xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision cd238eff)
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 <asm/pgtable.h>
43 #include <linux/uaccess.h>
44 #include <asm/debugfs.h>
45 #include <asm/io.h>
46 #include <asm/machdep.h>
47 #include <asm/rtas.h>
48 #include <asm/pmc.h>
49 #include <asm/reg.h>
50 #ifdef CONFIG_PMAC_BACKLIGHT
51 #include <asm/backlight.h>
52 #endif
53 #ifdef CONFIG_PPC64
54 #include <asm/firmware.h>
55 #include <asm/processor.h>
56 #include <asm/tm.h>
57 #endif
58 #include <asm/kexec.h>
59 #include <asm/ppc-opcode.h>
60 #include <asm/rio.h>
61 #include <asm/fadump.h>
62 #include <asm/switch_to.h>
63 #include <asm/tm.h>
64 #include <asm/debug.h>
65 #include <asm/asm-prototypes.h>
66 #include <asm/hmi.h>
67 #include <sysdev/fsl_pci.h>
68 #include <asm/kprobes.h>
69 #include <asm/stacktrace.h>
70 #include <asm/nmi.h>
71 
72 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
73 int (*__debugger)(struct pt_regs *regs) __read_mostly;
74 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
75 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
79 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
80 
81 EXPORT_SYMBOL(__debugger);
82 EXPORT_SYMBOL(__debugger_ipi);
83 EXPORT_SYMBOL(__debugger_bpt);
84 EXPORT_SYMBOL(__debugger_sstep);
85 EXPORT_SYMBOL(__debugger_iabr_match);
86 EXPORT_SYMBOL(__debugger_break_match);
87 EXPORT_SYMBOL(__debugger_fault_handler);
88 #endif
89 
90 /* Transactional Memory trap debug */
91 #ifdef TM_DEBUG_SW
92 #define TM_DEBUG(x...) printk(KERN_INFO x)
93 #else
94 #define TM_DEBUG(x...) do { } while(0)
95 #endif
96 
97 static const char *signame(int signr)
98 {
99 	switch (signr) {
100 	case SIGBUS:	return "bus error";
101 	case SIGFPE:	return "floating point exception";
102 	case SIGILL:	return "illegal instruction";
103 	case SIGSEGV:	return "segfault";
104 	case SIGTRAP:	return "unhandled trap";
105 	}
106 
107 	return "unknown signal";
108 }
109 
110 /*
111  * Trap & Exception support
112  */
113 
114 #ifdef CONFIG_PMAC_BACKLIGHT
115 static void pmac_backlight_unblank(void)
116 {
117 	mutex_lock(&pmac_backlight_mutex);
118 	if (pmac_backlight) {
119 		struct backlight_properties *props;
120 
121 		props = &pmac_backlight->props;
122 		props->brightness = props->max_brightness;
123 		props->power = FB_BLANK_UNBLANK;
124 		backlight_update_status(pmac_backlight);
125 	}
126 	mutex_unlock(&pmac_backlight_mutex);
127 }
128 #else
129 static inline void pmac_backlight_unblank(void) { }
130 #endif
131 
132 /*
133  * If oops/die is expected to crash the machine, return true here.
134  *
135  * This should not be expected to be 100% accurate, there may be
136  * notifiers registered or other unexpected conditions that may bring
137  * down the kernel. Or if the current process in the kernel is holding
138  * locks or has other critical state, the kernel may become effectively
139  * unusable anyway.
140  */
141 bool die_will_crash(void)
142 {
143 	if (should_fadump_crash())
144 		return true;
145 	if (kexec_should_crash(current))
146 		return true;
147 	if (in_interrupt() || panic_on_oops ||
148 			!current->pid || is_global_init(current))
149 		return true;
150 
151 	return false;
152 }
153 
154 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
155 static int die_owner = -1;
156 static unsigned int die_nest_count;
157 static int die_counter;
158 
159 extern void panic_flush_kmsg_start(void)
160 {
161 	/*
162 	 * These are mostly taken from kernel/panic.c, but tries to do
163 	 * relatively minimal work. Don't use delay functions (TB may
164 	 * be broken), don't crash dump (need to set a firmware log),
165 	 * don't run notifiers. We do want to get some information to
166 	 * Linux console.
167 	 */
168 	console_verbose();
169 	bust_spinlocks(1);
170 }
171 
172 extern void panic_flush_kmsg_end(void)
173 {
174 	printk_safe_flush_on_panic();
175 	kmsg_dump(KMSG_DUMP_PANIC);
176 	bust_spinlocks(0);
177 	debug_locks_off();
178 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
179 }
180 
181 static unsigned long oops_begin(struct pt_regs *regs)
182 {
183 	int cpu;
184 	unsigned long flags;
185 
186 	oops_enter();
187 
188 	/* racy, but better than risking deadlock. */
189 	raw_local_irq_save(flags);
190 	cpu = smp_processor_id();
191 	if (!arch_spin_trylock(&die_lock)) {
192 		if (cpu == die_owner)
193 			/* nested oops. should stop eventually */;
194 		else
195 			arch_spin_lock(&die_lock);
196 	}
197 	die_nest_count++;
198 	die_owner = cpu;
199 	console_verbose();
200 	bust_spinlocks(1);
201 	if (machine_is(powermac))
202 		pmac_backlight_unblank();
203 	return flags;
204 }
205 NOKPROBE_SYMBOL(oops_begin);
206 
207 static void oops_end(unsigned long flags, struct pt_regs *regs,
208 			       int signr)
209 {
210 	bust_spinlocks(0);
211 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
212 	die_nest_count--;
213 	oops_exit();
214 	printk("\n");
215 	if (!die_nest_count) {
216 		/* Nest count reaches zero, release the lock. */
217 		die_owner = -1;
218 		arch_spin_unlock(&die_lock);
219 	}
220 	raw_local_irq_restore(flags);
221 
222 	/*
223 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
224 	 */
225 	if (TRAP(regs) == 0x100)
226 		return;
227 
228 	crash_fadump(regs, "die oops");
229 
230 	if (kexec_should_crash(current))
231 		crash_kexec(regs);
232 
233 	if (!signr)
234 		return;
235 
236 	/*
237 	 * While our oops output is serialised by a spinlock, output
238 	 * from panic() called below can race and corrupt it. If we
239 	 * know we are going to panic, delay for 1 second so we have a
240 	 * chance to get clean backtraces from all CPUs that are oopsing.
241 	 */
242 	if (in_interrupt() || panic_on_oops || !current->pid ||
243 	    is_global_init(current)) {
244 		mdelay(MSEC_PER_SEC);
245 	}
246 
247 	if (panic_on_oops)
248 		panic("Fatal exception");
249 	do_exit(signr);
250 }
251 NOKPROBE_SYMBOL(oops_end);
252 
253 static int __die(const char *str, struct pt_regs *regs, long err)
254 {
255 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
256 
257 	printk("%s PAGE_SIZE=%luK%s%s%s%s%s%s%s %s\n",
258 	       IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE",
259 	       PAGE_SIZE / 1024,
260 	       early_radix_enabled() ? " MMU=Radix" : "",
261 	       early_mmu_has_feature(MMU_FTR_HPTE_TABLE) ? " MMU=Hash" : "",
262 	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
263 	       IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
264 	       IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "",
265 	       debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
266 	       IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "",
267 	       ppc_md.name ? ppc_md.name : "");
268 
269 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
270 		return 1;
271 
272 	print_modules();
273 	show_regs(regs);
274 
275 	return 0;
276 }
277 NOKPROBE_SYMBOL(__die);
278 
279 void die(const char *str, struct pt_regs *regs, long err)
280 {
281 	unsigned long flags;
282 
283 	/*
284 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
285 	 */
286 	if (TRAP(regs) != 0x100) {
287 		if (debugger(regs))
288 			return;
289 	}
290 
291 	flags = oops_begin(regs);
292 	if (__die(str, regs, err))
293 		err = 0;
294 	oops_end(flags, regs, err);
295 }
296 NOKPROBE_SYMBOL(die);
297 
298 void user_single_step_report(struct pt_regs *regs)
299 {
300 	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip, current);
301 }
302 
303 static void show_signal_msg(int signr, struct pt_regs *regs, int code,
304 			    unsigned long addr)
305 {
306 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
307 				      DEFAULT_RATELIMIT_BURST);
308 
309 	if (!show_unhandled_signals)
310 		return;
311 
312 	if (!unhandled_signal(current, signr))
313 		return;
314 
315 	if (!__ratelimit(&rs))
316 		return;
317 
318 	pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x",
319 		current->comm, current->pid, signame(signr), signr,
320 		addr, regs->nip, regs->link, code);
321 
322 	print_vma_addr(KERN_CONT " in ", regs->nip);
323 
324 	pr_cont("\n");
325 
326 	show_user_instructions(regs);
327 }
328 
329 static bool exception_common(int signr, struct pt_regs *regs, int code,
330 			      unsigned long addr)
331 {
332 	if (!user_mode(regs)) {
333 		die("Exception in kernel mode", regs, signr);
334 		return false;
335 	}
336 
337 	show_signal_msg(signr, regs, code, addr);
338 
339 	if (arch_irqs_disabled() && !arch_irq_disabled_regs(regs))
340 		local_irq_enable();
341 
342 	current->thread.trap_nr = code;
343 
344 	/*
345 	 * Save all the pkey registers AMR/IAMR/UAMOR. Eg: Core dumps need
346 	 * to capture the content, if the task gets killed.
347 	 */
348 	thread_pkey_regs_save(&current->thread);
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, current);
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 nested = in_nmi();
438 	bool saved_hsrrs = false;
439 
440 	/*
441 	 * Avoid crashes in case of nested NMI exceptions. Recoverability
442 	 * is determined by RI and in_nmi
443 	 */
444 	if (!nested)
445 		nmi_enter();
446 
447 	/*
448 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
449 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
450 	 * OPAL), so save them here and restore them before returning.
451 	 *
452 	 * Machine checks don't need to save HSRRs, as the real mode handler
453 	 * is careful to avoid them, and the regular handler is not delivered
454 	 * as an NMI.
455 	 */
456 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
457 		hsrr0 = mfspr(SPRN_HSRR0);
458 		hsrr1 = mfspr(SPRN_HSRR1);
459 		saved_hsrrs = true;
460 	}
461 
462 	hv_nmi_check_nonrecoverable(regs);
463 
464 	__this_cpu_inc(irq_stat.sreset_irqs);
465 
466 	/* See if any machine dependent calls */
467 	if (ppc_md.system_reset_exception) {
468 		if (ppc_md.system_reset_exception(regs))
469 			goto out;
470 	}
471 
472 	if (debugger(regs))
473 		goto out;
474 
475 	/*
476 	 * A system reset is a request to dump, so we always send
477 	 * it through the crashdump code (if fadump or kdump are
478 	 * registered).
479 	 */
480 	crash_fadump(regs, "System Reset");
481 
482 	crash_kexec(regs);
483 
484 	/*
485 	 * We aren't the primary crash CPU. We need to send it
486 	 * to a holding pattern to avoid it ending up in the panic
487 	 * code.
488 	 */
489 	crash_kexec_secondary(regs);
490 
491 	/*
492 	 * No debugger or crash dump registered, print logs then
493 	 * panic.
494 	 */
495 	die("System Reset", regs, SIGABRT);
496 
497 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
498 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
499 	nmi_panic(regs, "System Reset");
500 
501 out:
502 #ifdef CONFIG_PPC_BOOK3S_64
503 	BUG_ON(get_paca()->in_nmi == 0);
504 	if (get_paca()->in_nmi > 1)
505 		nmi_panic(regs, "Unrecoverable nested System Reset");
506 #endif
507 	/* Must die if the interrupt is not recoverable */
508 	if (!(regs->msr & MSR_RI))
509 		nmi_panic(regs, "Unrecoverable System Reset");
510 
511 	if (saved_hsrrs) {
512 		mtspr(SPRN_HSRR0, hsrr0);
513 		mtspr(SPRN_HSRR1, hsrr1);
514 	}
515 
516 	if (!nested)
517 		nmi_exit();
518 
519 	/* What should we do here? We could issue a shutdown or hard reset. */
520 }
521 
522 /*
523  * I/O accesses can cause machine checks on powermacs.
524  * Check if the NIP corresponds to the address of a sync
525  * instruction for which there is an entry in the exception
526  * table.
527  * Note that the 601 only takes a machine check on TEA
528  * (transfer error ack) signal assertion, and does not
529  * set any of the top 16 bits of SRR1.
530  *  -- paulus.
531  */
532 static inline int check_io_access(struct pt_regs *regs)
533 {
534 #ifdef CONFIG_PPC32
535 	unsigned long msr = regs->msr;
536 	const struct exception_table_entry *entry;
537 	unsigned int *nip = (unsigned int *)regs->nip;
538 
539 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
540 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
541 		/*
542 		 * Check that it's a sync instruction, or somewhere
543 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
544 		 * As the address is in the exception table
545 		 * we should be able to read the instr there.
546 		 * For the debug message, we look at the preceding
547 		 * load or store.
548 		 */
549 		if (*nip == PPC_INST_NOP)
550 			nip -= 2;
551 		else if (*nip == PPC_INST_ISYNC)
552 			--nip;
553 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
554 			unsigned int rb;
555 
556 			--nip;
557 			rb = (*nip >> 11) & 0x1f;
558 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
559 			       (*nip & 0x100)? "OUT to": "IN from",
560 			       regs->gpr[rb] - _IO_BASE, nip);
561 			regs->msr |= MSR_RI;
562 			regs->nip = extable_fixup(entry);
563 			return 1;
564 		}
565 	}
566 #endif /* CONFIG_PPC32 */
567 	return 0;
568 }
569 
570 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
571 /* On 4xx, the reason for the machine check or program exception
572    is in the ESR. */
573 #define get_reason(regs)	((regs)->dsisr)
574 #define REASON_FP		ESR_FP
575 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
576 #define REASON_PRIVILEGED	ESR_PPR
577 #define REASON_TRAP		ESR_PTR
578 
579 /* single-step stuff */
580 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
581 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
582 #define clear_br_trace(regs)	do {} while(0)
583 #else
584 /* On non-4xx, the reason for the machine check or program
585    exception is in the MSR. */
586 #define get_reason(regs)	((regs)->msr)
587 #define REASON_TM		SRR1_PROGTM
588 #define REASON_FP		SRR1_PROGFPE
589 #define REASON_ILLEGAL		SRR1_PROGILL
590 #define REASON_PRIVILEGED	SRR1_PROGPRIV
591 #define REASON_TRAP		SRR1_PROGTRAP
592 
593 #define single_stepping(regs)	((regs)->msr & MSR_SE)
594 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
595 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
596 #endif
597 
598 #if defined(CONFIG_E500)
599 int machine_check_e500mc(struct pt_regs *regs)
600 {
601 	unsigned long mcsr = mfspr(SPRN_MCSR);
602 	unsigned long pvr = mfspr(SPRN_PVR);
603 	unsigned long reason = mcsr;
604 	int recoverable = 1;
605 
606 	if (reason & MCSR_LD) {
607 		recoverable = fsl_rio_mcheck_exception(regs);
608 		if (recoverable == 1)
609 			goto silent_out;
610 	}
611 
612 	printk("Machine check in kernel mode.\n");
613 	printk("Caused by (from MCSR=%lx): ", reason);
614 
615 	if (reason & MCSR_MCP)
616 		pr_cont("Machine Check Signal\n");
617 
618 	if (reason & MCSR_ICPERR) {
619 		pr_cont("Instruction Cache Parity Error\n");
620 
621 		/*
622 		 * This is recoverable by invalidating the i-cache.
623 		 */
624 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
625 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
626 			;
627 
628 		/*
629 		 * This will generally be accompanied by an instruction
630 		 * fetch error report -- only treat MCSR_IF as fatal
631 		 * if it wasn't due to an L1 parity error.
632 		 */
633 		reason &= ~MCSR_IF;
634 	}
635 
636 	if (reason & MCSR_DCPERR_MC) {
637 		pr_cont("Data Cache Parity Error\n");
638 
639 		/*
640 		 * In write shadow mode we auto-recover from the error, but it
641 		 * may still get logged and cause a machine check.  We should
642 		 * only treat the non-write shadow case as non-recoverable.
643 		 */
644 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
645 		 * is not implemented but L1 data cache always runs in write
646 		 * shadow mode. Hence on data cache parity errors HW will
647 		 * automatically invalidate the L1 Data Cache.
648 		 */
649 		if (PVR_VER(pvr) != PVR_VER_E6500) {
650 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
651 				recoverable = 0;
652 		}
653 	}
654 
655 	if (reason & MCSR_L2MMU_MHIT) {
656 		pr_cont("Hit on multiple TLB entries\n");
657 		recoverable = 0;
658 	}
659 
660 	if (reason & MCSR_NMI)
661 		pr_cont("Non-maskable interrupt\n");
662 
663 	if (reason & MCSR_IF) {
664 		pr_cont("Instruction Fetch Error Report\n");
665 		recoverable = 0;
666 	}
667 
668 	if (reason & MCSR_LD) {
669 		pr_cont("Load Error Report\n");
670 		recoverable = 0;
671 	}
672 
673 	if (reason & MCSR_ST) {
674 		pr_cont("Store Error Report\n");
675 		recoverable = 0;
676 	}
677 
678 	if (reason & MCSR_LDG) {
679 		pr_cont("Guarded Load Error Report\n");
680 		recoverable = 0;
681 	}
682 
683 	if (reason & MCSR_TLBSYNC)
684 		pr_cont("Simultaneous tlbsync operations\n");
685 
686 	if (reason & MCSR_BSL2_ERR) {
687 		pr_cont("Level 2 Cache Error\n");
688 		recoverable = 0;
689 	}
690 
691 	if (reason & MCSR_MAV) {
692 		u64 addr;
693 
694 		addr = mfspr(SPRN_MCAR);
695 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
696 
697 		pr_cont("Machine Check %s Address: %#llx\n",
698 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
699 	}
700 
701 silent_out:
702 	mtspr(SPRN_MCSR, mcsr);
703 	return mfspr(SPRN_MCSR) == 0 && recoverable;
704 }
705 
706 int machine_check_e500(struct pt_regs *regs)
707 {
708 	unsigned long reason = mfspr(SPRN_MCSR);
709 
710 	if (reason & MCSR_BUS_RBERR) {
711 		if (fsl_rio_mcheck_exception(regs))
712 			return 1;
713 		if (fsl_pci_mcheck_exception(regs))
714 			return 1;
715 	}
716 
717 	printk("Machine check in kernel mode.\n");
718 	printk("Caused by (from MCSR=%lx): ", reason);
719 
720 	if (reason & MCSR_MCP)
721 		pr_cont("Machine Check Signal\n");
722 	if (reason & MCSR_ICPERR)
723 		pr_cont("Instruction Cache Parity Error\n");
724 	if (reason & MCSR_DCP_PERR)
725 		pr_cont("Data Cache Push Parity Error\n");
726 	if (reason & MCSR_DCPERR)
727 		pr_cont("Data Cache Parity Error\n");
728 	if (reason & MCSR_BUS_IAERR)
729 		pr_cont("Bus - Instruction Address Error\n");
730 	if (reason & MCSR_BUS_RAERR)
731 		pr_cont("Bus - Read Address Error\n");
732 	if (reason & MCSR_BUS_WAERR)
733 		pr_cont("Bus - Write Address Error\n");
734 	if (reason & MCSR_BUS_IBERR)
735 		pr_cont("Bus - Instruction Data Error\n");
736 	if (reason & MCSR_BUS_RBERR)
737 		pr_cont("Bus - Read Data Bus Error\n");
738 	if (reason & MCSR_BUS_WBERR)
739 		pr_cont("Bus - Write Data Bus Error\n");
740 	if (reason & MCSR_BUS_IPERR)
741 		pr_cont("Bus - Instruction Parity Error\n");
742 	if (reason & MCSR_BUS_RPERR)
743 		pr_cont("Bus - Read Parity Error\n");
744 
745 	return 0;
746 }
747 
748 int machine_check_generic(struct pt_regs *regs)
749 {
750 	return 0;
751 }
752 #elif defined(CONFIG_E200)
753 int machine_check_e200(struct pt_regs *regs)
754 {
755 	unsigned long reason = mfspr(SPRN_MCSR);
756 
757 	printk("Machine check in kernel mode.\n");
758 	printk("Caused by (from MCSR=%lx): ", reason);
759 
760 	if (reason & MCSR_MCP)
761 		pr_cont("Machine Check Signal\n");
762 	if (reason & MCSR_CP_PERR)
763 		pr_cont("Cache Push Parity Error\n");
764 	if (reason & MCSR_CPERR)
765 		pr_cont("Cache Parity Error\n");
766 	if (reason & MCSR_EXCP_ERR)
767 		pr_cont("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
768 	if (reason & MCSR_BUS_IRERR)
769 		pr_cont("Bus - Read Bus Error on instruction fetch\n");
770 	if (reason & MCSR_BUS_DRERR)
771 		pr_cont("Bus - Read Bus Error on data load\n");
772 	if (reason & MCSR_BUS_WRERR)
773 		pr_cont("Bus - Write Bus Error on buffered store or cache line push\n");
774 
775 	return 0;
776 }
777 #elif defined(CONFIG_PPC32)
778 int machine_check_generic(struct pt_regs *regs)
779 {
780 	unsigned long reason = regs->msr;
781 
782 	printk("Machine check in kernel mode.\n");
783 	printk("Caused by (from SRR1=%lx): ", reason);
784 	switch (reason & 0x601F0000) {
785 	case 0x80000:
786 		pr_cont("Machine check signal\n");
787 		break;
788 	case 0:		/* for 601 */
789 	case 0x40000:
790 	case 0x140000:	/* 7450 MSS error and TEA */
791 		pr_cont("Transfer error ack signal\n");
792 		break;
793 	case 0x20000:
794 		pr_cont("Data parity error signal\n");
795 		break;
796 	case 0x10000:
797 		pr_cont("Address parity error signal\n");
798 		break;
799 	case 0x20000000:
800 		pr_cont("L1 Data Cache error\n");
801 		break;
802 	case 0x40000000:
803 		pr_cont("L1 Instruction Cache error\n");
804 		break;
805 	case 0x00100000:
806 		pr_cont("L2 data cache parity error\n");
807 		break;
808 	default:
809 		pr_cont("Unknown values in msr\n");
810 	}
811 	return 0;
812 }
813 #endif /* everything else */
814 
815 void machine_check_exception(struct pt_regs *regs)
816 {
817 	int recover = 0;
818 	bool nested = in_nmi();
819 	if (!nested)
820 		nmi_enter();
821 
822 	__this_cpu_inc(irq_stat.mce_exceptions);
823 
824 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
825 
826 	/* See if any machine dependent calls. In theory, we would want
827 	 * to call the CPU first, and call the ppc_md. one if the CPU
828 	 * one returns a positive number. However there is existing code
829 	 * that assumes the board gets a first chance, so let's keep it
830 	 * that way for now and fix things later. --BenH.
831 	 */
832 	if (ppc_md.machine_check_exception)
833 		recover = ppc_md.machine_check_exception(regs);
834 	else if (cur_cpu_spec->machine_check)
835 		recover = cur_cpu_spec->machine_check(regs);
836 
837 	if (recover > 0)
838 		goto bail;
839 
840 	if (debugger_fault_handler(regs))
841 		goto bail;
842 
843 	if (check_io_access(regs))
844 		goto bail;
845 
846 	if (!nested)
847 		nmi_exit();
848 
849 	die("Machine check", regs, SIGBUS);
850 
851 	/* Must die if the interrupt is not recoverable */
852 	if (!(regs->msr & MSR_RI))
853 		nmi_panic(regs, "Unrecoverable Machine check");
854 
855 	return;
856 
857 bail:
858 	if (!nested)
859 		nmi_exit();
860 }
861 
862 void SMIException(struct pt_regs *regs)
863 {
864 	die("System Management Interrupt", regs, SIGABRT);
865 }
866 
867 #ifdef CONFIG_VSX
868 static void p9_hmi_special_emu(struct pt_regs *regs)
869 {
870 	unsigned int ra, rb, t, i, sel, instr, rc;
871 	const void __user *addr;
872 	u8 vbuf[16], *vdst;
873 	unsigned long ea, msr, msr_mask;
874 	bool swap;
875 
876 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
877 		return;
878 
879 	/*
880 	 * lxvb16x	opcode: 0x7c0006d8
881 	 * lxvd2x	opcode: 0x7c000698
882 	 * lxvh8x	opcode: 0x7c000658
883 	 * lxvw4x	opcode: 0x7c000618
884 	 */
885 	if ((instr & 0xfc00073e) != 0x7c000618) {
886 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
887 			 " instr=%08x\n",
888 			 smp_processor_id(), current->comm, current->pid,
889 			 regs->nip, instr);
890 		return;
891 	}
892 
893 	/* Grab vector registers into the task struct */
894 	msr = regs->msr; /* Grab msr before we flush the bits */
895 	flush_vsx_to_thread(current);
896 	enable_kernel_altivec();
897 
898 	/*
899 	 * Is userspace running with a different endian (this is rare but
900 	 * not impossible)
901 	 */
902 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
903 
904 	/* Decode the instruction */
905 	ra = (instr >> 16) & 0x1f;
906 	rb = (instr >> 11) & 0x1f;
907 	t = (instr >> 21) & 0x1f;
908 	if (instr & 1)
909 		vdst = (u8 *)&current->thread.vr_state.vr[t];
910 	else
911 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
912 
913 	/* Grab the vector address */
914 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
915 	if (is_32bit_task())
916 		ea &= 0xfffffffful;
917 	addr = (__force const void __user *)ea;
918 
919 	/* Check it */
920 	if (!access_ok(addr, 16)) {
921 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
922 			 " instr=%08x addr=%016lx\n",
923 			 smp_processor_id(), current->comm, current->pid,
924 			 regs->nip, instr, (unsigned long)addr);
925 		return;
926 	}
927 
928 	/* Read the vector */
929 	rc = 0;
930 	if ((unsigned long)addr & 0xfUL)
931 		/* unaligned case */
932 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
933 	else
934 		__get_user_atomic_128_aligned(vbuf, addr, rc);
935 	if (rc) {
936 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
937 			 " instr=%08x addr=%016lx\n",
938 			 smp_processor_id(), current->comm, current->pid,
939 			 regs->nip, instr, (unsigned long)addr);
940 		return;
941 	}
942 
943 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
944 		 " instr=%08x addr=%016lx\n",
945 		 smp_processor_id(), current->comm, current->pid, regs->nip,
946 		 instr, (unsigned long) addr);
947 
948 	/* Grab instruction "selector" */
949 	sel = (instr >> 6) & 3;
950 
951 	/*
952 	 * Check to make sure the facility is actually enabled. This
953 	 * could happen if we get a false positive hit.
954 	 *
955 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
956 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
957 	 */
958 	msr_mask = MSR_VSX;
959 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
960 		msr_mask = MSR_VEC;
961 	if (!(msr & msr_mask)) {
962 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
963 			 " instr=%08x msr:%016lx\n",
964 			 smp_processor_id(), current->comm, current->pid,
965 			 regs->nip, instr, msr);
966 		return;
967 	}
968 
969 	/* Do logging here before we modify sel based on endian */
970 	switch (sel) {
971 	case 0:	/* lxvw4x */
972 		PPC_WARN_EMULATED(lxvw4x, regs);
973 		break;
974 	case 1: /* lxvh8x */
975 		PPC_WARN_EMULATED(lxvh8x, regs);
976 		break;
977 	case 2: /* lxvd2x */
978 		PPC_WARN_EMULATED(lxvd2x, regs);
979 		break;
980 	case 3: /* lxvb16x */
981 		PPC_WARN_EMULATED(lxvb16x, regs);
982 		break;
983 	}
984 
985 #ifdef __LITTLE_ENDIAN__
986 	/*
987 	 * An LE kernel stores the vector in the task struct as an LE
988 	 * byte array (effectively swapping both the components and
989 	 * the content of the components). Those instructions expect
990 	 * the components to remain in ascending address order, so we
991 	 * swap them back.
992 	 *
993 	 * If we are running a BE user space, the expectation is that
994 	 * of a simple memcpy, so forcing the emulation to look like
995 	 * a lxvb16x should do the trick.
996 	 */
997 	if (swap)
998 		sel = 3;
999 
1000 	switch (sel) {
1001 	case 0:	/* lxvw4x */
1002 		for (i = 0; i < 4; i++)
1003 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1004 		break;
1005 	case 1: /* lxvh8x */
1006 		for (i = 0; i < 8; i++)
1007 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1008 		break;
1009 	case 2: /* lxvd2x */
1010 		for (i = 0; i < 2; i++)
1011 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1012 		break;
1013 	case 3: /* lxvb16x */
1014 		for (i = 0; i < 16; i++)
1015 			vdst[i] = vbuf[15-i];
1016 		break;
1017 	}
1018 #else /* __LITTLE_ENDIAN__ */
1019 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1020 	if (!swap)
1021 		sel = 3;
1022 
1023 	/* Otherwise, we need to swap the content of the components */
1024 	switch (sel) {
1025 	case 0:	/* lxvw4x */
1026 		for (i = 0; i < 4; i++)
1027 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1028 		break;
1029 	case 1: /* lxvh8x */
1030 		for (i = 0; i < 8; i++)
1031 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1032 		break;
1033 	case 2: /* lxvd2x */
1034 		for (i = 0; i < 2; i++)
1035 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1036 		break;
1037 	case 3: /* lxvb16x */
1038 		memcpy(vdst, vbuf, 16);
1039 		break;
1040 	}
1041 #endif /* !__LITTLE_ENDIAN__ */
1042 
1043 	/* Go to next instruction */
1044 	regs->nip += 4;
1045 }
1046 #endif /* CONFIG_VSX */
1047 
1048 void handle_hmi_exception(struct pt_regs *regs)
1049 {
1050 	struct pt_regs *old_regs;
1051 
1052 	old_regs = set_irq_regs(regs);
1053 	irq_enter();
1054 
1055 #ifdef CONFIG_VSX
1056 	/* Real mode flagged P9 special emu is needed */
1057 	if (local_paca->hmi_p9_special_emu) {
1058 		local_paca->hmi_p9_special_emu = 0;
1059 
1060 		/*
1061 		 * We don't want to take page faults while doing the
1062 		 * emulation, we just replay the instruction if necessary.
1063 		 */
1064 		pagefault_disable();
1065 		p9_hmi_special_emu(regs);
1066 		pagefault_enable();
1067 	}
1068 #endif /* CONFIG_VSX */
1069 
1070 	if (ppc_md.handle_hmi_exception)
1071 		ppc_md.handle_hmi_exception(regs);
1072 
1073 	irq_exit();
1074 	set_irq_regs(old_regs);
1075 }
1076 
1077 void unknown_exception(struct pt_regs *regs)
1078 {
1079 	enum ctx_state prev_state = exception_enter();
1080 
1081 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1082 	       regs->nip, regs->msr, regs->trap);
1083 
1084 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1085 
1086 	exception_exit(prev_state);
1087 }
1088 
1089 void instruction_breakpoint_exception(struct pt_regs *regs)
1090 {
1091 	enum ctx_state prev_state = exception_enter();
1092 
1093 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1094 					5, SIGTRAP) == NOTIFY_STOP)
1095 		goto bail;
1096 	if (debugger_iabr_match(regs))
1097 		goto bail;
1098 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1099 
1100 bail:
1101 	exception_exit(prev_state);
1102 }
1103 
1104 void RunModeException(struct pt_regs *regs)
1105 {
1106 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1107 }
1108 
1109 void single_step_exception(struct pt_regs *regs)
1110 {
1111 	enum ctx_state prev_state = exception_enter();
1112 
1113 	clear_single_step(regs);
1114 	clear_br_trace(regs);
1115 
1116 	if (kprobe_post_handler(regs))
1117 		return;
1118 
1119 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1120 					5, SIGTRAP) == NOTIFY_STOP)
1121 		goto bail;
1122 	if (debugger_sstep(regs))
1123 		goto bail;
1124 
1125 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1126 
1127 bail:
1128 	exception_exit(prev_state);
1129 }
1130 NOKPROBE_SYMBOL(single_step_exception);
1131 
1132 /*
1133  * After we have successfully emulated an instruction, we have to
1134  * check if the instruction was being single-stepped, and if so,
1135  * pretend we got a single-step exception.  This was pointed out
1136  * by Kumar Gala.  -- paulus
1137  */
1138 static void emulate_single_step(struct pt_regs *regs)
1139 {
1140 	if (single_stepping(regs))
1141 		single_step_exception(regs);
1142 }
1143 
1144 static inline int __parse_fpscr(unsigned long fpscr)
1145 {
1146 	int ret = FPE_FLTUNK;
1147 
1148 	/* Invalid operation */
1149 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1150 		ret = FPE_FLTINV;
1151 
1152 	/* Overflow */
1153 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1154 		ret = FPE_FLTOVF;
1155 
1156 	/* Underflow */
1157 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1158 		ret = FPE_FLTUND;
1159 
1160 	/* Divide by zero */
1161 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1162 		ret = FPE_FLTDIV;
1163 
1164 	/* Inexact result */
1165 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1166 		ret = FPE_FLTRES;
1167 
1168 	return ret;
1169 }
1170 
1171 static void parse_fpe(struct pt_regs *regs)
1172 {
1173 	int code = 0;
1174 
1175 	flush_fp_to_thread(current);
1176 
1177 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1178 
1179 	_exception(SIGFPE, regs, code, regs->nip);
1180 }
1181 
1182 /*
1183  * Illegal instruction emulation support.  Originally written to
1184  * provide the PVR to user applications using the mfspr rd, PVR.
1185  * Return non-zero if we can't emulate, or -EFAULT if the associated
1186  * memory access caused an access fault.  Return zero on success.
1187  *
1188  * There are a couple of ways to do this, either "decode" the instruction
1189  * or directly match lots of bits.  In this case, matching lots of
1190  * bits is faster and easier.
1191  *
1192  */
1193 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1194 {
1195 	u8 rT = (instword >> 21) & 0x1f;
1196 	u8 rA = (instword >> 16) & 0x1f;
1197 	u8 NB_RB = (instword >> 11) & 0x1f;
1198 	u32 num_bytes;
1199 	unsigned long EA;
1200 	int pos = 0;
1201 
1202 	/* Early out if we are an invalid form of lswx */
1203 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1204 		if ((rT == rA) || (rT == NB_RB))
1205 			return -EINVAL;
1206 
1207 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1208 
1209 	switch (instword & PPC_INST_STRING_MASK) {
1210 		case PPC_INST_LSWX:
1211 		case PPC_INST_STSWX:
1212 			EA += NB_RB;
1213 			num_bytes = regs->xer & 0x7f;
1214 			break;
1215 		case PPC_INST_LSWI:
1216 		case PPC_INST_STSWI:
1217 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1218 			break;
1219 		default:
1220 			return -EINVAL;
1221 	}
1222 
1223 	while (num_bytes != 0)
1224 	{
1225 		u8 val;
1226 		u32 shift = 8 * (3 - (pos & 0x3));
1227 
1228 		/* if process is 32-bit, clear upper 32 bits of EA */
1229 		if ((regs->msr & MSR_64BIT) == 0)
1230 			EA &= 0xFFFFFFFF;
1231 
1232 		switch ((instword & PPC_INST_STRING_MASK)) {
1233 			case PPC_INST_LSWX:
1234 			case PPC_INST_LSWI:
1235 				if (get_user(val, (u8 __user *)EA))
1236 					return -EFAULT;
1237 				/* first time updating this reg,
1238 				 * zero it out */
1239 				if (pos == 0)
1240 					regs->gpr[rT] = 0;
1241 				regs->gpr[rT] |= val << shift;
1242 				break;
1243 			case PPC_INST_STSWI:
1244 			case PPC_INST_STSWX:
1245 				val = regs->gpr[rT] >> shift;
1246 				if (put_user(val, (u8 __user *)EA))
1247 					return -EFAULT;
1248 				break;
1249 		}
1250 		/* move EA to next address */
1251 		EA += 1;
1252 		num_bytes--;
1253 
1254 		/* manage our position within the register */
1255 		if (++pos == 4) {
1256 			pos = 0;
1257 			if (++rT == 32)
1258 				rT = 0;
1259 		}
1260 	}
1261 
1262 	return 0;
1263 }
1264 
1265 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1266 {
1267 	u32 ra,rs;
1268 	unsigned long tmp;
1269 
1270 	ra = (instword >> 16) & 0x1f;
1271 	rs = (instword >> 21) & 0x1f;
1272 
1273 	tmp = regs->gpr[rs];
1274 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1275 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1276 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1277 	regs->gpr[ra] = tmp;
1278 
1279 	return 0;
1280 }
1281 
1282 static int emulate_isel(struct pt_regs *regs, u32 instword)
1283 {
1284 	u8 rT = (instword >> 21) & 0x1f;
1285 	u8 rA = (instword >> 16) & 0x1f;
1286 	u8 rB = (instword >> 11) & 0x1f;
1287 	u8 BC = (instword >> 6) & 0x1f;
1288 	u8 bit;
1289 	unsigned long tmp;
1290 
1291 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1292 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1293 
1294 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1295 
1296 	return 0;
1297 }
1298 
1299 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1300 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1301 {
1302         /* If we're emulating a load/store in an active transaction, we cannot
1303          * emulate it as the kernel operates in transaction suspended context.
1304          * We need to abort the transaction.  This creates a persistent TM
1305          * abort so tell the user what caused it with a new code.
1306 	 */
1307 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1308 		tm_enable();
1309 		tm_abort(cause);
1310 		return true;
1311 	}
1312 	return false;
1313 }
1314 #else
1315 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1316 {
1317 	return false;
1318 }
1319 #endif
1320 
1321 static int emulate_instruction(struct pt_regs *regs)
1322 {
1323 	u32 instword;
1324 	u32 rd;
1325 
1326 	if (!user_mode(regs))
1327 		return -EINVAL;
1328 	CHECK_FULL_REGS(regs);
1329 
1330 	if (get_user(instword, (u32 __user *)(regs->nip)))
1331 		return -EFAULT;
1332 
1333 	/* Emulate the mfspr rD, PVR. */
1334 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1335 		PPC_WARN_EMULATED(mfpvr, regs);
1336 		rd = (instword >> 21) & 0x1f;
1337 		regs->gpr[rd] = mfspr(SPRN_PVR);
1338 		return 0;
1339 	}
1340 
1341 	/* Emulating the dcba insn is just a no-op.  */
1342 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1343 		PPC_WARN_EMULATED(dcba, regs);
1344 		return 0;
1345 	}
1346 
1347 	/* Emulate the mcrxr insn.  */
1348 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1349 		int shift = (instword >> 21) & 0x1c;
1350 		unsigned long msk = 0xf0000000UL >> shift;
1351 
1352 		PPC_WARN_EMULATED(mcrxr, regs);
1353 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1354 		regs->xer &= ~0xf0000000UL;
1355 		return 0;
1356 	}
1357 
1358 	/* Emulate load/store string insn. */
1359 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1360 		if (tm_abort_check(regs,
1361 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1362 			return -EINVAL;
1363 		PPC_WARN_EMULATED(string, regs);
1364 		return emulate_string_inst(regs, instword);
1365 	}
1366 
1367 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1368 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1369 		PPC_WARN_EMULATED(popcntb, regs);
1370 		return emulate_popcntb_inst(regs, instword);
1371 	}
1372 
1373 	/* Emulate isel (Integer Select) instruction */
1374 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1375 		PPC_WARN_EMULATED(isel, regs);
1376 		return emulate_isel(regs, instword);
1377 	}
1378 
1379 	/* Emulate sync instruction variants */
1380 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1381 		PPC_WARN_EMULATED(sync, regs);
1382 		asm volatile("sync");
1383 		return 0;
1384 	}
1385 
1386 #ifdef CONFIG_PPC64
1387 	/* Emulate the mfspr rD, DSCR. */
1388 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1389 		PPC_INST_MFSPR_DSCR_USER) ||
1390 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1391 		PPC_INST_MFSPR_DSCR)) &&
1392 			cpu_has_feature(CPU_FTR_DSCR)) {
1393 		PPC_WARN_EMULATED(mfdscr, regs);
1394 		rd = (instword >> 21) & 0x1f;
1395 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1396 		return 0;
1397 	}
1398 	/* Emulate the mtspr DSCR, rD. */
1399 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1400 		PPC_INST_MTSPR_DSCR_USER) ||
1401 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1402 		PPC_INST_MTSPR_DSCR)) &&
1403 			cpu_has_feature(CPU_FTR_DSCR)) {
1404 		PPC_WARN_EMULATED(mtdscr, regs);
1405 		rd = (instword >> 21) & 0x1f;
1406 		current->thread.dscr = regs->gpr[rd];
1407 		current->thread.dscr_inherit = 1;
1408 		mtspr(SPRN_DSCR, current->thread.dscr);
1409 		return 0;
1410 	}
1411 #endif
1412 
1413 	return -EINVAL;
1414 }
1415 
1416 int is_valid_bugaddr(unsigned long addr)
1417 {
1418 	return is_kernel_addr(addr);
1419 }
1420 
1421 #ifdef CONFIG_MATH_EMULATION
1422 static int emulate_math(struct pt_regs *regs)
1423 {
1424 	int ret;
1425 	extern int do_mathemu(struct pt_regs *regs);
1426 
1427 	ret = do_mathemu(regs);
1428 	if (ret >= 0)
1429 		PPC_WARN_EMULATED(math, regs);
1430 
1431 	switch (ret) {
1432 	case 0:
1433 		emulate_single_step(regs);
1434 		return 0;
1435 	case 1: {
1436 			int code = 0;
1437 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1438 			_exception(SIGFPE, regs, code, regs->nip);
1439 			return 0;
1440 		}
1441 	case -EFAULT:
1442 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1443 		return 0;
1444 	}
1445 
1446 	return -1;
1447 }
1448 #else
1449 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1450 #endif
1451 
1452 void program_check_exception(struct pt_regs *regs)
1453 {
1454 	enum ctx_state prev_state = exception_enter();
1455 	unsigned int reason = get_reason(regs);
1456 
1457 	/* We can now get here via a FP Unavailable exception if the core
1458 	 * has no FPU, in that case the reason flags will be 0 */
1459 
1460 	if (reason & REASON_FP) {
1461 		/* IEEE FP exception */
1462 		parse_fpe(regs);
1463 		goto bail;
1464 	}
1465 	if (reason & REASON_TRAP) {
1466 		unsigned long bugaddr;
1467 		/* Debugger is first in line to stop recursive faults in
1468 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1469 		if (debugger_bpt(regs))
1470 			goto bail;
1471 
1472 		if (kprobe_handler(regs))
1473 			goto bail;
1474 
1475 		/* trap exception */
1476 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1477 				== NOTIFY_STOP)
1478 			goto bail;
1479 
1480 		bugaddr = regs->nip;
1481 		/*
1482 		 * Fixup bugaddr for BUG_ON() in real mode
1483 		 */
1484 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1485 			bugaddr += PAGE_OFFSET;
1486 
1487 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1488 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1489 			regs->nip += 4;
1490 			goto bail;
1491 		}
1492 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1493 		goto bail;
1494 	}
1495 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1496 	if (reason & REASON_TM) {
1497 		/* This is a TM "Bad Thing Exception" program check.
1498 		 * This occurs when:
1499 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1500 		 *    transition in TM states.
1501 		 * -  A trechkpt is attempted when transactional.
1502 		 * -  A treclaim is attempted when non transactional.
1503 		 * -  A tend is illegally attempted.
1504 		 * -  writing a TM SPR when transactional.
1505 		 *
1506 		 * If usermode caused this, it's done something illegal and
1507 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1508 		 * operand to distinguish from the instruction just being bad
1509 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1510 		 * illegal /placement/ of a valid instruction.
1511 		 */
1512 		if (user_mode(regs)) {
1513 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1514 			goto bail;
1515 		} else {
1516 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1517 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1518 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1519 			die("Unrecoverable exception", regs, SIGABRT);
1520 		}
1521 	}
1522 #endif
1523 
1524 	/*
1525 	 * If we took the program check in the kernel skip down to sending a
1526 	 * SIGILL. The subsequent cases all relate to emulating instructions
1527 	 * which we should only do for userspace. We also do not want to enable
1528 	 * interrupts for kernel faults because that might lead to further
1529 	 * faults, and loose the context of the original exception.
1530 	 */
1531 	if (!user_mode(regs))
1532 		goto sigill;
1533 
1534 	/* We restore the interrupt state now */
1535 	if (!arch_irq_disabled_regs(regs))
1536 		local_irq_enable();
1537 
1538 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1539 	 * but there seems to be a hardware bug on the 405GP (RevD)
1540 	 * that means ESR is sometimes set incorrectly - either to
1541 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1542 	 * hardware people - not sure if it can happen on any illegal
1543 	 * instruction or only on FP instructions, whether there is a
1544 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1545 	 */
1546 	if (!emulate_math(regs))
1547 		goto bail;
1548 
1549 	/* Try to emulate it if we should. */
1550 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1551 		switch (emulate_instruction(regs)) {
1552 		case 0:
1553 			regs->nip += 4;
1554 			emulate_single_step(regs);
1555 			goto bail;
1556 		case -EFAULT:
1557 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1558 			goto bail;
1559 		}
1560 	}
1561 
1562 sigill:
1563 	if (reason & REASON_PRIVILEGED)
1564 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1565 	else
1566 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1567 
1568 bail:
1569 	exception_exit(prev_state);
1570 }
1571 NOKPROBE_SYMBOL(program_check_exception);
1572 
1573 /*
1574  * This occurs when running in hypervisor mode on POWER6 or later
1575  * and an illegal instruction is encountered.
1576  */
1577 void emulation_assist_interrupt(struct pt_regs *regs)
1578 {
1579 	regs->msr |= REASON_ILLEGAL;
1580 	program_check_exception(regs);
1581 }
1582 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1583 
1584 void alignment_exception(struct pt_regs *regs)
1585 {
1586 	enum ctx_state prev_state = exception_enter();
1587 	int sig, code, fixed = 0;
1588 
1589 	/* We restore the interrupt state now */
1590 	if (!arch_irq_disabled_regs(regs))
1591 		local_irq_enable();
1592 
1593 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1594 		goto bail;
1595 
1596 	/* we don't implement logging of alignment exceptions */
1597 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1598 		fixed = fix_alignment(regs);
1599 
1600 	if (fixed == 1) {
1601 		regs->nip += 4;	/* skip over emulated instruction */
1602 		emulate_single_step(regs);
1603 		goto bail;
1604 	}
1605 
1606 	/* Operand address was bad */
1607 	if (fixed == -EFAULT) {
1608 		sig = SIGSEGV;
1609 		code = SEGV_ACCERR;
1610 	} else {
1611 		sig = SIGBUS;
1612 		code = BUS_ADRALN;
1613 	}
1614 	if (user_mode(regs))
1615 		_exception(sig, regs, code, regs->dar);
1616 	else
1617 		bad_page_fault(regs, regs->dar, sig);
1618 
1619 bail:
1620 	exception_exit(prev_state);
1621 }
1622 
1623 void StackOverflow(struct pt_regs *regs)
1624 {
1625 	pr_crit("Kernel stack overflow in process %s[%d], r1=%lx\n",
1626 		current->comm, task_pid_nr(current), regs->gpr[1]);
1627 	debugger(regs);
1628 	show_regs(regs);
1629 	panic("kernel stack overflow");
1630 }
1631 
1632 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1633 {
1634 	enum ctx_state prev_state = exception_enter();
1635 
1636 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1637 			  "%lx at %lx\n", regs->trap, regs->nip);
1638 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1639 
1640 	exception_exit(prev_state);
1641 }
1642 
1643 void altivec_unavailable_exception(struct pt_regs *regs)
1644 {
1645 	enum ctx_state prev_state = exception_enter();
1646 
1647 	if (user_mode(regs)) {
1648 		/* A user program has executed an altivec instruction,
1649 		   but this kernel doesn't support altivec. */
1650 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1651 		goto bail;
1652 	}
1653 
1654 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1655 			"%lx at %lx\n", regs->trap, regs->nip);
1656 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1657 
1658 bail:
1659 	exception_exit(prev_state);
1660 }
1661 
1662 void vsx_unavailable_exception(struct pt_regs *regs)
1663 {
1664 	if (user_mode(regs)) {
1665 		/* A user program has executed an vsx instruction,
1666 		   but this kernel doesn't support vsx. */
1667 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1668 		return;
1669 	}
1670 
1671 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1672 			"%lx at %lx\n", regs->trap, regs->nip);
1673 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1674 }
1675 
1676 #ifdef CONFIG_PPC64
1677 static void tm_unavailable(struct pt_regs *regs)
1678 {
1679 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1680 	if (user_mode(regs)) {
1681 		current->thread.load_tm++;
1682 		regs->msr |= MSR_TM;
1683 		tm_enable();
1684 		tm_restore_sprs(&current->thread);
1685 		return;
1686 	}
1687 #endif
1688 	pr_emerg("Unrecoverable TM Unavailable Exception "
1689 			"%lx at %lx\n", regs->trap, regs->nip);
1690 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1691 }
1692 
1693 void facility_unavailable_exception(struct pt_regs *regs)
1694 {
1695 	static char *facility_strings[] = {
1696 		[FSCR_FP_LG] = "FPU",
1697 		[FSCR_VECVSX_LG] = "VMX/VSX",
1698 		[FSCR_DSCR_LG] = "DSCR",
1699 		[FSCR_PM_LG] = "PMU SPRs",
1700 		[FSCR_BHRB_LG] = "BHRB",
1701 		[FSCR_TM_LG] = "TM",
1702 		[FSCR_EBB_LG] = "EBB",
1703 		[FSCR_TAR_LG] = "TAR",
1704 		[FSCR_MSGP_LG] = "MSGP",
1705 		[FSCR_SCV_LG] = "SCV",
1706 	};
1707 	char *facility = "unknown";
1708 	u64 value;
1709 	u32 instword, rd;
1710 	u8 status;
1711 	bool hv;
1712 
1713 	hv = (TRAP(regs) == 0xf80);
1714 	if (hv)
1715 		value = mfspr(SPRN_HFSCR);
1716 	else
1717 		value = mfspr(SPRN_FSCR);
1718 
1719 	status = value >> 56;
1720 	if ((hv || status >= 2) &&
1721 	    (status < ARRAY_SIZE(facility_strings)) &&
1722 	    facility_strings[status])
1723 		facility = facility_strings[status];
1724 
1725 	/* We should not have taken this interrupt in kernel */
1726 	if (!user_mode(regs)) {
1727 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1728 			 facility, status, regs->nip);
1729 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1730 	}
1731 
1732 	/* We restore the interrupt state now */
1733 	if (!arch_irq_disabled_regs(regs))
1734 		local_irq_enable();
1735 
1736 	if (status == FSCR_DSCR_LG) {
1737 		/*
1738 		 * User is accessing the DSCR register using the problem
1739 		 * state only SPR number (0x03) either through a mfspr or
1740 		 * a mtspr instruction. If it is a write attempt through
1741 		 * a mtspr, then we set the inherit bit. This also allows
1742 		 * the user to write or read the register directly in the
1743 		 * future by setting via the FSCR DSCR bit. But in case it
1744 		 * is a read DSCR attempt through a mfspr instruction, we
1745 		 * just emulate the instruction instead. This code path will
1746 		 * always emulate all the mfspr instructions till the user
1747 		 * has attempted at least one mtspr instruction. This way it
1748 		 * preserves the same behaviour when the user is accessing
1749 		 * the DSCR through privilege level only SPR number (0x11)
1750 		 * which is emulated through illegal instruction exception.
1751 		 * We always leave HFSCR DSCR set.
1752 		 */
1753 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1754 			pr_err("Failed to fetch the user instruction\n");
1755 			return;
1756 		}
1757 
1758 		/* Write into DSCR (mtspr 0x03, RS) */
1759 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1760 				== PPC_INST_MTSPR_DSCR_USER) {
1761 			rd = (instword >> 21) & 0x1f;
1762 			current->thread.dscr = regs->gpr[rd];
1763 			current->thread.dscr_inherit = 1;
1764 			current->thread.fscr |= FSCR_DSCR;
1765 			mtspr(SPRN_FSCR, current->thread.fscr);
1766 		}
1767 
1768 		/* Read from DSCR (mfspr RT, 0x03) */
1769 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1770 				== PPC_INST_MFSPR_DSCR_USER) {
1771 			if (emulate_instruction(regs)) {
1772 				pr_err("DSCR based mfspr emulation failed\n");
1773 				return;
1774 			}
1775 			regs->nip += 4;
1776 			emulate_single_step(regs);
1777 		}
1778 		return;
1779 	}
1780 
1781 	if (status == FSCR_TM_LG) {
1782 		/*
1783 		 * If we're here then the hardware is TM aware because it
1784 		 * generated an exception with FSRM_TM set.
1785 		 *
1786 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1787 		 * told us not to do TM, or the kernel is not built with TM
1788 		 * support.
1789 		 *
1790 		 * If both of those things are true, then userspace can spam the
1791 		 * console by triggering the printk() below just by continually
1792 		 * doing tbegin (or any TM instruction). So in that case just
1793 		 * send the process a SIGILL immediately.
1794 		 */
1795 		if (!cpu_has_feature(CPU_FTR_TM))
1796 			goto out;
1797 
1798 		tm_unavailable(regs);
1799 		return;
1800 	}
1801 
1802 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1803 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1804 
1805 out:
1806 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1807 }
1808 #endif
1809 
1810 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1811 
1812 void fp_unavailable_tm(struct pt_regs *regs)
1813 {
1814 	/* Note:  This does not handle any kind of FP laziness. */
1815 
1816 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1817 		 regs->nip, regs->msr);
1818 
1819         /* We can only have got here if the task started using FP after
1820          * beginning the transaction.  So, the transactional regs are just a
1821          * copy of the checkpointed ones.  But, we still need to recheckpoint
1822          * as we're enabling FP for the process; it will return, abort the
1823          * transaction, and probably retry but now with FP enabled.  So the
1824          * checkpointed FP registers need to be loaded.
1825 	 */
1826 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1827 
1828 	/*
1829 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1830 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1831 	 *
1832 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1833 	 * recheckpoint.
1834 	 */
1835 
1836 	/* Enable FP for the task: */
1837 	current->thread.load_fp = 1;
1838 
1839 	/*
1840 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1841 	 */
1842 	tm_recheckpoint(&current->thread);
1843 }
1844 
1845 void altivec_unavailable_tm(struct pt_regs *regs)
1846 {
1847 	/* See the comments in fp_unavailable_tm().  This function operates
1848 	 * the same way.
1849 	 */
1850 
1851 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1852 		 "MSR=%lx\n",
1853 		 regs->nip, regs->msr);
1854 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1855 	current->thread.load_vec = 1;
1856 	tm_recheckpoint(&current->thread);
1857 	current->thread.used_vr = 1;
1858 }
1859 
1860 void vsx_unavailable_tm(struct pt_regs *regs)
1861 {
1862 	/* See the comments in fp_unavailable_tm().  This works similarly,
1863 	 * though we're loading both FP and VEC registers in here.
1864 	 *
1865 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1866 	 * regs.  Either way, set MSR_VSX.
1867 	 */
1868 
1869 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1870 		 "MSR=%lx\n",
1871 		 regs->nip, regs->msr);
1872 
1873 	current->thread.used_vsr = 1;
1874 
1875 	/* This reclaims FP and/or VR regs if they're already enabled */
1876 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1877 
1878 	current->thread.load_vec = 1;
1879 	current->thread.load_fp = 1;
1880 
1881 	tm_recheckpoint(&current->thread);
1882 }
1883 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1884 
1885 void performance_monitor_exception(struct pt_regs *regs)
1886 {
1887 	__this_cpu_inc(irq_stat.pmu_irqs);
1888 
1889 	perf_irq(regs);
1890 }
1891 
1892 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1893 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1894 {
1895 	int changed = 0;
1896 	/*
1897 	 * Determine the cause of the debug event, clear the
1898 	 * event flags and send a trap to the handler. Torez
1899 	 */
1900 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1901 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1902 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1903 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1904 #endif
1905 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1906 			     5);
1907 		changed |= 0x01;
1908 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1909 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1910 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1911 			     6);
1912 		changed |= 0x01;
1913 	}  else if (debug_status & DBSR_IAC1) {
1914 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1915 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1916 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1917 			     1);
1918 		changed |= 0x01;
1919 	}  else if (debug_status & DBSR_IAC2) {
1920 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1921 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1922 			     2);
1923 		changed |= 0x01;
1924 	}  else if (debug_status & DBSR_IAC3) {
1925 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1926 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1927 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1928 			     3);
1929 		changed |= 0x01;
1930 	}  else if (debug_status & DBSR_IAC4) {
1931 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1932 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1933 			     4);
1934 		changed |= 0x01;
1935 	}
1936 	/*
1937 	 * At the point this routine was called, the MSR(DE) was turned off.
1938 	 * Check all other debug flags and see if that bit needs to be turned
1939 	 * back on or not.
1940 	 */
1941 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1942 			       current->thread.debug.dbcr1))
1943 		regs->msr |= MSR_DE;
1944 	else
1945 		/* Make sure the IDM flag is off */
1946 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1947 
1948 	if (changed & 0x01)
1949 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1950 }
1951 
1952 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1953 {
1954 	current->thread.debug.dbsr = debug_status;
1955 
1956 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1957 	 * on server, it stops on the target of the branch. In order to simulate
1958 	 * the server behaviour, we thus restart right away with a single step
1959 	 * instead of stopping here when hitting a BT
1960 	 */
1961 	if (debug_status & DBSR_BT) {
1962 		regs->msr &= ~MSR_DE;
1963 
1964 		/* Disable BT */
1965 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1966 		/* Clear the BT event */
1967 		mtspr(SPRN_DBSR, DBSR_BT);
1968 
1969 		/* Do the single step trick only when coming from userspace */
1970 		if (user_mode(regs)) {
1971 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1972 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1973 			regs->msr |= MSR_DE;
1974 			return;
1975 		}
1976 
1977 		if (kprobe_post_handler(regs))
1978 			return;
1979 
1980 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1981 			       5, SIGTRAP) == NOTIFY_STOP) {
1982 			return;
1983 		}
1984 		if (debugger_sstep(regs))
1985 			return;
1986 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1987 		regs->msr &= ~MSR_DE;
1988 
1989 		/* Disable instruction completion */
1990 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
1991 		/* Clear the instruction completion event */
1992 		mtspr(SPRN_DBSR, DBSR_IC);
1993 
1994 		if (kprobe_post_handler(regs))
1995 			return;
1996 
1997 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1998 			       5, SIGTRAP) == NOTIFY_STOP) {
1999 			return;
2000 		}
2001 
2002 		if (debugger_sstep(regs))
2003 			return;
2004 
2005 		if (user_mode(regs)) {
2006 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2007 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2008 					       current->thread.debug.dbcr1))
2009 				regs->msr |= MSR_DE;
2010 			else
2011 				/* Make sure the IDM bit is off */
2012 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2013 		}
2014 
2015 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2016 	} else
2017 		handle_debug(regs, debug_status);
2018 }
2019 NOKPROBE_SYMBOL(DebugException);
2020 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2021 
2022 #if !defined(CONFIG_TAU_INT)
2023 void TAUException(struct pt_regs *regs)
2024 {
2025 	printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
2026 	       regs->nip, regs->msr, regs->trap, print_tainted());
2027 }
2028 #endif /* CONFIG_INT_TAU */
2029 
2030 #ifdef CONFIG_ALTIVEC
2031 void altivec_assist_exception(struct pt_regs *regs)
2032 {
2033 	int err;
2034 
2035 	if (!user_mode(regs)) {
2036 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2037 		       " at %lx\n", regs->nip);
2038 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2039 	}
2040 
2041 	flush_altivec_to_thread(current);
2042 
2043 	PPC_WARN_EMULATED(altivec, regs);
2044 	err = emulate_altivec(regs);
2045 	if (err == 0) {
2046 		regs->nip += 4;		/* skip emulated instruction */
2047 		emulate_single_step(regs);
2048 		return;
2049 	}
2050 
2051 	if (err == -EFAULT) {
2052 		/* got an error reading the instruction */
2053 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2054 	} else {
2055 		/* didn't recognize the instruction */
2056 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2057 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2058 				   "in %s at %lx\n", current->comm, regs->nip);
2059 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2060 	}
2061 }
2062 #endif /* CONFIG_ALTIVEC */
2063 
2064 #ifdef CONFIG_FSL_BOOKE
2065 void CacheLockingException(struct pt_regs *regs, unsigned long address,
2066 			   unsigned long error_code)
2067 {
2068 	/* We treat cache locking instructions from the user
2069 	 * as priv ops, in the future we could try to do
2070 	 * something smarter
2071 	 */
2072 	if (error_code & (ESR_DLK|ESR_ILK))
2073 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2074 	return;
2075 }
2076 #endif /* CONFIG_FSL_BOOKE */
2077 
2078 #ifdef CONFIG_SPE
2079 void SPEFloatingPointException(struct pt_regs *regs)
2080 {
2081 	extern int do_spe_mathemu(struct pt_regs *regs);
2082 	unsigned long spefscr;
2083 	int fpexc_mode;
2084 	int code = FPE_FLTUNK;
2085 	int err;
2086 
2087 	/* We restore the interrupt state now */
2088 	if (!arch_irq_disabled_regs(regs))
2089 		local_irq_enable();
2090 
2091 	flush_spe_to_thread(current);
2092 
2093 	spefscr = current->thread.spefscr;
2094 	fpexc_mode = current->thread.fpexc_mode;
2095 
2096 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2097 		code = FPE_FLTOVF;
2098 	}
2099 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2100 		code = FPE_FLTUND;
2101 	}
2102 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2103 		code = FPE_FLTDIV;
2104 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2105 		code = FPE_FLTINV;
2106 	}
2107 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2108 		code = FPE_FLTRES;
2109 
2110 	err = do_spe_mathemu(regs);
2111 	if (err == 0) {
2112 		regs->nip += 4;		/* skip emulated instruction */
2113 		emulate_single_step(regs);
2114 		return;
2115 	}
2116 
2117 	if (err == -EFAULT) {
2118 		/* got an error reading the instruction */
2119 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2120 	} else if (err == -EINVAL) {
2121 		/* didn't recognize the instruction */
2122 		printk(KERN_ERR "unrecognized spe instruction "
2123 		       "in %s at %lx\n", current->comm, regs->nip);
2124 	} else {
2125 		_exception(SIGFPE, regs, code, regs->nip);
2126 	}
2127 
2128 	return;
2129 }
2130 
2131 void SPEFloatingPointRoundException(struct pt_regs *regs)
2132 {
2133 	extern int speround_handler(struct pt_regs *regs);
2134 	int err;
2135 
2136 	/* We restore the interrupt state now */
2137 	if (!arch_irq_disabled_regs(regs))
2138 		local_irq_enable();
2139 
2140 	preempt_disable();
2141 	if (regs->msr & MSR_SPE)
2142 		giveup_spe(current);
2143 	preempt_enable();
2144 
2145 	regs->nip -= 4;
2146 	err = speround_handler(regs);
2147 	if (err == 0) {
2148 		regs->nip += 4;		/* skip emulated instruction */
2149 		emulate_single_step(regs);
2150 		return;
2151 	}
2152 
2153 	if (err == -EFAULT) {
2154 		/* got an error reading the instruction */
2155 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2156 	} else if (err == -EINVAL) {
2157 		/* didn't recognize the instruction */
2158 		printk(KERN_ERR "unrecognized spe instruction "
2159 		       "in %s at %lx\n", current->comm, regs->nip);
2160 	} else {
2161 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2162 		return;
2163 	}
2164 }
2165 #endif
2166 
2167 /*
2168  * We enter here if we get an unrecoverable exception, that is, one
2169  * that happened at a point where the RI (recoverable interrupt) bit
2170  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2171  * we therefore lost state by taking this exception.
2172  */
2173 void unrecoverable_exception(struct pt_regs *regs)
2174 {
2175 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2176 		 regs->trap, regs->nip, regs->msr);
2177 	die("Unrecoverable exception", regs, SIGABRT);
2178 }
2179 NOKPROBE_SYMBOL(unrecoverable_exception);
2180 
2181 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2182 /*
2183  * Default handler for a Watchdog exception,
2184  * spins until a reboot occurs
2185  */
2186 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2187 {
2188 	/* Generic WatchdogHandler, implement your own */
2189 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2190 	return;
2191 }
2192 
2193 void WatchdogException(struct pt_regs *regs)
2194 {
2195 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2196 	WatchdogHandler(regs);
2197 }
2198 #endif
2199 
2200 /*
2201  * We enter here if we discover during exception entry that we are
2202  * running in supervisor mode with a userspace value in the stack pointer.
2203  */
2204 void kernel_bad_stack(struct pt_regs *regs)
2205 {
2206 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2207 	       regs->gpr[1], regs->nip);
2208 	die("Bad kernel stack pointer", regs, SIGABRT);
2209 }
2210 NOKPROBE_SYMBOL(kernel_bad_stack);
2211 
2212 void __init trap_init(void)
2213 {
2214 }
2215 
2216 
2217 #ifdef CONFIG_PPC_EMULATED_STATS
2218 
2219 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2220 
2221 struct ppc_emulated ppc_emulated = {
2222 #ifdef CONFIG_ALTIVEC
2223 	WARN_EMULATED_SETUP(altivec),
2224 #endif
2225 	WARN_EMULATED_SETUP(dcba),
2226 	WARN_EMULATED_SETUP(dcbz),
2227 	WARN_EMULATED_SETUP(fp_pair),
2228 	WARN_EMULATED_SETUP(isel),
2229 	WARN_EMULATED_SETUP(mcrxr),
2230 	WARN_EMULATED_SETUP(mfpvr),
2231 	WARN_EMULATED_SETUP(multiple),
2232 	WARN_EMULATED_SETUP(popcntb),
2233 	WARN_EMULATED_SETUP(spe),
2234 	WARN_EMULATED_SETUP(string),
2235 	WARN_EMULATED_SETUP(sync),
2236 	WARN_EMULATED_SETUP(unaligned),
2237 #ifdef CONFIG_MATH_EMULATION
2238 	WARN_EMULATED_SETUP(math),
2239 #endif
2240 #ifdef CONFIG_VSX
2241 	WARN_EMULATED_SETUP(vsx),
2242 #endif
2243 #ifdef CONFIG_PPC64
2244 	WARN_EMULATED_SETUP(mfdscr),
2245 	WARN_EMULATED_SETUP(mtdscr),
2246 	WARN_EMULATED_SETUP(lq_stq),
2247 	WARN_EMULATED_SETUP(lxvw4x),
2248 	WARN_EMULATED_SETUP(lxvh8x),
2249 	WARN_EMULATED_SETUP(lxvd2x),
2250 	WARN_EMULATED_SETUP(lxvb16x),
2251 #endif
2252 };
2253 
2254 u32 ppc_warn_emulated;
2255 
2256 void ppc_warn_emulated_print(const char *type)
2257 {
2258 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2259 			    type);
2260 }
2261 
2262 static int __init ppc_warn_emulated_init(void)
2263 {
2264 	struct dentry *dir, *d;
2265 	unsigned int i;
2266 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2267 
2268 	if (!powerpc_debugfs_root)
2269 		return -ENODEV;
2270 
2271 	dir = debugfs_create_dir("emulated_instructions",
2272 				 powerpc_debugfs_root);
2273 	if (!dir)
2274 		return -ENOMEM;
2275 
2276 	d = debugfs_create_u32("do_warn", 0644, dir,
2277 			       &ppc_warn_emulated);
2278 	if (!d)
2279 		goto fail;
2280 
2281 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) {
2282 		d = debugfs_create_u32(entries[i].name, 0644, dir,
2283 				       (u32 *)&entries[i].val.counter);
2284 		if (!d)
2285 			goto fail;
2286 	}
2287 
2288 	return 0;
2289 
2290 fail:
2291 	debugfs_remove_recursive(dir);
2292 	return -ENOMEM;
2293 }
2294 
2295 device_initcall(ppc_warn_emulated_init);
2296 
2297 #endif /* CONFIG_PPC_EMULATED_STATS */
2298