1099 
1100 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1101 
1102 	exception_exit(prev_state);
1103 }
1104 
1105 void instruction_breakpoint_exception(struct pt_regs *regs)
1106 {
1107 	enum ctx_state prev_state = exception_enter();
1108 
1109 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1110 					5, SIGTRAP) == NOTIFY_STOP)
1111 		goto bail;
1112 	if (debugger_iabr_match(regs))
1113 		goto bail;
1114 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1115 
1116 bail:
1117 	exception_exit(prev_state);
1118 }
1119 
1120 void RunModeException(struct pt_regs *regs)
1121 {
1122 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1123 }
1124 
1125 void single_step_exception(struct pt_regs *regs)
1126 {
1127 	enum ctx_state prev_state = exception_enter();
1128 
1129 	clear_single_step(regs);
1130 	clear_br_trace(regs);
1131 
1132 	if (kprobe_post_handler(regs))
1133 		return;
1134 
1135 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1136 					5, SIGTRAP) == NOTIFY_STOP)
1137 		goto bail;
1138 	if (debugger_sstep(regs))
1139 		goto bail;
1140 
1141 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1142 
1143 bail:
1144 	exception_exit(prev_state);
1145 }
1146 NOKPROBE_SYMBOL(single_step_exception);
1147 
1148 /*
1149  * After we have successfully emulated an instruction, we have to
1150  * check if the instruction was being single-stepped, and if so,
1151  * pretend we got a single-step exception.  This was pointed out
1152  * by Kumar Gala.  -- paulus
1153  */
1154 static void emulate_single_step(struct pt_regs *regs)
1155 {
1156 	if (single_stepping(regs))
1157 		single_step_exception(regs);
1158 }
1159 
1160 static inline int __parse_fpscr(unsigned long fpscr)
1161 {
1162 	int ret = FPE_FLTUNK;
1163 
1164 	/* Invalid operation */
1165 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1166 		ret = FPE_FLTINV;
1167 
1168 	/* Overflow */
1169 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1170 		ret = FPE_FLTOVF;
1171 
1172 	/* Underflow */
1173 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1174 		ret = FPE_FLTUND;
1175 
1176 	/* Divide by zero */
1177 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1178 		ret = FPE_FLTDIV;
1179 
1180 	/* Inexact result */
1181 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1182 		ret = FPE_FLTRES;
1183 
1184 	return ret;
1185 }
1186 
1187 static void parse_fpe(struct pt_regs *regs)
1188 {
1189 	int code = 0;
1190 
1191 	flush_fp_to_thread(current);
1192 
1193 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1194 
1195 	_exception(SIGFPE, regs, code, regs->nip);
1196 }
1197 
1198 /*
1199  * Illegal instruction emulation support.  Originally written to
1200  * provide the PVR to user applications using the mfspr rd, PVR.
1201  * Return non-zero if we can't emulate, or -EFAULT if the associated
1202  * memory access caused an access fault.  Return zero on success.
1203  *
1204  * There are a couple of ways to do this, either "decode" the instruction
1205  * or directly match lots of bits.  In this case, matching lots of
1206  * bits is faster and easier.
1207  *
1208  */
1209 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1210 {
1211 	u8 rT = (instword >> 21) & 0x1f;
1212 	u8 rA = (instword >> 16) & 0x1f;
1213 	u8 NB_RB = (instword >> 11) & 0x1f;
1214 	u32 num_bytes;
1215 	unsigned long EA;
1216 	int pos = 0;
1217 
1218 	/* Early out if we are an invalid form of lswx */
1219 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1220 		if ((rT == rA) || (rT == NB_RB))
1221 			return -EINVAL;
1222 
1223 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1224 
1225 	switch (instword & PPC_INST_STRING_MASK) {
1226 		case PPC_INST_LSWX:
1227 		case PPC_INST_STSWX:
1228 			EA += NB_RB;
1229 			num_bytes = regs->xer & 0x7f;
1230 			break;
1231 		case PPC_INST_LSWI:
1232 		case PPC_INST_STSWI:
1233 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1234 			break;
1235 		default:
1236 			return -EINVAL;
1237 	}
1238 
1239 	while (num_bytes != 0)
1240 	{
1241 		u8 val;
1242 		u32 shift = 8 * (3 - (pos & 0x3));
1243 
1244 		/* if process is 32-bit, clear upper 32 bits of EA */
1245 		if ((regs->msr & MSR_64BIT) == 0)
1246 			EA &= 0xFFFFFFFF;
1247 
1248 		switch ((instword & PPC_INST_STRING_MASK)) {
1249 			case PPC_INST_LSWX:
1250 			case PPC_INST_LSWI:
1251 				if (get_user(val, (u8 __user *)EA))
1252 					return -EFAULT;
1253 				/* first time updating this reg,
1254 				 * zero it out */
1255 				if (pos == 0)
1256 					regs->gpr[rT] = 0;
1257 				regs->gpr[rT] |= val << shift;
1258 				break;
1259 			case PPC_INST_STSWI:
1260 			case PPC_INST_STSWX:
1261 				val = regs->gpr[rT] >> shift;
1262 				if (put_user(val, (u8 __user *)EA))
1263 					return -EFAULT;
1264 				break;
1265 		}
1266 		/* move EA to next address */
1267 		EA += 1;
1268 		num_bytes--;
1269 
1270 		/* manage our position within the register */
1271 		if (++pos == 4) {
1272 			pos = 0;
1273 			if (++rT == 32)
1274 				rT = 0;
1275 		}
1276 	}
1277 
1278 	return 0;
1279 }
1280 
1281 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1282 {
1283 	u32 ra,rs;
1284 	unsigned long tmp;
1285 
1286 	ra = (instword >> 16) & 0x1f;
1287 	rs = (instword >> 21) & 0x1f;
1288 
1289 	tmp = regs->gpr[rs];
1290 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1291 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1292 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1293 	regs->gpr[ra] = tmp;
1294 
1295 	return 0;
1296 }
1297 
1298 static int emulate_isel(struct pt_regs *regs, u32 instword)
1299 {
1300 	u8 rT = (instword >> 21) & 0x1f;
1301 	u8 rA = (instword >> 16) & 0x1f;
1302 	u8 rB = (instword >> 11) & 0x1f;
1303 	u8 BC = (instword >> 6) & 0x1f;
1304 	u8 bit;
1305 	unsigned long tmp;
1306 
1307 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1308 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1309 
1310 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1311 
1312 	return 0;
1313 }
1314 
1315 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1316 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1317 {
1318         /* If we're emulating a load/store in an active transaction, we cannot
1319          * emulate it as the kernel operates in transaction suspended context.
1320          * We need to abort the transaction.  This creates a persistent TM
1321          * abort so tell the user what caused it with a new code.
1322 	 */
1323 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1324 		tm_enable();
1325 		tm_abort(cause);
1326 		return true;
1327 	}
1328 	return false;
1329 }
1330 #else
1331 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1332 {
1333 	return false;
1334 }
1335 #endif
1336 
1337 static int emulate_instruction(struct pt_regs *regs)
1338 {
1339 	u32 instword;
1340 	u32 rd;
1341 
1342 	if (!user_mode(regs))
1343 		return -EINVAL;
1344 	CHECK_FULL_REGS(regs);
1345 
1346 	if (get_user(instword, (u32 __user *)(regs->nip)))
1347 		return -EFAULT;
1348 
1349 	/* Emulate the mfspr rD, PVR. */
1350 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1351 		PPC_WARN_EMULATED(mfpvr, regs);
1352 		rd = (instword >> 21) & 0x1f;
1353 		regs->gpr[rd] = mfspr(SPRN_PVR);
1354 		return 0;
1355 	}
1356 
1357 	/* Emulating the dcba insn is just a no-op.  */
1358 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1359 		PPC_WARN_EMULATED(dcba, regs);
1360 		return 0;
1361 	}
1362 
1363 	/* Emulate the mcrxr insn.  */
1364 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1365 		int shift = (instword >> 21) & 0x1c;
1366 		unsigned long msk = 0xf0000000UL >> shift;
1367 
1368 		PPC_WARN_EMULATED(mcrxr, regs);
1369 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1370 		regs->xer &= ~0xf0000000UL;
1371 		return 0;
1372 	}
1373 
1374 	/* Emulate load/store string insn. */
1375 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1376 		if (tm_abort_check(regs,
1377 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1378 			return -EINVAL;
1379 		PPC_WARN_EMULATED(string, regs);
1380 		return emulate_string_inst(regs, instword);
1381 	}
1382 
1383 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1384 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1385 		PPC_WARN_EMULATED(popcntb, regs);
1386 		return emulate_popcntb_inst(regs, instword);
1387 	}
1388 
1389 	/* Emulate isel (Integer Select) instruction */
1390 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1391 		PPC_WARN_EMULATED(isel, regs);
1392 		return emulate_isel(regs, instword);
1393 	}
1394 
1395 	/* Emulate sync instruction variants */
1396 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1397 		PPC_WARN_EMULATED(sync, regs);
1398 		asm volatile("sync");
1399 		return 0;
1400 	}
1401 
1402 #ifdef CONFIG_PPC64
1403 	/* Emulate the mfspr rD, DSCR. */
1404 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1405 		PPC_INST_MFSPR_DSCR_USER) ||
1406 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1407 		PPC_INST_MFSPR_DSCR)) &&
1408 			cpu_has_feature(CPU_FTR_DSCR)) {
1409 		PPC_WARN_EMULATED(mfdscr, regs);
1410 		rd = (instword >> 21) & 0x1f;
1411 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1412 		return 0;
1413 	}
1414 	/* Emulate the mtspr DSCR, rD. */
1415 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1416 		PPC_INST_MTSPR_DSCR_USER) ||
1417 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1418 		PPC_INST_MTSPR_DSCR)) &&
1419 			cpu_has_feature(CPU_FTR_DSCR)) {
1420 		PPC_WARN_EMULATED(mtdscr, regs);
1421 		rd = (instword >> 21) & 0x1f;
1422 		current->thread.dscr = regs->gpr[rd];
1423 		current->thread.dscr_inherit = 1;
1424 		mtspr(SPRN_DSCR, current->thread.dscr);
1425 		return 0;
1426 	}
1427 #endif
1428 
1429 	return -EINVAL;
1430 }
1431 
1432 int is_valid_bugaddr(unsigned long addr)
1433 {
1434 	return is_kernel_addr(addr);
1435 }
1436 
1437 #ifdef CONFIG_MATH_EMULATION
1438 static int emulate_math(struct pt_regs *regs)
1439 {
1440 	int ret;
1441 	extern int do_mathemu(struct pt_regs *regs);
1442 
1443 	ret = do_mathemu(regs);
1444 	if (ret >= 0)
1445 		PPC_WARN_EMULATED(math, regs);
1446 
1447 	switch (ret) {
1448 	case 0:
1449 		emulate_single_step(regs);
1450 		return 0;
1451 	case 1: {
1452 			int code = 0;
1453 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1454 			_exception(SIGFPE, regs, code, regs->nip);
1455 			return 0;
1456 		}
1457 	case -EFAULT:
1458 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1459 		return 0;
1460 	}
1461 
1462 	return -1;
1463 }
1464 #else
1465 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1466 #endif
1467 
1468 void program_check_exception(struct pt_regs *regs)
1469 {
1470 	enum ctx_state prev_state = exception_enter();
1471 	unsigned int reason = get_reason(regs);
1472 
1473 	/* We can now get here via a FP Unavailable exception if the core
1474 	 * has no FPU, in that case the reason flags will be 0 */
1475 
1476 	if (reason & REASON_FP) {
1477 		/* IEEE FP exception */
1478 		parse_fpe(regs);
1479 		goto bail;
1480 	}
1481 	if (reason & REASON_TRAP) {
1482 		unsigned long bugaddr;
1483 		/* Debugger is first in line to stop recursive faults in
1484 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1485 		if (debugger_bpt(regs))
1486 			goto bail;
1487 
1488 		if (kprobe_handler(regs))
1489 			goto bail;
1490 
1491 		/* trap exception */
1492 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1493 				== NOTIFY_STOP)
1494 			goto bail;
1495 
1496 		bugaddr = regs->nip;
1497 		/*
1498 		 * Fixup bugaddr for BUG_ON() in real mode
1499 		 */
1500 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1501 			bugaddr += PAGE_OFFSET;
1502 
1503 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1504 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1505 			regs->nip += 4;
1506 			goto bail;
1507 		}
1508 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1509 		goto bail;
1510 	}
1511 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1512 	if (reason & REASON_TM) {
1513 		/* This is a TM "Bad Thing Exception" program check.
1514 		 * This occurs when:
1515 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1516 		 *    transition in TM states.
1517 		 * -  A trechkpt is attempted when transactional.
1518 		 * -  A treclaim is attempted when non transactional.
1519 		 * -  A tend is illegally attempted.
1520 		 * -  writing a TM SPR when transactional.
1521 		 *
1522 		 * If usermode caused this, it's done something illegal and
1523 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1524 		 * operand to distinguish from the instruction just being bad
1525 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1526 		 * illegal /placement/ of a valid instruction.
1527 		 */
1528 		if (user_mode(regs)) {
1529 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1530 			goto bail;
1531 		} else {
1532 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1533 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1534 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1535 			die("Unrecoverable exception", regs, SIGABRT);
1536 		}
1537 	}
1538 #endif
1539 
1540 	/*
1541 	 * If we took the program check in the kernel skip down to sending a
1542 	 * SIGILL. The subsequent cases all relate to emulating instructions
1543 	 * which we should only do for userspace. We also do not want to enable
1544 	 * interrupts for kernel faults because that might lead to further
1545 	 * faults, and loose the context of the original exception.
1546 	 */
1547 	if (!user_mode(regs))
1548 		goto sigill;
1549 
1550 	/* We restore the interrupt state now */
1551 	if (!arch_irq_disabled_regs(regs))
1552 		local_irq_enable();
1553 
1554 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1555 	 * but there seems to be a hardware bug on the 405GP (RevD)
1556 	 * that means ESR is sometimes set incorrectly - either to
1557 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1558 	 * hardware people - not sure if it can happen on any illegal
1559 	 * instruction or only on FP instructions, whether there is a
1560 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1561 	 */
1562 	if (!emulate_math(regs))
1563 		goto bail;
1564 
1565 	/* Try to emulate it if we should. */
1566 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1567 		switch (emulate_instruction(regs)) {
1568 		case 0:
1569 			regs->nip += 4;
1570 			emulate_single_step(regs);
1571 			goto bail;
1572 		case -EFAULT:
1573 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1574 			goto bail;
1575 		}
1576 	}
1577 
1578 sigill:
1579 	if (reason & REASON_PRIVILEGED)
1580 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1581 	else
1582 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1583 
1584 bail:
1585 	exception_exit(prev_state);
1586 }
1587 NOKPROBE_SYMBOL(program_check_exception);
1588 
1589 /*
1590  * This occurs when running in hypervisor mode on POWER6 or later
1591  * and an illegal instruction is encountered.
1592  */
1593 void emulation_assist_interrupt(struct pt_regs *regs)
1594 {
1595 	regs->msr |= REASON_ILLEGAL;
1596 	program_check_exception(regs);
1597 }
1598 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1599 
1600 void alignment_exception(struct pt_regs *regs)
1601 {
1602 	enum ctx_state prev_state = exception_enter();
1603 	int sig, code, fixed = 0;
1604 	unsigned long  reason;
1605 
1606 	/* We restore the interrupt state now */
1607 	if (!arch_irq_disabled_regs(regs))
1608 		local_irq_enable();
1609 
1610 	reason = get_reason(regs);
1611 
1612 	if (reason & REASON_BOUNDARY) {
1613 		sig = SIGBUS;
1614 		code = BUS_ADRALN;
1615 		goto bad;
1616 	}
1617 
1618 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1619 		goto bail;
1620 
1621 	/* we don't implement logging of alignment exceptions */
1622 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1623 		fixed = fix_alignment(regs);
1624 
1625 	if (fixed == 1) {
1626 		/* skip over emulated instruction */
1627 		regs->nip += inst_length(reason);
1628 		emulate_single_step(regs);
1629 		goto bail;
1630 	}
1631 
1632 	/* Operand address was bad */
1633 	if (fixed == -EFAULT) {
1634 		sig = SIGSEGV;
1635 		code = SEGV_ACCERR;
1636 	} else {
1637 		sig = SIGBUS;
1638 		code = BUS_ADRALN;
1639 	}
1640 bad:
1641 	if (user_mode(regs))
1642 		_exception(sig, regs, code, regs->dar);
1643 	else
1644 		bad_page_fault(regs, regs->dar, sig);
1645 
1646 bail:
1647 	exception_exit(prev_state);
1648 }
1649 
1650 void StackOverflow(struct pt_regs *regs)
1651 {
1652 	pr_crit("Kernel stack overflow in process %s[%d], r1=%lx\n",
1653 		current->comm, task_pid_nr(current), regs->gpr[1]);
1654 	debugger(regs);
1655 	show_regs(regs);
1656 	panic("kernel stack overflow");
1657 }
1658 
1659 void stack_overflow_exception(struct pt_regs *regs)
1660 {
1661 	enum ctx_state prev_state = exception_enter();
1662 
1663 	die("Kernel stack overflow", regs, SIGSEGV);
1664 
1665 	exception_exit(prev_state);
1666 }
1667 
1668 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1669 {
1670 	enum ctx_state prev_state = exception_enter();
1671 
1672 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1673 			  "%lx at %lx\n", regs->trap, regs->nip);
1674 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1675 
1676 	exception_exit(prev_state);
1677 }
1678 
1679 void altivec_unavailable_exception(struct pt_regs *regs)
1680 {
1681 	enum ctx_state prev_state = exception_enter();
1682 
1683 	if (user_mode(regs)) {
1684 		/* A user program has executed an altivec instruction,
1685 		   but this kernel doesn't support altivec. */
1686 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1687 		goto bail;
1688 	}
1689 
1690 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1691 			"%lx at %lx\n", regs->trap, regs->nip);
1692 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1693 
1694 bail:
1695 	exception_exit(prev_state);
1696 }
1697 
1698 void vsx_unavailable_exception(struct pt_regs *regs)
1699 {
1700 	if (user_mode(regs)) {
1701 		/* A user program has executed an vsx instruction,
1702 		   but this kernel doesn't support vsx. */
1703 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1704 		return;
1705 	}
1706 
1707 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1708 			"%lx at %lx\n", regs->trap, regs->nip);
1709 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1710 }
1711 
1712 #ifdef CONFIG_PPC64
1713 static void tm_unavailable(struct pt_regs *regs)
1714 {
1715 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1716 	if (user_mode(regs)) {
1717 		current->thread.load_tm++;
1718 		regs->msr |= MSR_TM;
1719 		tm_enable();
1720 		tm_restore_sprs(&current->thread);
1721 		return;
1722 	}
1723 #endif
1724 	pr_emerg("Unrecoverable TM Unavailable Exception "
1725 			"%lx at %lx\n", regs->trap, regs->nip);
1726 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1727 }
1728 
1729 void facility_unavailable_exception(struct pt_regs *regs)
1730 {
1731 	static char *facility_strings[] = {
1732 		[FSCR_FP_LG] = "FPU",
1733 		[FSCR_VECVSX_LG] = "VMX/VSX",
1734 		[FSCR_DSCR_LG] = "DSCR",
1735 		[FSCR_PM_LG] = "PMU SPRs",
1736 		[FSCR_BHRB_LG] = "BHRB",
1737 		[FSCR_TM_LG] = "TM",
1738 		[FSCR_EBB_LG] = "EBB",
1739 		[FSCR_TAR_LG] = "TAR",
1740 		[FSCR_MSGP_LG] = "MSGP",
1741 		[FSCR_SCV_LG] = "SCV",
1742 		[FSCR_PREFIX_LG] = "PREFIX",
1743 	};
1744 	char *facility = "unknown";
1745 	u64 value;
1746 	u32 instword, rd;
1747 	u8 status;
1748 	bool hv;
1749 
1750 	hv = (TRAP(regs) == 0xf80);
1751 	if (hv)
1752 		value = mfspr(SPRN_HFSCR);
1753 	else
1754 		value = mfspr(SPRN_FSCR);
1755 
1756 	status = value >> 56;
1757 	if ((hv || status >= 2) &&
1758 	    (status < ARRAY_SIZE(facility_strings)) &&
1759 	    facility_strings[status])
1760 		facility = facility_strings[status];
1761 
1762 	/* We should not have taken this interrupt in kernel */
1763 	if (!user_mode(regs)) {
1764 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1765 			 facility, status, regs->nip);
1766 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1767 	}
1768 
1769 	/* We restore the interrupt state now */
1770 	if (!arch_irq_disabled_regs(regs))
1771 		local_irq_enable();
1772 
1773 	if (status == FSCR_DSCR_LG) {
1774 		/*
1775 		 * User is accessing the DSCR register using the problem
1776 		 * state only SPR number (0x03) either through a mfspr or
1777 		 * a mtspr instruction. If it is a write attempt through
1778 		 * a mtspr, then we set the inherit bit. This also allows
1779 		 * the user to write or read the register directly in the
1780 		 * future by setting via the FSCR DSCR bit. But in case it
1781 		 * is a read DSCR attempt through a mfspr instruction, we
1782 		 * just emulate the instruction instead. This code path will
1783 		 * always emulate all the mfspr instructions till the user
1784 		 * has attempted at least one mtspr instruction. This way it
1785 		 * preserves the same behaviour when the user is accessing
1786 		 * the DSCR through privilege level only SPR number (0x11)
1787 		 * which is emulated through illegal instruction exception.
1788 		 * We always leave HFSCR DSCR set.
1789 		 */
1790 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1791 			pr_err("Failed to fetch the user instruction\n");
1792 			return;
1793 		}
1794 
1795 		/* Write into DSCR (mtspr 0x03, RS) */
1796 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1797 				== PPC_INST_MTSPR_DSCR_USER) {
1798 			rd = (instword >> 21) & 0x1f;
1799 			current->thread.dscr = regs->gpr[rd];
1800 			current->thread.dscr_inherit = 1;
1801 			current->thread.fscr |= FSCR_DSCR;
1802 			mtspr(SPRN_FSCR, current->thread.fscr);
1803 		}
1804 
1805 		/* Read from DSCR (mfspr RT, 0x03) */
1806 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1807 				== PPC_INST_MFSPR_DSCR_USER) {
1808 			if (emulate_instruction(regs)) {
1809 				pr_err("DSCR based mfspr emulation failed\n");
1810 				return;
1811 			}
1812 			regs->nip += 4;
1813 			emulate_single_step(regs);
1814 		}
1815 		return;
1816 	}
1817 
1818 	if (status == FSCR_TM_LG) {
1819 		/*
1820 		 * If we're here then the hardware is TM aware because it
1821 		 * generated an exception with FSRM_TM set.
1822 		 *
1823 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1824 		 * told us not to do TM, or the kernel is not built with TM
1825 		 * support.
1826 		 *
1827 		 * If both of those things are true, then userspace can spam the
1828 		 * console by triggering the printk() below just by continually
1829 		 * doing tbegin (or any TM instruction). So in that case just
1830 		 * send the process a SIGILL immediately.
1831 		 */
1832 		if (!cpu_has_feature(CPU_FTR_TM))
1833 			goto out;
1834 
1835 		tm_unavailable(regs);
1836 		return;
1837 	}
1838 
1839 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1840 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1841 
1842 out:
1843 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1844 }
1845 #endif
1846 
1847 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1848 
1849 void fp_unavailable_tm(struct pt_regs *regs)
1850 {
1851 	/* Note:  This does not handle any kind of FP laziness. */
1852 
1853 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1854 		 regs->nip, regs->msr);
1855 
1856         /* We can only have got here if the task started using FP after
1857          * beginning the transaction.  So, the transactional regs are just a
1858          * copy of the checkpointed ones.  But, we still need to recheckpoint
1859          * as we're enabling FP for the process; it will return, abort the
1860          * transaction, and probably retry but now with FP enabled.  So the
1861          * checkpointed FP registers need to be loaded.
1862 	 */
1863 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1864 
1865 	/*
1866 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1867 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1868 	 *
1869 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1870 	 * recheckpoint.
1871 	 */
1872 
1873 	/* Enable FP for the task: */
1874 	current->thread.load_fp = 1;
1875 
1876 	/*
1877 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1878 	 */
1879 	tm_recheckpoint(&current->thread);
1880 }
1881 
1882 void altivec_unavailable_tm(struct pt_regs *regs)
1883 {
1884 	/* See the comments in fp_unavailable_tm().  This function operates
1885 	 * the same way.
1886 	 */
1887 
1888 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1889 		 "MSR=%lx\n",
1890 		 regs->nip, regs->msr);
1891 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1892 	current->thread.load_vec = 1;
1893 	tm_recheckpoint(&current->thread);
1894 	current->thread.used_vr = 1;
1895 }
1896 
1897 void vsx_unavailable_tm(struct pt_regs *regs)
1898 {
1899 	/* See the comments in fp_unavailable_tm().  This works similarly,
1900 	 * though we're loading both FP and VEC registers in here.
1901 	 *
1902 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1903 	 * regs.  Either way, set MSR_VSX.
1904 	 */
1905 
1906 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1907 		 "MSR=%lx\n",
1908 		 regs->nip, regs->msr);
1909 
1910 	current->thread.used_vsr = 1;
1911 
1912 	/* This reclaims FP and/or VR regs if they're already enabled */
1913 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1914 
1915 	current->thread.load_vec = 1;
1916 	current->thread.load_fp = 1;
1917 
1918 	tm_recheckpoint(&current->thread);
1919 }
1920 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1921 
1922 void performance_monitor_exception(struct pt_regs *regs)
1923 {
1924 	__this_cpu_inc(irq_stat.pmu_irqs);
1925 
1926 	perf_irq(regs);
1927 }
1928 
1929 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1930 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1931 {
1932 	int changed = 0;
1933 	/*
1934 	 * Determine the cause of the debug event, clear the
1935 	 * event flags and send a trap to the handler. Torez
1936 	 */
1937 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1938 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1939 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1940 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1941 #endif
1942 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1943 			     5);
1944 		changed |= 0x01;
1945 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1946 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1947 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1948 			     6);
1949 		changed |= 0x01;
1950 	}  else if (debug_status & DBSR_IAC1) {
1951 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1952 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1953 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1954 			     1);
1955 		changed |= 0x01;
1956 	}  else if (debug_status & DBSR_IAC2) {
1957 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1958 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1959 			     2);
1960 		changed |= 0x01;
1961 	}  else if (debug_status & DBSR_IAC3) {
1962 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1963 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1964 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1965 			     3);
1966 		changed |= 0x01;
1967 	}  else if (debug_status & DBSR_IAC4) {
1968 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1969 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1970 			     4);
1971 		changed |= 0x01;
1972 	}
1973 	/*
1974 	 * At the point this routine was called, the MSR(DE) was turned off.
1975 	 * Check all other debug flags and see if that bit needs to be turned
1976 	 * back on or not.
1977 	 */
1978 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1979 			       current->thread.debug.dbcr1))
1980 		regs->msr |= MSR_DE;
1981 	else
1982 		/* Make sure the IDM flag is off */
1983 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1984 
1985 	if (changed & 0x01)
1986 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1987 }
1988 
1989 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1990 {
1991 	current->thread.debug.dbsr = debug_status;
1992 
1993 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1994 	 * on server, it stops on the target of the branch. In order to simulate
1995 	 * the server behaviour, we thus restart right away with a single step
1996 	 * instead of stopping here when hitting a BT
1997 	 */
1998 	if (debug_status & DBSR_BT) {
1999 		regs->msr &= ~MSR_DE;
2000 
2001 		/* Disable BT */
2002 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
2003 		/* Clear the BT event */
2004 		mtspr(SPRN_DBSR, DBSR_BT);
2005 
2006 		/* Do the single step trick only when coming from userspace */
2007 		if (user_mode(regs)) {
2008 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
2009 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
2010 			regs->msr |= MSR_DE;
2011 			return;
2012 		}
2013 
2014 		if (kprobe_post_handler(regs))
2015 			return;
2016 
2017 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
2018 			       5, SIGTRAP) == NOTIFY_STOP) {
2019 			return;
2020 		}
2021 		if (debugger_sstep(regs))
2022 			return;
2023 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
2024 		regs->msr &= ~MSR_DE;
2025 
2026 		/* Disable instruction completion */
2027 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
2028 		/* Clear the instruction completion event */
2029 		mtspr(SPRN_DBSR, DBSR_IC);
2030 
2031 		if (kprobe_post_handler(regs))
2032 			return;
2033 
2034 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2035 			       5, SIGTRAP) == NOTIFY_STOP) {
2036 			return;
2037 		}
2038 
2039 		if (debugger_sstep(regs))
2040 			return;
2041 
2042 		if (user_mode(regs)) {
2043 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2044 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2045 					       current->thread.debug.dbcr1))
2046 				regs->msr |= MSR_DE;
2047 			else
2048 				/* Make sure the IDM bit is off */
2049 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2050 		}
2051 
2052 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2053 	} else
2054 		handle_debug(regs, debug_status);
2055 }
2056 NOKPROBE_SYMBOL(DebugException);
2057 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2058 
2059 #ifdef CONFIG_ALTIVEC
2060 void altivec_assist_exception(struct pt_regs *regs)
2061 {
2062 	int err;
2063 
2064 	if (!user_mode(regs)) {
2065 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2066 		       " at %lx\n", regs->nip);
2067 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2068 	}
2069 
2070 	flush_altivec_to_thread(current);
2071 
2072 	PPC_WARN_EMULATED(altivec, regs);
2073 	err = emulate_altivec(regs);
2074 	if (err == 0) {
2075 		regs->nip += 4;		/* skip emulated instruction */
2076 		emulate_single_step(regs);
2077 		return;
2078 	}
2079 
2080 	if (err == -EFAULT) {
2081 		/* got an error reading the instruction */
2082 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2083 	} else {
2084 		/* didn't recognize the instruction */
2085 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2086 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2087 				   "in %s at %lx\n", current->comm, regs->nip);
2088 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2089 	}
2090 }
2091 #endif /* CONFIG_ALTIVEC */
2092 
2093 #ifdef CONFIG_FSL_BOOKE
2094 void CacheLockingException(struct pt_regs *regs, unsigned long address,
2095 			   unsigned long error_code)
2096 {
2097 	/* We treat cache locking instructions from the user
2098 	 * as priv ops, in the future we could try to do
2099 	 * something smarter
2100 	 */
2101 	if (error_code & (ESR_DLK|ESR_ILK))
2102 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2103 	return;
2104 }
2105 #endif /* CONFIG_FSL_BOOKE */
2106 
2107 #ifdef CONFIG_SPE
2108 void SPEFloatingPointException(struct pt_regs *regs)
2109 {
2110 	extern int do_spe_mathemu(struct pt_regs *regs);
2111 	unsigned long spefscr;
2112 	int fpexc_mode;
2113 	int code = FPE_FLTUNK;
2114 	int err;
2115 
2116 	/* We restore the interrupt state now */
2117 	if (!arch_irq_disabled_regs(regs))
2118 		local_irq_enable();
2119 
2120 	flush_spe_to_thread(current);
2121 
2122 	spefscr = current->thread.spefscr;
2123 	fpexc_mode = current->thread.fpexc_mode;
2124 
2125 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2126 		code = FPE_FLTOVF;
2127 	}
2128 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2129 		code = FPE_FLTUND;
2130 	}
2131 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2132 		code = FPE_FLTDIV;
2133 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2134 		code = FPE_FLTINV;
2135 	}
2136 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2137 		code = FPE_FLTRES;
2138 
2139 	err = do_spe_mathemu(regs);
2140 	if (err == 0) {
2141 		regs->nip += 4;		/* skip emulated instruction */
2142 		emulate_single_step(regs);
2143 		return;
2144 	}
2145 
2146 	if (err == -EFAULT) {
2147 		/* got an error reading the instruction */
2148 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2149 	} else if (err == -EINVAL) {
2150 		/* didn't recognize the instruction */
2151 		printk(KERN_ERR "unrecognized spe instruction "
2152 		       "in %s at %lx\n", current->comm, regs->nip);
2153 	} else {
2154 		_exception(SIGFPE, regs, code, regs->nip);
2155 	}
2156 
2157 	return;
2158 }
2159 
2160 void SPEFloatingPointRoundException(struct pt_regs *regs)
2161 {
2162 	extern int speround_handler(struct pt_regs *regs);
2163 	int err;
2164 
2165 	/* We restore the interrupt state now */
2166 	if (!arch_irq_disabled_regs(regs))
2167 		local_irq_enable();
2168 
2169 	preempt_disable();
2170 	if (regs->msr & MSR_SPE)
2171 		giveup_spe(current);
2172 	preempt_enable();
2173 
2174 	regs->nip -= 4;
2175 	err = speround_handler(regs);
2176 	if (err == 0) {
2177 		regs->nip += 4;		/* skip emulated instruction */
2178 		emulate_single_step(regs);
2179 		return;
2180 	}
2181 
2182 	if (err == -EFAULT) {
2183 		/* got an error reading the instruction */
2184 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2185 	} else if (err == -EINVAL) {
2186 		/* didn't recognize the instruction */
2187 		printk(KERN_ERR "unrecognized spe instruction "
2188 		       "in %s at %lx\n", current->comm, regs->nip);
2189 	} else {
2190 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2191 		return;
2192 	}
2193 }
2194 #endif
2195 
2196 /*
2197  * We enter here if we get an unrecoverable exception, that is, one
2198  * that happened at a point where the RI (recoverable interrupt) bit
2199  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2200  * we therefore lost state by taking this exception.
2201  */
2202 void unrecoverable_exception(struct pt_regs *regs)
2203 {
2204 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2205 		 regs->trap, regs->nip, regs->msr);
2206 	die("Unrecoverable exception", regs, SIGABRT);
2207 }
2208 NOKPROBE_SYMBOL(unrecoverable_exception);
2209 
2210 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2211 /*
2212  * Default handler for a Watchdog exception,
2213  * spins until a reboot occurs
2214  */
2215 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2216 {
2217 	/* Generic WatchdogHandler, implement your own */
2218 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2219 	return;
2220 }
2221 
2222 void WatchdogException(struct pt_regs *regs)
2223 {
2224 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2225 	WatchdogHandler(regs);
2226 }
2227 #endif
2228 
2229 /*
2230  * We enter here if we discover during exception entry that we are
2231  * running in supervisor mode with a userspace value in the stack pointer.
2232  */
2233 void kernel_bad_stack(struct pt_regs *regs)
2234 {
2235 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2236 	       regs->gpr[1], regs->nip);
2237 	die("Bad kernel stack pointer", regs, SIGABRT);
2238 }
2239 NOKPROBE_SYMBOL(kernel_bad_stack);
2240 
2241 void __init trap_init(void)
2242 {
2243 }
2244 
2245 
2246 #ifdef CONFIG_PPC_EMULATED_STATS
2247 
2248 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2249 
2250 struct ppc_emulated ppc_emulated = {
2251 #ifdef CONFIG_ALTIVEC
2252 	WARN_EMULATED_SETUP(altivec),
2253 #endif
2254 	WARN_EMULATED_SETUP(dcba),
2255 	WARN_EMULATED_SETUP(dcbz),
2256 	WARN_EMULATED_SETUP(fp_pair),
2257 	WARN_EMULATED_SETUP(isel),
2258 	WARN_EMULATED_SETUP(mcrxr),
2259 	WARN_EMULATED_SETUP(mfpvr),
2260 	WARN_EMULATED_SETUP(multiple),
2261 	WARN_EMULATED_SETUP(popcntb),
2262 	WARN_EMULATED_SETUP(spe),
2263 	WARN_EMULATED_SETUP(string),
2264 	WARN_EMULATED_SETUP(sync),
2265 	WARN_EMULATED_SETUP(unaligned),
2266 #ifdef CONFIG_MATH_EMULATION
2267 	WARN_EMULATED_SETUP(math),
2268 #endif
2269 #ifdef CONFIG_VSX
2270 	WARN_EMULATED_SETUP(vsx),
2271 #endif
2272 #ifdef CONFIG_PPC64
2273 	WARN_EMULATED_SETUP(mfdscr),
2274 	WARN_EMULATED_SETUP(mtdscr),
2275 	WARN_EMULATED_SETUP(lq_stq),
2276 	WARN_EMULATED_SETUP(lxvw4x),
2277 	WARN_EMULATED_SETUP(lxvh8x),
2278 	WARN_EMULATED_SETUP(lxvd2x),
2279 	WARN_EMULATED_SETUP(lxvb16x),
2280 #endif
2281 };
2282 
2283 u32 ppc_warn_emulated;
2284 
2285 void ppc_warn_emulated_print(const char *type)
2286 {
2287 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2288 			    type);
2289 }
2290 
2291 static int __init ppc_warn_emulated_init(void)
2292 {
2293 	struct dentry *dir;
2294 	unsigned int i;
2295 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2296 
2297 	dir = debugfs_create_dir("emulated_instructions",
2298 				 powerpc_debugfs_root);
2299 
2300 	debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated);
2301 
2302 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++)
2303 		debugfs_create_u32(entries[i].name, 0644, dir,
2304 				   (u32 *)&entries[i].val.counter);
2305 
2306 	return 0;
2307 }
2308 
2309 device_initcall(ppc_warn_emulated_init);
2310 
2311 #endif /* CONFIG_PPC_EMULATED_STATS */
2312