xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision 3a83e4e6)
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  * Note that the 601 only takes a machine check on TEA
533  * (transfer error ack) signal assertion, and does not
534  * set any of the top 16 bits of SRR1.
535  *  -- paulus.
536  */
537 static inline int check_io_access(struct pt_regs *regs)
538 {
539 #ifdef CONFIG_PPC32
540 	unsigned long msr = regs->msr;
541 	const struct exception_table_entry *entry;
542 	unsigned int *nip = (unsigned int *)regs->nip;
543 
544 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
545 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
546 		/*
547 		 * Check that it's a sync instruction, or somewhere
548 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
549 		 * As the address is in the exception table
550 		 * we should be able to read the instr there.
551 		 * For the debug message, we look at the preceding
552 		 * load or store.
553 		 */
554 		if (*nip == PPC_INST_NOP)
555 			nip -= 2;
556 		else if (*nip == PPC_INST_ISYNC)
557 			--nip;
558 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
559 			unsigned int rb;
560 
561 			--nip;
562 			rb = (*nip >> 11) & 0x1f;
563 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
564 			       (*nip & 0x100)? "OUT to": "IN from",
565 			       regs->gpr[rb] - _IO_BASE, nip);
566 			regs->msr |= MSR_RI;
567 			regs->nip = extable_fixup(entry);
568 			return 1;
569 		}
570 	}
571 #endif /* CONFIG_PPC32 */
572 	return 0;
573 }
574 
575 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
576 /* On 4xx, the reason for the machine check or program exception
577    is in the ESR. */
578 #define get_reason(regs)	((regs)->dsisr)
579 #define REASON_FP		ESR_FP
580 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
581 #define REASON_PRIVILEGED	ESR_PPR
582 #define REASON_TRAP		ESR_PTR
583 #define REASON_PREFIXED		0
584 #define REASON_BOUNDARY		0
585 
586 /* single-step stuff */
587 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
588 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
589 #define clear_br_trace(regs)	do {} while(0)
590 #else
591 /* On non-4xx, the reason for the machine check or program
592    exception is in the MSR. */
593 #define get_reason(regs)	((regs)->msr)
594 #define REASON_TM		SRR1_PROGTM
595 #define REASON_FP		SRR1_PROGFPE
596 #define REASON_ILLEGAL		SRR1_PROGILL
597 #define REASON_PRIVILEGED	SRR1_PROGPRIV
598 #define REASON_TRAP		SRR1_PROGTRAP
599 #define REASON_PREFIXED		SRR1_PREFIXED
600 #define REASON_BOUNDARY		SRR1_BOUNDARY
601 
602 #define single_stepping(regs)	((regs)->msr & MSR_SE)
603 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
604 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
605 #endif
606 
607 #define inst_length(reason)	(((reason) & REASON_PREFIXED) ? 8 : 4)
608 
609 #if defined(CONFIG_E500)
610 int machine_check_e500mc(struct pt_regs *regs)
611 {
612 	unsigned long mcsr = mfspr(SPRN_MCSR);
613 	unsigned long pvr = mfspr(SPRN_PVR);
614 	unsigned long reason = mcsr;
615 	int recoverable = 1;
616 
617 	if (reason & MCSR_LD) {
618 		recoverable = fsl_rio_mcheck_exception(regs);
619 		if (recoverable == 1)
620 			goto silent_out;
621 	}
622 
623 	printk("Machine check in kernel mode.\n");
624 	printk("Caused by (from MCSR=%lx): ", reason);
625 
626 	if (reason & MCSR_MCP)
627 		pr_cont("Machine Check Signal\n");
628 
629 	if (reason & MCSR_ICPERR) {
630 		pr_cont("Instruction Cache Parity Error\n");
631 
632 		/*
633 		 * This is recoverable by invalidating the i-cache.
634 		 */
635 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
636 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
637 			;
638 
639 		/*
640 		 * This will generally be accompanied by an instruction
641 		 * fetch error report -- only treat MCSR_IF as fatal
642 		 * if it wasn't due to an L1 parity error.
643 		 */
644 		reason &= ~MCSR_IF;
645 	}
646 
647 	if (reason & MCSR_DCPERR_MC) {
648 		pr_cont("Data Cache Parity Error\n");
649 
650 		/*
651 		 * In write shadow mode we auto-recover from the error, but it
652 		 * may still get logged and cause a machine check.  We should
653 		 * only treat the non-write shadow case as non-recoverable.
654 		 */
655 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
656 		 * is not implemented but L1 data cache always runs in write
657 		 * shadow mode. Hence on data cache parity errors HW will
658 		 * automatically invalidate the L1 Data Cache.
659 		 */
660 		if (PVR_VER(pvr) != PVR_VER_E6500) {
661 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
662 				recoverable = 0;
663 		}
664 	}
665 
666 	if (reason & MCSR_L2MMU_MHIT) {
667 		pr_cont("Hit on multiple TLB entries\n");
668 		recoverable = 0;
669 	}
670 
671 	if (reason & MCSR_NMI)
672 		pr_cont("Non-maskable interrupt\n");
673 
674 	if (reason & MCSR_IF) {
675 		pr_cont("Instruction Fetch Error Report\n");
676 		recoverable = 0;
677 	}
678 
679 	if (reason & MCSR_LD) {
680 		pr_cont("Load Error Report\n");
681 		recoverable = 0;
682 	}
683 
684 	if (reason & MCSR_ST) {
685 		pr_cont("Store Error Report\n");
686 		recoverable = 0;
687 	}
688 
689 	if (reason & MCSR_LDG) {
690 		pr_cont("Guarded Load Error Report\n");
691 		recoverable = 0;
692 	}
693 
694 	if (reason & MCSR_TLBSYNC)
695 		pr_cont("Simultaneous tlbsync operations\n");
696 
697 	if (reason & MCSR_BSL2_ERR) {
698 		pr_cont("Level 2 Cache Error\n");
699 		recoverable = 0;
700 	}
701 
702 	if (reason & MCSR_MAV) {
703 		u64 addr;
704 
705 		addr = mfspr(SPRN_MCAR);
706 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
707 
708 		pr_cont("Machine Check %s Address: %#llx\n",
709 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
710 	}
711 
712 silent_out:
713 	mtspr(SPRN_MCSR, mcsr);
714 	return mfspr(SPRN_MCSR) == 0 && recoverable;
715 }
716 
717 int machine_check_e500(struct pt_regs *regs)
718 {
719 	unsigned long reason = mfspr(SPRN_MCSR);
720 
721 	if (reason & MCSR_BUS_RBERR) {
722 		if (fsl_rio_mcheck_exception(regs))
723 			return 1;
724 		if (fsl_pci_mcheck_exception(regs))
725 			return 1;
726 	}
727 
728 	printk("Machine check in kernel mode.\n");
729 	printk("Caused by (from MCSR=%lx): ", reason);
730 
731 	if (reason & MCSR_MCP)
732 		pr_cont("Machine Check Signal\n");
733 	if (reason & MCSR_ICPERR)
734 		pr_cont("Instruction Cache Parity Error\n");
735 	if (reason & MCSR_DCP_PERR)
736 		pr_cont("Data Cache Push Parity Error\n");
737 	if (reason & MCSR_DCPERR)
738 		pr_cont("Data Cache Parity Error\n");
739 	if (reason & MCSR_BUS_IAERR)
740 		pr_cont("Bus - Instruction Address Error\n");
741 	if (reason & MCSR_BUS_RAERR)
742 		pr_cont("Bus - Read Address Error\n");
743 	if (reason & MCSR_BUS_WAERR)
744 		pr_cont("Bus - Write Address Error\n");
745 	if (reason & MCSR_BUS_IBERR)
746 		pr_cont("Bus - Instruction Data Error\n");
747 	if (reason & MCSR_BUS_RBERR)
748 		pr_cont("Bus - Read Data Bus Error\n");
749 	if (reason & MCSR_BUS_WBERR)
750 		pr_cont("Bus - Write Data Bus Error\n");
751 	if (reason & MCSR_BUS_IPERR)
752 		pr_cont("Bus - Instruction Parity Error\n");
753 	if (reason & MCSR_BUS_RPERR)
754 		pr_cont("Bus - Read Parity Error\n");
755 
756 	return 0;
757 }
758 
759 int machine_check_generic(struct pt_regs *regs)
760 {
761 	return 0;
762 }
763 #elif defined(CONFIG_E200)
764 int machine_check_e200(struct pt_regs *regs)
765 {
766 	unsigned long reason = mfspr(SPRN_MCSR);
767 
768 	printk("Machine check in kernel mode.\n");
769 	printk("Caused by (from MCSR=%lx): ", reason);
770 
771 	if (reason & MCSR_MCP)
772 		pr_cont("Machine Check Signal\n");
773 	if (reason & MCSR_CP_PERR)
774 		pr_cont("Cache Push Parity Error\n");
775 	if (reason & MCSR_CPERR)
776 		pr_cont("Cache Parity Error\n");
777 	if (reason & MCSR_EXCP_ERR)
778 		pr_cont("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
779 	if (reason & MCSR_BUS_IRERR)
780 		pr_cont("Bus - Read Bus Error on instruction fetch\n");
781 	if (reason & MCSR_BUS_DRERR)
782 		pr_cont("Bus - Read Bus Error on data load\n");
783 	if (reason & MCSR_BUS_WRERR)
784 		pr_cont("Bus - Write Bus Error on buffered store or cache line push\n");
785 
786 	return 0;
787 }
788 #elif defined(CONFIG_PPC32)
789 int machine_check_generic(struct pt_regs *regs)
790 {
791 	unsigned long reason = regs->msr;
792 
793 	printk("Machine check in kernel mode.\n");
794 	printk("Caused by (from SRR1=%lx): ", reason);
795 	switch (reason & 0x601F0000) {
796 	case 0x80000:
797 		pr_cont("Machine check signal\n");
798 		break;
799 	case 0:		/* for 601 */
800 	case 0x40000:
801 	case 0x140000:	/* 7450 MSS error and TEA */
802 		pr_cont("Transfer error ack signal\n");
803 		break;
804 	case 0x20000:
805 		pr_cont("Data parity error signal\n");
806 		break;
807 	case 0x10000:
808 		pr_cont("Address parity error signal\n");
809 		break;
810 	case 0x20000000:
811 		pr_cont("L1 Data Cache error\n");
812 		break;
813 	case 0x40000000:
814 		pr_cont("L1 Instruction Cache error\n");
815 		break;
816 	case 0x00100000:
817 		pr_cont("L2 data cache parity error\n");
818 		break;
819 	default:
820 		pr_cont("Unknown values in msr\n");
821 	}
822 	return 0;
823 }
824 #endif /* everything else */
825 
826 void machine_check_exception(struct pt_regs *regs)
827 {
828 	int recover = 0;
829 
830 	/*
831 	 * BOOK3S_64 does not call this handler as a non-maskable interrupt
832 	 * (it uses its own early real-mode handler to handle the MCE proper
833 	 * and then raises irq_work to call this handler when interrupts are
834 	 * enabled).
835 	 *
836 	 * This is silly. The BOOK3S_64 should just call a different function
837 	 * rather than expecting semantics to magically change. Something
838 	 * like 'non_nmi_machine_check_exception()', perhaps?
839 	 */
840 	const bool nmi = !IS_ENABLED(CONFIG_PPC_BOOK3S_64);
841 
842 	if (nmi) nmi_enter();
843 
844 	__this_cpu_inc(irq_stat.mce_exceptions);
845 
846 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
847 
848 	/* See if any machine dependent calls. In theory, we would want
849 	 * to call the CPU first, and call the ppc_md. one if the CPU
850 	 * one returns a positive number. However there is existing code
851 	 * that assumes the board gets a first chance, so let's keep it
852 	 * that way for now and fix things later. --BenH.
853 	 */
854 	if (ppc_md.machine_check_exception)
855 		recover = ppc_md.machine_check_exception(regs);
856 	else if (cur_cpu_spec->machine_check)
857 		recover = cur_cpu_spec->machine_check(regs);
858 
859 	if (recover > 0)
860 		goto bail;
861 
862 	if (debugger_fault_handler(regs))
863 		goto bail;
864 
865 	if (check_io_access(regs))
866 		goto bail;
867 
868 	if (nmi) nmi_exit();
869 
870 	die("Machine check", regs, SIGBUS);
871 
872 	/* Must die if the interrupt is not recoverable */
873 	if (!(regs->msr & MSR_RI))
874 		die("Unrecoverable Machine check", regs, SIGBUS);
875 
876 	return;
877 
878 bail:
879 	if (nmi) nmi_exit();
880 }
881 
882 void SMIException(struct pt_regs *regs)
883 {
884 	die("System Management Interrupt", regs, SIGABRT);
885 }
886 
887 #ifdef CONFIG_VSX
888 static void p9_hmi_special_emu(struct pt_regs *regs)
889 {
890 	unsigned int ra, rb, t, i, sel, instr, rc;
891 	const void __user *addr;
892 	u8 vbuf[16], *vdst;
893 	unsigned long ea, msr, msr_mask;
894 	bool swap;
895 
896 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
897 		return;
898 
899 	/*
900 	 * lxvb16x	opcode: 0x7c0006d8
901 	 * lxvd2x	opcode: 0x7c000698
902 	 * lxvh8x	opcode: 0x7c000658
903 	 * lxvw4x	opcode: 0x7c000618
904 	 */
905 	if ((instr & 0xfc00073e) != 0x7c000618) {
906 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
907 			 " instr=%08x\n",
908 			 smp_processor_id(), current->comm, current->pid,
909 			 regs->nip, instr);
910 		return;
911 	}
912 
913 	/* Grab vector registers into the task struct */
914 	msr = regs->msr; /* Grab msr before we flush the bits */
915 	flush_vsx_to_thread(current);
916 	enable_kernel_altivec();
917 
918 	/*
919 	 * Is userspace running with a different endian (this is rare but
920 	 * not impossible)
921 	 */
922 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
923 
924 	/* Decode the instruction */
925 	ra = (instr >> 16) & 0x1f;
926 	rb = (instr >> 11) & 0x1f;
927 	t = (instr >> 21) & 0x1f;
928 	if (instr & 1)
929 		vdst = (u8 *)&current->thread.vr_state.vr[t];
930 	else
931 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
932 
933 	/* Grab the vector address */
934 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
935 	if (is_32bit_task())
936 		ea &= 0xfffffffful;
937 	addr = (__force const void __user *)ea;
938 
939 	/* Check it */
940 	if (!access_ok(addr, 16)) {
941 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
942 			 " instr=%08x addr=%016lx\n",
943 			 smp_processor_id(), current->comm, current->pid,
944 			 regs->nip, instr, (unsigned long)addr);
945 		return;
946 	}
947 
948 	/* Read the vector */
949 	rc = 0;
950 	if ((unsigned long)addr & 0xfUL)
951 		/* unaligned case */
952 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
953 	else
954 		__get_user_atomic_128_aligned(vbuf, addr, rc);
955 	if (rc) {
956 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
957 			 " instr=%08x addr=%016lx\n",
958 			 smp_processor_id(), current->comm, current->pid,
959 			 regs->nip, instr, (unsigned long)addr);
960 		return;
961 	}
962 
963 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
964 		 " instr=%08x addr=%016lx\n",
965 		 smp_processor_id(), current->comm, current->pid, regs->nip,
966 		 instr, (unsigned long) addr);
967 
968 	/* Grab instruction "selector" */
969 	sel = (instr >> 6) & 3;
970 
971 	/*
972 	 * Check to make sure the facility is actually enabled. This
973 	 * could happen if we get a false positive hit.
974 	 *
975 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
976 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
977 	 */
978 	msr_mask = MSR_VSX;
979 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
980 		msr_mask = MSR_VEC;
981 	if (!(msr & msr_mask)) {
982 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
983 			 " instr=%08x msr:%016lx\n",
984 			 smp_processor_id(), current->comm, current->pid,
985 			 regs->nip, instr, msr);
986 		return;
987 	}
988 
989 	/* Do logging here before we modify sel based on endian */
990 	switch (sel) {
991 	case 0:	/* lxvw4x */
992 		PPC_WARN_EMULATED(lxvw4x, regs);
993 		break;
994 	case 1: /* lxvh8x */
995 		PPC_WARN_EMULATED(lxvh8x, regs);
996 		break;
997 	case 2: /* lxvd2x */
998 		PPC_WARN_EMULATED(lxvd2x, regs);
999 		break;
1000 	case 3: /* lxvb16x */
1001 		PPC_WARN_EMULATED(lxvb16x, regs);
1002 		break;
1003 	}
1004 
1005 #ifdef __LITTLE_ENDIAN__
1006 	/*
1007 	 * An LE kernel stores the vector in the task struct as an LE
1008 	 * byte array (effectively swapping both the components and
1009 	 * the content of the components). Those instructions expect
1010 	 * the components to remain in ascending address order, so we
1011 	 * swap them back.
1012 	 *
1013 	 * If we are running a BE user space, the expectation is that
1014 	 * of a simple memcpy, so forcing the emulation to look like
1015 	 * a lxvb16x should do the trick.
1016 	 */
1017 	if (swap)
1018 		sel = 3;
1019 
1020 	switch (sel) {
1021 	case 0:	/* lxvw4x */
1022 		for (i = 0; i < 4; i++)
1023 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1024 		break;
1025 	case 1: /* lxvh8x */
1026 		for (i = 0; i < 8; i++)
1027 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1028 		break;
1029 	case 2: /* lxvd2x */
1030 		for (i = 0; i < 2; i++)
1031 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1032 		break;
1033 	case 3: /* lxvb16x */
1034 		for (i = 0; i < 16; i++)
1035 			vdst[i] = vbuf[15-i];
1036 		break;
1037 	}
1038 #else /* __LITTLE_ENDIAN__ */
1039 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1040 	if (!swap)
1041 		sel = 3;
1042 
1043 	/* Otherwise, we need to swap the content of the components */
1044 	switch (sel) {
1045 	case 0:	/* lxvw4x */
1046 		for (i = 0; i < 4; i++)
1047 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1048 		break;
1049 	case 1: /* lxvh8x */
1050 		for (i = 0; i < 8; i++)
1051 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1052 		break;
1053 	case 2: /* lxvd2x */
1054 		for (i = 0; i < 2; i++)
1055 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1056 		break;
1057 	case 3: /* lxvb16x */
1058 		memcpy(vdst, vbuf, 16);
1059 		break;
1060 	}
1061 #endif /* !__LITTLE_ENDIAN__ */
1062 
1063 	/* Go to next instruction */
1064 	regs->nip += 4;
1065 }
1066 #endif /* CONFIG_VSX */
1067 
1068 void handle_hmi_exception(struct pt_regs *regs)
1069 {
1070 	struct pt_regs *old_regs;
1071 
1072 	old_regs = set_irq_regs(regs);
1073 	irq_enter();
1074 
1075 #ifdef CONFIG_VSX
1076 	/* Real mode flagged P9 special emu is needed */
1077 	if (local_paca->hmi_p9_special_emu) {
1078 		local_paca->hmi_p9_special_emu = 0;
1079 
1080 		/*
1081 		 * We don't want to take page faults while doing the
1082 		 * emulation, we just replay the instruction if necessary.
1083 		 */
1084 		pagefault_disable();
1085 		p9_hmi_special_emu(regs);
1086 		pagefault_enable();
1087 	}
1088 #endif /* CONFIG_VSX */
1089 
1090 	if (ppc_md.handle_hmi_exception)
1091 		ppc_md.handle_hmi_exception(regs);
1092 
1093 	irq_exit();
1094 	set_irq_regs(old_regs);
1095 }
1096 
1097 void unknown_exception(struct pt_regs *regs)
1098 {
1099 	enum ctx_state prev_state = exception_enter();
1100 
1101 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1102 	       regs->nip, regs->msr, regs->trap);
1103 
1104 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1105 
1106 	exception_exit(prev_state);
1107 }
1108 
1109 void instruction_breakpoint_exception(struct pt_regs *regs)
1110 {
1111 	enum ctx_state prev_state = exception_enter();
1112 
1113 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1114 					5, SIGTRAP) == NOTIFY_STOP)
1115 		goto bail;
1116 	if (debugger_iabr_match(regs))
1117 		goto bail;
1118 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1119 
1120 bail:
1121 	exception_exit(prev_state);
1122 }
1123 
1124 void RunModeException(struct pt_regs *regs)
1125 {
1126 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1127 }
1128 
1129 void single_step_exception(struct pt_regs *regs)
1130 {
1131 	enum ctx_state prev_state = exception_enter();
1132 
1133 	clear_single_step(regs);
1134 	clear_br_trace(regs);
1135 
1136 	if (kprobe_post_handler(regs))
1137 		return;
1138 
1139 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1140 					5, SIGTRAP) == NOTIFY_STOP)
1141 		goto bail;
1142 	if (debugger_sstep(regs))
1143 		goto bail;
1144 
1145 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1146 
1147 bail:
1148 	exception_exit(prev_state);
1149 }
1150 NOKPROBE_SYMBOL(single_step_exception);
1151 
1152 /*
1153  * After we have successfully emulated an instruction, we have to
1154  * check if the instruction was being single-stepped, and if so,
1155  * pretend we got a single-step exception.  This was pointed out
1156  * by Kumar Gala.  -- paulus
1157  */
1158 static void emulate_single_step(struct pt_regs *regs)
1159 {
1160 	if (single_stepping(regs))
1161 		single_step_exception(regs);
1162 }
1163 
1164 static inline int __parse_fpscr(unsigned long fpscr)
1165 {
1166 	int ret = FPE_FLTUNK;
1167 
1168 	/* Invalid operation */
1169 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1170 		ret = FPE_FLTINV;
1171 
1172 	/* Overflow */
1173 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1174 		ret = FPE_FLTOVF;
1175 
1176 	/* Underflow */
1177 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1178 		ret = FPE_FLTUND;
1179 
1180 	/* Divide by zero */
1181 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1182 		ret = FPE_FLTDIV;
1183 
1184 	/* Inexact result */
1185 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1186 		ret = FPE_FLTRES;
1187 
1188 	return ret;
1189 }
1190 
1191 static void parse_fpe(struct pt_regs *regs)
1192 {
1193 	int code = 0;
1194 
1195 	flush_fp_to_thread(current);
1196 
1197 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1198 
1199 	_exception(SIGFPE, regs, code, regs->nip);
1200 }
1201 
1202 /*
1203  * Illegal instruction emulation support.  Originally written to
1204  * provide the PVR to user applications using the mfspr rd, PVR.
1205  * Return non-zero if we can't emulate, or -EFAULT if the associated
1206  * memory access caused an access fault.  Return zero on success.
1207  *
1208  * There are a couple of ways to do this, either "decode" the instruction
1209  * or directly match lots of bits.  In this case, matching lots of
1210  * bits is faster and easier.
1211  *
1212  */
1213 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1214 {
1215 	u8 rT = (instword >> 21) & 0x1f;
1216 	u8 rA = (instword >> 16) & 0x1f;
1217 	u8 NB_RB = (instword >> 11) & 0x1f;
1218 	u32 num_bytes;
1219 	unsigned long EA;
1220 	int pos = 0;
1221 
1222 	/* Early out if we are an invalid form of lswx */
1223 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1224 		if ((rT == rA) || (rT == NB_RB))
1225 			return -EINVAL;
1226 
1227 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1228 
1229 	switch (instword & PPC_INST_STRING_MASK) {
1230 		case PPC_INST_LSWX:
1231 		case PPC_INST_STSWX:
1232 			EA += NB_RB;
1233 			num_bytes = regs->xer & 0x7f;
1234 			break;
1235 		case PPC_INST_LSWI:
1236 		case PPC_INST_STSWI:
1237 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1238 			break;
1239 		default:
1240 			return -EINVAL;
1241 	}
1242 
1243 	while (num_bytes != 0)
1244 	{
1245 		u8 val;
1246 		u32 shift = 8 * (3 - (pos & 0x3));
1247 
1248 		/* if process is 32-bit, clear upper 32 bits of EA */
1249 		if ((regs->msr & MSR_64BIT) == 0)
1250 			EA &= 0xFFFFFFFF;
1251 
1252 		switch ((instword & PPC_INST_STRING_MASK)) {
1253 			case PPC_INST_LSWX:
1254 			case PPC_INST_LSWI:
1255 				if (get_user(val, (u8 __user *)EA))
1256 					return -EFAULT;
1257 				/* first time updating this reg,
1258 				 * zero it out */
1259 				if (pos == 0)
1260 					regs->gpr[rT] = 0;
1261 				regs->gpr[rT] |= val << shift;
1262 				break;
1263 			case PPC_INST_STSWI:
1264 			case PPC_INST_STSWX:
1265 				val = regs->gpr[rT] >> shift;
1266 				if (put_user(val, (u8 __user *)EA))
1267 					return -EFAULT;
1268 				break;
1269 		}
1270 		/* move EA to next address */
1271 		EA += 1;
1272 		num_bytes--;
1273 
1274 		/* manage our position within the register */
1275 		if (++pos == 4) {
1276 			pos = 0;
1277 			if (++rT == 32)
1278 				rT = 0;
1279 		}
1280 	}
1281 
1282 	return 0;
1283 }
1284 
1285 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1286 {
1287 	u32 ra,rs;
1288 	unsigned long tmp;
1289 
1290 	ra = (instword >> 16) & 0x1f;
1291 	rs = (instword >> 21) & 0x1f;
1292 
1293 	tmp = regs->gpr[rs];
1294 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1295 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1296 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1297 	regs->gpr[ra] = tmp;
1298 
1299 	return 0;
1300 }
1301 
1302 static int emulate_isel(struct pt_regs *regs, u32 instword)
1303 {
1304 	u8 rT = (instword >> 21) & 0x1f;
1305 	u8 rA = (instword >> 16) & 0x1f;
1306 	u8 rB = (instword >> 11) & 0x1f;
1307 	u8 BC = (instword >> 6) & 0x1f;
1308 	u8 bit;
1309 	unsigned long tmp;
1310 
1311 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1312 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1313 
1314 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1315 
1316 	return 0;
1317 }
1318 
1319 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1320 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1321 {
1322         /* If we're emulating a load/store in an active transaction, we cannot
1323          * emulate it as the kernel operates in transaction suspended context.
1324          * We need to abort the transaction.  This creates a persistent TM
1325          * abort so tell the user what caused it with a new code.
1326 	 */
1327 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1328 		tm_enable();
1329 		tm_abort(cause);
1330 		return true;
1331 	}
1332 	return false;
1333 }
1334 #else
1335 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1336 {
1337 	return false;
1338 }
1339 #endif
1340 
1341 static int emulate_instruction(struct pt_regs *regs)
1342 {
1343 	u32 instword;
1344 	u32 rd;
1345 
1346 	if (!user_mode(regs))
1347 		return -EINVAL;
1348 	CHECK_FULL_REGS(regs);
1349 
1350 	if (get_user(instword, (u32 __user *)(regs->nip)))
1351 		return -EFAULT;
1352 
1353 	/* Emulate the mfspr rD, PVR. */
1354 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1355 		PPC_WARN_EMULATED(mfpvr, regs);
1356 		rd = (instword >> 21) & 0x1f;
1357 		regs->gpr[rd] = mfspr(SPRN_PVR);
1358 		return 0;
1359 	}
1360 
1361 	/* Emulating the dcba insn is just a no-op.  */
1362 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1363 		PPC_WARN_EMULATED(dcba, regs);
1364 		return 0;
1365 	}
1366 
1367 	/* Emulate the mcrxr insn.  */
1368 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1369 		int shift = (instword >> 21) & 0x1c;
1370 		unsigned long msk = 0xf0000000UL >> shift;
1371 
1372 		PPC_WARN_EMULATED(mcrxr, regs);
1373 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1374 		regs->xer &= ~0xf0000000UL;
1375 		return 0;
1376 	}
1377 
1378 	/* Emulate load/store string insn. */
1379 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1380 		if (tm_abort_check(regs,
1381 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1382 			return -EINVAL;
1383 		PPC_WARN_EMULATED(string, regs);
1384 		return emulate_string_inst(regs, instword);
1385 	}
1386 
1387 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1388 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1389 		PPC_WARN_EMULATED(popcntb, regs);
1390 		return emulate_popcntb_inst(regs, instword);
1391 	}
1392 
1393 	/* Emulate isel (Integer Select) instruction */
1394 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1395 		PPC_WARN_EMULATED(isel, regs);
1396 		return emulate_isel(regs, instword);
1397 	}
1398 
1399 	/* Emulate sync instruction variants */
1400 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1401 		PPC_WARN_EMULATED(sync, regs);
1402 		asm volatile("sync");
1403 		return 0;
1404 	}
1405 
1406 #ifdef CONFIG_PPC64
1407 	/* Emulate the mfspr rD, DSCR. */
1408 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1409 		PPC_INST_MFSPR_DSCR_USER) ||
1410 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1411 		PPC_INST_MFSPR_DSCR)) &&
1412 			cpu_has_feature(CPU_FTR_DSCR)) {
1413 		PPC_WARN_EMULATED(mfdscr, regs);
1414 		rd = (instword >> 21) & 0x1f;
1415 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1416 		return 0;
1417 	}
1418 	/* Emulate the mtspr DSCR, rD. */
1419 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1420 		PPC_INST_MTSPR_DSCR_USER) ||
1421 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1422 		PPC_INST_MTSPR_DSCR)) &&
1423 			cpu_has_feature(CPU_FTR_DSCR)) {
1424 		PPC_WARN_EMULATED(mtdscr, regs);
1425 		rd = (instword >> 21) & 0x1f;
1426 		current->thread.dscr = regs->gpr[rd];
1427 		current->thread.dscr_inherit = 1;
1428 		mtspr(SPRN_DSCR, current->thread.dscr);
1429 		return 0;
1430 	}
1431 #endif
1432 
1433 	return -EINVAL;
1434 }
1435 
1436 int is_valid_bugaddr(unsigned long addr)
1437 {
1438 	return is_kernel_addr(addr);
1439 }
1440 
1441 #ifdef CONFIG_MATH_EMULATION
1442 static int emulate_math(struct pt_regs *regs)
1443 {
1444 	int ret;
1445 	extern int do_mathemu(struct pt_regs *regs);
1446 
1447 	ret = do_mathemu(regs);
1448 	if (ret >= 0)
1449 		PPC_WARN_EMULATED(math, regs);
1450 
1451 	switch (ret) {
1452 	case 0:
1453 		emulate_single_step(regs);
1454 		return 0;
1455 	case 1: {
1456 			int code = 0;
1457 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1458 			_exception(SIGFPE, regs, code, regs->nip);
1459 			return 0;
1460 		}
1461 	case -EFAULT:
1462 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1463 		return 0;
1464 	}
1465 
1466 	return -1;
1467 }
1468 #else
1469 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1470 #endif
1471 
1472 void program_check_exception(struct pt_regs *regs)
1473 {
1474 	enum ctx_state prev_state = exception_enter();
1475 	unsigned int reason = get_reason(regs);
1476 
1477 	/* We can now get here via a FP Unavailable exception if the core
1478 	 * has no FPU, in that case the reason flags will be 0 */
1479 
1480 	if (reason & REASON_FP) {
1481 		/* IEEE FP exception */
1482 		parse_fpe(regs);
1483 		goto bail;
1484 	}
1485 	if (reason & REASON_TRAP) {
1486 		unsigned long bugaddr;
1487 		/* Debugger is first in line to stop recursive faults in
1488 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1489 		if (debugger_bpt(regs))
1490 			goto bail;
1491 
1492 		if (kprobe_handler(regs))
1493 			goto bail;
1494 
1495 		/* trap exception */
1496 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1497 				== NOTIFY_STOP)
1498 			goto bail;
1499 
1500 		bugaddr = regs->nip;
1501 		/*
1502 		 * Fixup bugaddr for BUG_ON() in real mode
1503 		 */
1504 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1505 			bugaddr += PAGE_OFFSET;
1506 
1507 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1508 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1509 			regs->nip += 4;
1510 			goto bail;
1511 		}
1512 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1513 		goto bail;
1514 	}
1515 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1516 	if (reason & REASON_TM) {
1517 		/* This is a TM "Bad Thing Exception" program check.
1518 		 * This occurs when:
1519 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1520 		 *    transition in TM states.
1521 		 * -  A trechkpt is attempted when transactional.
1522 		 * -  A treclaim is attempted when non transactional.
1523 		 * -  A tend is illegally attempted.
1524 		 * -  writing a TM SPR when transactional.
1525 		 *
1526 		 * If usermode caused this, it's done something illegal and
1527 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1528 		 * operand to distinguish from the instruction just being bad
1529 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1530 		 * illegal /placement/ of a valid instruction.
1531 		 */
1532 		if (user_mode(regs)) {
1533 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1534 			goto bail;
1535 		} else {
1536 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1537 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1538 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1539 			die("Unrecoverable exception", regs, SIGABRT);
1540 		}
1541 	}
1542 #endif
1543 
1544 	/*
1545 	 * If we took the program check in the kernel skip down to sending a
1546 	 * SIGILL. The subsequent cases all relate to emulating instructions
1547 	 * which we should only do for userspace. We also do not want to enable
1548 	 * interrupts for kernel faults because that might lead to further
1549 	 * faults, and loose the context of the original exception.
1550 	 */
1551 	if (!user_mode(regs))
1552 		goto sigill;
1553 
1554 	/* We restore the interrupt state now */
1555 	if (!arch_irq_disabled_regs(regs))
1556 		local_irq_enable();
1557 
1558 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1559 	 * but there seems to be a hardware bug on the 405GP (RevD)
1560 	 * that means ESR is sometimes set incorrectly - either to
1561 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1562 	 * hardware people - not sure if it can happen on any illegal
1563 	 * instruction or only on FP instructions, whether there is a
1564 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1565 	 */
1566 	if (!emulate_math(regs))
1567 		goto bail;
1568 
1569 	/* Try to emulate it if we should. */
1570 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1571 		switch (emulate_instruction(regs)) {
1572 		case 0:
1573 			regs->nip += 4;
1574 			emulate_single_step(regs);
1575 			goto bail;
1576 		case -EFAULT:
1577 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1578 			goto bail;
1579 		}
1580 	}
1581 
1582 sigill:
1583 	if (reason & REASON_PRIVILEGED)
1584 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1585 	else
1586 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1587 
1588 bail:
1589 	exception_exit(prev_state);
1590 }
1591 NOKPROBE_SYMBOL(program_check_exception);
1592 
1593 /*
1594  * This occurs when running in hypervisor mode on POWER6 or later
1595  * and an illegal instruction is encountered.
1596  */
1597 void emulation_assist_interrupt(struct pt_regs *regs)
1598 {
1599 	regs->msr |= REASON_ILLEGAL;
1600 	program_check_exception(regs);
1601 }
1602 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1603 
1604 void alignment_exception(struct pt_regs *regs)
1605 {
1606 	enum ctx_state prev_state = exception_enter();
1607 	int sig, code, fixed = 0;
1608 	unsigned long  reason;
1609 
1610 	/* We restore the interrupt state now */
1611 	if (!arch_irq_disabled_regs(regs))
1612 		local_irq_enable();
1613 
1614 	reason = get_reason(regs);
1615 
1616 	if (reason & REASON_BOUNDARY) {
1617 		sig = SIGBUS;
1618 		code = BUS_ADRALN;
1619 		goto bad;
1620 	}
1621 
1622 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1623 		goto bail;
1624 
1625 	/* we don't implement logging of alignment exceptions */
1626 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1627 		fixed = fix_alignment(regs);
1628 
1629 	if (fixed == 1) {
1630 		/* skip over emulated instruction */
1631 		regs->nip += inst_length(reason);
1632 		emulate_single_step(regs);
1633 		goto bail;
1634 	}
1635 
1636 	/* Operand address was bad */
1637 	if (fixed == -EFAULT) {
1638 		sig = SIGSEGV;
1639 		code = SEGV_ACCERR;
1640 	} else {
1641 		sig = SIGBUS;
1642 		code = BUS_ADRALN;
1643 	}
1644 bad:
1645 	if (user_mode(regs))
1646 		_exception(sig, regs, code, regs->dar);
1647 	else
1648 		bad_page_fault(regs, regs->dar, sig);
1649 
1650 bail:
1651 	exception_exit(prev_state);
1652 }
1653 
1654 void StackOverflow(struct pt_regs *regs)
1655 {
1656 	pr_crit("Kernel stack overflow in process %s[%d], r1=%lx\n",
1657 		current->comm, task_pid_nr(current), regs->gpr[1]);
1658 	debugger(regs);
1659 	show_regs(regs);
1660 	panic("kernel stack overflow");
1661 }
1662 
1663 void stack_overflow_exception(struct pt_regs *regs)
1664 {
1665 	enum ctx_state prev_state = exception_enter();
1666 
1667 	die("Kernel stack overflow", regs, SIGSEGV);
1668 
1669 	exception_exit(prev_state);
1670 }
1671 
1672 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1673 {
1674 	enum ctx_state prev_state = exception_enter();
1675 
1676 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1677 			  "%lx at %lx\n", regs->trap, regs->nip);
1678 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1679 
1680 	exception_exit(prev_state);
1681 }
1682 
1683 void altivec_unavailable_exception(struct pt_regs *regs)
1684 {
1685 	enum ctx_state prev_state = exception_enter();
1686 
1687 	if (user_mode(regs)) {
1688 		/* A user program has executed an altivec instruction,
1689 		   but this kernel doesn't support altivec. */
1690 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1691 		goto bail;
1692 	}
1693 
1694 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1695 			"%lx at %lx\n", regs->trap, regs->nip);
1696 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1697 
1698 bail:
1699 	exception_exit(prev_state);
1700 }
1701 
1702 void vsx_unavailable_exception(struct pt_regs *regs)
1703 {
1704 	if (user_mode(regs)) {
1705 		/* A user program has executed an vsx instruction,
1706 		   but this kernel doesn't support vsx. */
1707 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1708 		return;
1709 	}
1710 
1711 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1712 			"%lx at %lx\n", regs->trap, regs->nip);
1713 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1714 }
1715 
1716 #ifdef CONFIG_PPC64
1717 static void tm_unavailable(struct pt_regs *regs)
1718 {
1719 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1720 	if (user_mode(regs)) {
1721 		current->thread.load_tm++;
1722 		regs->msr |= MSR_TM;
1723 		tm_enable();
1724 		tm_restore_sprs(&current->thread);
1725 		return;
1726 	}
1727 #endif
1728 	pr_emerg("Unrecoverable TM Unavailable Exception "
1729 			"%lx at %lx\n", regs->trap, regs->nip);
1730 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1731 }
1732 
1733 void facility_unavailable_exception(struct pt_regs *regs)
1734 {
1735 	static char *facility_strings[] = {
1736 		[FSCR_FP_LG] = "FPU",
1737 		[FSCR_VECVSX_LG] = "VMX/VSX",
1738 		[FSCR_DSCR_LG] = "DSCR",
1739 		[FSCR_PM_LG] = "PMU SPRs",
1740 		[FSCR_BHRB_LG] = "BHRB",
1741 		[FSCR_TM_LG] = "TM",
1742 		[FSCR_EBB_LG] = "EBB",
1743 		[FSCR_TAR_LG] = "TAR",
1744 		[FSCR_MSGP_LG] = "MSGP",
1745 		[FSCR_SCV_LG] = "SCV",
1746 		[FSCR_PREFIX_LG] = "PREFIX",
1747 	};
1748 	char *facility = "unknown";
1749 	u64 value;
1750 	u32 instword, rd;
1751 	u8 status;
1752 	bool hv;
1753 
1754 	hv = (TRAP(regs) == 0xf80);
1755 	if (hv)
1756 		value = mfspr(SPRN_HFSCR);
1757 	else
1758 		value = mfspr(SPRN_FSCR);
1759 
1760 	status = value >> 56;
1761 	if ((hv || status >= 2) &&
1762 	    (status < ARRAY_SIZE(facility_strings)) &&
1763 	    facility_strings[status])
1764 		facility = facility_strings[status];
1765 
1766 	/* We should not have taken this interrupt in kernel */
1767 	if (!user_mode(regs)) {
1768 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1769 			 facility, status, regs->nip);
1770 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1771 	}
1772 
1773 	/* We restore the interrupt state now */
1774 	if (!arch_irq_disabled_regs(regs))
1775 		local_irq_enable();
1776 
1777 	if (status == FSCR_DSCR_LG) {
1778 		/*
1779 		 * User is accessing the DSCR register using the problem
1780 		 * state only SPR number (0x03) either through a mfspr or
1781 		 * a mtspr instruction. If it is a write attempt through
1782 		 * a mtspr, then we set the inherit bit. This also allows
1783 		 * the user to write or read the register directly in the
1784 		 * future by setting via the FSCR DSCR bit. But in case it
1785 		 * is a read DSCR attempt through a mfspr instruction, we
1786 		 * just emulate the instruction instead. This code path will
1787 		 * always emulate all the mfspr instructions till the user
1788 		 * has attempted at least one mtspr instruction. This way it
1789 		 * preserves the same behaviour when the user is accessing
1790 		 * the DSCR through privilege level only SPR number (0x11)
1791 		 * which is emulated through illegal instruction exception.
1792 		 * We always leave HFSCR DSCR set.
1793 		 */
1794 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1795 			pr_err("Failed to fetch the user instruction\n");
1796 			return;
1797 		}
1798 
1799 		/* Write into DSCR (mtspr 0x03, RS) */
1800 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1801 				== PPC_INST_MTSPR_DSCR_USER) {
1802 			rd = (instword >> 21) & 0x1f;
1803 			current->thread.dscr = regs->gpr[rd];
1804 			current->thread.dscr_inherit = 1;
1805 			current->thread.fscr |= FSCR_DSCR;
1806 			mtspr(SPRN_FSCR, current->thread.fscr);
1807 		}
1808 
1809 		/* Read from DSCR (mfspr RT, 0x03) */
1810 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1811 				== PPC_INST_MFSPR_DSCR_USER) {
1812 			if (emulate_instruction(regs)) {
1813 				pr_err("DSCR based mfspr emulation failed\n");
1814 				return;
1815 			}
1816 			regs->nip += 4;
1817 			emulate_single_step(regs);
1818 		}
1819 		return;
1820 	}
1821 
1822 	if (status == FSCR_TM_LG) {
1823 		/*
1824 		 * If we're here then the hardware is TM aware because it
1825 		 * generated an exception with FSRM_TM set.
1826 		 *
1827 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1828 		 * told us not to do TM, or the kernel is not built with TM
1829 		 * support.
1830 		 *
1831 		 * If both of those things are true, then userspace can spam the
1832 		 * console by triggering the printk() below just by continually
1833 		 * doing tbegin (or any TM instruction). So in that case just
1834 		 * send the process a SIGILL immediately.
1835 		 */
1836 		if (!cpu_has_feature(CPU_FTR_TM))
1837 			goto out;
1838 
1839 		tm_unavailable(regs);
1840 		return;
1841 	}
1842 
1843 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1844 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1845 
1846 out:
1847 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1848 }
1849 #endif
1850 
1851 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1852 
1853 void fp_unavailable_tm(struct pt_regs *regs)
1854 {
1855 	/* Note:  This does not handle any kind of FP laziness. */
1856 
1857 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1858 		 regs->nip, regs->msr);
1859 
1860         /* We can only have got here if the task started using FP after
1861          * beginning the transaction.  So, the transactional regs are just a
1862          * copy of the checkpointed ones.  But, we still need to recheckpoint
1863          * as we're enabling FP for the process; it will return, abort the
1864          * transaction, and probably retry but now with FP enabled.  So the
1865          * checkpointed FP registers need to be loaded.
1866 	 */
1867 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1868 
1869 	/*
1870 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1871 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1872 	 *
1873 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1874 	 * recheckpoint.
1875 	 */
1876 
1877 	/* Enable FP for the task: */
1878 	current->thread.load_fp = 1;
1879 
1880 	/*
1881 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1882 	 */
1883 	tm_recheckpoint(&current->thread);
1884 }
1885 
1886 void altivec_unavailable_tm(struct pt_regs *regs)
1887 {
1888 	/* See the comments in fp_unavailable_tm().  This function operates
1889 	 * the same way.
1890 	 */
1891 
1892 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1893 		 "MSR=%lx\n",
1894 		 regs->nip, regs->msr);
1895 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1896 	current->thread.load_vec = 1;
1897 	tm_recheckpoint(&current->thread);
1898 	current->thread.used_vr = 1;
1899 }
1900 
1901 void vsx_unavailable_tm(struct pt_regs *regs)
1902 {
1903 	/* See the comments in fp_unavailable_tm().  This works similarly,
1904 	 * though we're loading both FP and VEC registers in here.
1905 	 *
1906 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1907 	 * regs.  Either way, set MSR_VSX.
1908 	 */
1909 
1910 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1911 		 "MSR=%lx\n",
1912 		 regs->nip, regs->msr);
1913 
1914 	current->thread.used_vsr = 1;
1915 
1916 	/* This reclaims FP and/or VR regs if they're already enabled */
1917 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1918 
1919 	current->thread.load_vec = 1;
1920 	current->thread.load_fp = 1;
1921 
1922 	tm_recheckpoint(&current->thread);
1923 }
1924 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1925 
1926 void performance_monitor_exception(struct pt_regs *regs)
1927 {
1928 	__this_cpu_inc(irq_stat.pmu_irqs);
1929 
1930 	perf_irq(regs);
1931 }
1932 
1933 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1934 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1935 {
1936 	int changed = 0;
1937 	/*
1938 	 * Determine the cause of the debug event, clear the
1939 	 * event flags and send a trap to the handler. Torez
1940 	 */
1941 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1942 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1943 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1944 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1945 #endif
1946 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1947 			     5);
1948 		changed |= 0x01;
1949 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1950 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1951 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1952 			     6);
1953 		changed |= 0x01;
1954 	}  else if (debug_status & DBSR_IAC1) {
1955 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1956 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1957 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1958 			     1);
1959 		changed |= 0x01;
1960 	}  else if (debug_status & DBSR_IAC2) {
1961 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1962 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1963 			     2);
1964 		changed |= 0x01;
1965 	}  else if (debug_status & DBSR_IAC3) {
1966 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1967 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1968 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1969 			     3);
1970 		changed |= 0x01;
1971 	}  else if (debug_status & DBSR_IAC4) {
1972 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1973 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1974 			     4);
1975 		changed |= 0x01;
1976 	}
1977 	/*
1978 	 * At the point this routine was called, the MSR(DE) was turned off.
1979 	 * Check all other debug flags and see if that bit needs to be turned
1980 	 * back on or not.
1981 	 */
1982 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1983 			       current->thread.debug.dbcr1))
1984 		regs->msr |= MSR_DE;
1985 	else
1986 		/* Make sure the IDM flag is off */
1987 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1988 
1989 	if (changed & 0x01)
1990 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1991 }
1992 
1993 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1994 {
1995 	current->thread.debug.dbsr = debug_status;
1996 
1997 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1998 	 * on server, it stops on the target of the branch. In order to simulate
1999 	 * the server behaviour, we thus restart right away with a single step
2000 	 * instead of stopping here when hitting a BT
2001 	 */
2002 	if (debug_status & DBSR_BT) {
2003 		regs->msr &= ~MSR_DE;
2004 
2005 		/* Disable BT */
2006 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
2007 		/* Clear the BT event */
2008 		mtspr(SPRN_DBSR, DBSR_BT);
2009 
2010 		/* Do the single step trick only when coming from userspace */
2011 		if (user_mode(regs)) {
2012 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
2013 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
2014 			regs->msr |= MSR_DE;
2015 			return;
2016 		}
2017 
2018 		if (kprobe_post_handler(regs))
2019 			return;
2020 
2021 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
2022 			       5, SIGTRAP) == NOTIFY_STOP) {
2023 			return;
2024 		}
2025 		if (debugger_sstep(regs))
2026 			return;
2027 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
2028 		regs->msr &= ~MSR_DE;
2029 
2030 		/* Disable instruction completion */
2031 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
2032 		/* Clear the instruction completion event */
2033 		mtspr(SPRN_DBSR, DBSR_IC);
2034 
2035 		if (kprobe_post_handler(regs))
2036 			return;
2037 
2038 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2039 			       5, SIGTRAP) == NOTIFY_STOP) {
2040 			return;
2041 		}
2042 
2043 		if (debugger_sstep(regs))
2044 			return;
2045 
2046 		if (user_mode(regs)) {
2047 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2048 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2049 					       current->thread.debug.dbcr1))
2050 				regs->msr |= MSR_DE;
2051 			else
2052 				/* Make sure the IDM bit is off */
2053 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2054 		}
2055 
2056 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2057 	} else
2058 		handle_debug(regs, debug_status);
2059 }
2060 NOKPROBE_SYMBOL(DebugException);
2061 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2062 
2063 #ifdef CONFIG_ALTIVEC
2064 void altivec_assist_exception(struct pt_regs *regs)
2065 {
2066 	int err;
2067 
2068 	if (!user_mode(regs)) {
2069 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2070 		       " at %lx\n", regs->nip);
2071 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2072 	}
2073 
2074 	flush_altivec_to_thread(current);
2075 
2076 	PPC_WARN_EMULATED(altivec, regs);
2077 	err = emulate_altivec(regs);
2078 	if (err == 0) {
2079 		regs->nip += 4;		/* skip emulated instruction */
2080 		emulate_single_step(regs);
2081 		return;
2082 	}
2083 
2084 	if (err == -EFAULT) {
2085 		/* got an error reading the instruction */
2086 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2087 	} else {
2088 		/* didn't recognize the instruction */
2089 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2090 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2091 				   "in %s at %lx\n", current->comm, regs->nip);
2092 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2093 	}
2094 }
2095 #endif /* CONFIG_ALTIVEC */
2096 
2097 #ifdef CONFIG_FSL_BOOKE
2098 void CacheLockingException(struct pt_regs *regs, unsigned long address,
2099 			   unsigned long error_code)
2100 {
2101 	/* We treat cache locking instructions from the user
2102 	 * as priv ops, in the future we could try to do
2103 	 * something smarter
2104 	 */
2105 	if (error_code & (ESR_DLK|ESR_ILK))
2106 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2107 	return;
2108 }
2109 #endif /* CONFIG_FSL_BOOKE */
2110 
2111 #ifdef CONFIG_SPE
2112 void SPEFloatingPointException(struct pt_regs *regs)
2113 {
2114 	extern int do_spe_mathemu(struct pt_regs *regs);
2115 	unsigned long spefscr;
2116 	int fpexc_mode;
2117 	int code = FPE_FLTUNK;
2118 	int err;
2119 
2120 	/* We restore the interrupt state now */
2121 	if (!arch_irq_disabled_regs(regs))
2122 		local_irq_enable();
2123 
2124 	flush_spe_to_thread(current);
2125 
2126 	spefscr = current->thread.spefscr;
2127 	fpexc_mode = current->thread.fpexc_mode;
2128 
2129 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2130 		code = FPE_FLTOVF;
2131 	}
2132 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2133 		code = FPE_FLTUND;
2134 	}
2135 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2136 		code = FPE_FLTDIV;
2137 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2138 		code = FPE_FLTINV;
2139 	}
2140 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2141 		code = FPE_FLTRES;
2142 
2143 	err = do_spe_mathemu(regs);
2144 	if (err == 0) {
2145 		regs->nip += 4;		/* skip emulated instruction */
2146 		emulate_single_step(regs);
2147 		return;
2148 	}
2149 
2150 	if (err == -EFAULT) {
2151 		/* got an error reading the instruction */
2152 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2153 	} else if (err == -EINVAL) {
2154 		/* didn't recognize the instruction */
2155 		printk(KERN_ERR "unrecognized spe instruction "
2156 		       "in %s at %lx\n", current->comm, regs->nip);
2157 	} else {
2158 		_exception(SIGFPE, regs, code, regs->nip);
2159 	}
2160 
2161 	return;
2162 }
2163 
2164 void SPEFloatingPointRoundException(struct pt_regs *regs)
2165 {
2166 	extern int speround_handler(struct pt_regs *regs);
2167 	int err;
2168 
2169 	/* We restore the interrupt state now */
2170 	if (!arch_irq_disabled_regs(regs))
2171 		local_irq_enable();
2172 
2173 	preempt_disable();
2174 	if (regs->msr & MSR_SPE)
2175 		giveup_spe(current);
2176 	preempt_enable();
2177 
2178 	regs->nip -= 4;
2179 	err = speround_handler(regs);
2180 	if (err == 0) {
2181 		regs->nip += 4;		/* skip emulated instruction */
2182 		emulate_single_step(regs);
2183 		return;
2184 	}
2185 
2186 	if (err == -EFAULT) {
2187 		/* got an error reading the instruction */
2188 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2189 	} else if (err == -EINVAL) {
2190 		/* didn't recognize the instruction */
2191 		printk(KERN_ERR "unrecognized spe instruction "
2192 		       "in %s at %lx\n", current->comm, regs->nip);
2193 	} else {
2194 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2195 		return;
2196 	}
2197 }
2198 #endif
2199 
2200 /*
2201  * We enter here if we get an unrecoverable exception, that is, one
2202  * that happened at a point where the RI (recoverable interrupt) bit
2203  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2204  * we therefore lost state by taking this exception.
2205  */
2206 void unrecoverable_exception(struct pt_regs *regs)
2207 {
2208 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2209 		 regs->trap, regs->nip, regs->msr);
2210 	die("Unrecoverable exception", regs, SIGABRT);
2211 }
2212 NOKPROBE_SYMBOL(unrecoverable_exception);
2213 
2214 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2215 /*
2216  * Default handler for a Watchdog exception,
2217  * spins until a reboot occurs
2218  */
2219 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2220 {
2221 	/* Generic WatchdogHandler, implement your own */
2222 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2223 	return;
2224 }
2225 
2226 void WatchdogException(struct pt_regs *regs)
2227 {
2228 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2229 	WatchdogHandler(regs);
2230 }
2231 #endif
2232 
2233 /*
2234  * We enter here if we discover during exception entry that we are
2235  * running in supervisor mode with a userspace value in the stack pointer.
2236  */
2237 void kernel_bad_stack(struct pt_regs *regs)
2238 {
2239 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2240 	       regs->gpr[1], regs->nip);
2241 	die("Bad kernel stack pointer", regs, SIGABRT);
2242 }
2243 NOKPROBE_SYMBOL(kernel_bad_stack);
2244 
2245 void __init trap_init(void)
2246 {
2247 }
2248 
2249 
2250 #ifdef CONFIG_PPC_EMULATED_STATS
2251 
2252 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2253 
2254 struct ppc_emulated ppc_emulated = {
2255 #ifdef CONFIG_ALTIVEC
2256 	WARN_EMULATED_SETUP(altivec),
2257 #endif
2258 	WARN_EMULATED_SETUP(dcba),
2259 	WARN_EMULATED_SETUP(dcbz),
2260 	WARN_EMULATED_SETUP(fp_pair),
2261 	WARN_EMULATED_SETUP(isel),
2262 	WARN_EMULATED_SETUP(mcrxr),
2263 	WARN_EMULATED_SETUP(mfpvr),
2264 	WARN_EMULATED_SETUP(multiple),
2265 	WARN_EMULATED_SETUP(popcntb),
2266 	WARN_EMULATED_SETUP(spe),
2267 	WARN_EMULATED_SETUP(string),
2268 	WARN_EMULATED_SETUP(sync),
2269 	WARN_EMULATED_SETUP(unaligned),
2270 #ifdef CONFIG_MATH_EMULATION
2271 	WARN_EMULATED_SETUP(math),
2272 #endif
2273 #ifdef CONFIG_VSX
2274 	WARN_EMULATED_SETUP(vsx),
2275 #endif
2276 #ifdef CONFIG_PPC64
2277 	WARN_EMULATED_SETUP(mfdscr),
2278 	WARN_EMULATED_SETUP(mtdscr),
2279 	WARN_EMULATED_SETUP(lq_stq),
2280 	WARN_EMULATED_SETUP(lxvw4x),
2281 	WARN_EMULATED_SETUP(lxvh8x),
2282 	WARN_EMULATED_SETUP(lxvd2x),
2283 	WARN_EMULATED_SETUP(lxvb16x),
2284 #endif
2285 };
2286 
2287 u32 ppc_warn_emulated;
2288 
2289 void ppc_warn_emulated_print(const char *type)
2290 {
2291 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2292 			    type);
2293 }
2294 
2295 static int __init ppc_warn_emulated_init(void)
2296 {
2297 	struct dentry *dir;
2298 	unsigned int i;
2299 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2300 
2301 	dir = debugfs_create_dir("emulated_instructions",
2302 				 powerpc_debugfs_root);
2303 
2304 	debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated);
2305 
2306 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++)
2307 		debugfs_create_u32(entries[i].name, 0644, dir,
2308 				   (u32 *)&entries[i].val.counter);
2309 
2310 	return 0;
2311 }
2312 
2313 device_initcall(ppc_warn_emulated_init);
2314 
2315 #endif /* CONFIG_PPC_EMULATED_STATS */
2316