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 #include <linux/debugfs.h>
41
42 #include <asm/emulated_ops.h>
43 #include <linux/uaccess.h>
44 #include <asm/interrupt.h>
45 #include <asm/io.h>
46 #include <asm/machdep.h>
47 #include <asm/rtas.h>
48 #include <asm/pmc.h>
49 #include <asm/reg.h>
50 #ifdef CONFIG_PMAC_BACKLIGHT
51 #include <asm/backlight.h>
52 #endif
53 #ifdef CONFIG_PPC64
54 #include <asm/firmware.h>
55 #include <asm/processor.h>
56 #endif
57 #include <asm/kexec.h>
58 #include <asm/ppc-opcode.h>
59 #include <asm/rio.h>
60 #include <asm/fadump.h>
61 #include <asm/switch_to.h>
62 #include <asm/tm.h>
63 #include <asm/debug.h>
64 #include <asm/asm-prototypes.h>
65 #include <asm/hmi.h>
66 #include <sysdev/fsl_pci.h>
67 #include <asm/kprobes.h>
68 #include <asm/stacktrace.h>
69 #include <asm/nmi.h>
70 #include <asm/disassemble.h>
71
72 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
73 int (*__debugger)(struct pt_regs *regs) __read_mostly;
74 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
75 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
79 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
80
81 EXPORT_SYMBOL(__debugger);
82 EXPORT_SYMBOL(__debugger_ipi);
83 EXPORT_SYMBOL(__debugger_bpt);
84 EXPORT_SYMBOL(__debugger_sstep);
85 EXPORT_SYMBOL(__debugger_iabr_match);
86 EXPORT_SYMBOL(__debugger_break_match);
87 EXPORT_SYMBOL(__debugger_fault_handler);
88 #endif
89
90 /* Transactional Memory trap debug */
91 #ifdef TM_DEBUG_SW
92 #define TM_DEBUG(x...) printk(KERN_INFO x)
93 #else
94 #define TM_DEBUG(x...) do { } while(0)
95 #endif
96
signame(int signr)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
pmac_backlight_unblank(void)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
pmac_backlight_unblank(void)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 */
die_will_crash(void)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
panic_flush_kmsg_start(void)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
panic_flush_kmsg_end(void)172 extern void panic_flush_kmsg_end(void)
173 {
174 kmsg_dump(KMSG_DUMP_PANIC);
175 bust_spinlocks(0);
176 debug_locks_off();
177 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
178 }
179
oops_begin(struct pt_regs * regs)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
oops_end(unsigned long flags,struct pt_regs * regs,int signr)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) == INTERRUPT_SYSTEM_RESET)
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 make_task_dead(signr);
249 }
250 NOKPROBE_SYMBOL(oops_end);
251
get_mmu_str(void)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
__die(const char * str,struct pt_regs * regs,long err)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
die(const char * str,struct pt_regs * regs,long err)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) != INTERRUPT_SYSTEM_RESET) {
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
user_single_step_report(struct pt_regs * regs)304 void user_single_step_report(struct pt_regs *regs)
305 {
306 force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip);
307 }
308
show_signal_msg(int signr,struct pt_regs * regs,int code,unsigned long addr)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
exception_common(int signr,struct pt_regs * regs,int code,unsigned long addr)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 /*
344 * Must not enable interrupts even for user-mode exception, because
345 * this can be called from machine check, which may be a NMI or IRQ
346 * which don't like interrupts being enabled. Could check for
347 * in_hardirq || in_nmi perhaps, but there doesn't seem to be a good
348 * reason why _exception() should enable irqs for an exception handler,
349 * the handlers themselves do that directly.
350 */
351
352 show_signal_msg(signr, regs, code, addr);
353
354 current->thread.trap_nr = code;
355
356 return true;
357 }
358
_exception_pkey(struct pt_regs * regs,unsigned long addr,int key)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
_exception(int signr,struct pt_regs * regs,int code,unsigned long addr)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 */
hv_nmi_check_nonrecoverable(struct pt_regs * regs)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 dropping 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_set_unrecoverable(regs);
437 #endif
438 }
DEFINE_INTERRUPT_HANDLER_NMI(system_reset_exception)439 DEFINE_INTERRUPT_HANDLER_NMI(system_reset_exception)
440 {
441 unsigned long hsrr0, hsrr1;
442 bool saved_hsrrs = false;
443
444 /*
445 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
446 * The system reset interrupt itself may clobber HSRRs (e.g., to call
447 * OPAL), so save them here and restore them before returning.
448 *
449 * Machine checks don't need to save HSRRs, as the real mode handler
450 * is careful to avoid them, and the regular handler is not delivered
451 * as an NMI.
452 */
453 if (cpu_has_feature(CPU_FTR_HVMODE)) {
454 hsrr0 = mfspr(SPRN_HSRR0);
455 hsrr1 = mfspr(SPRN_HSRR1);
456 saved_hsrrs = true;
457 }
458
459 hv_nmi_check_nonrecoverable(regs);
460
461 __this_cpu_inc(irq_stat.sreset_irqs);
462
463 /* See if any machine dependent calls */
464 if (ppc_md.system_reset_exception) {
465 if (ppc_md.system_reset_exception(regs))
466 goto out;
467 }
468
469 if (debugger(regs))
470 goto out;
471
472 kmsg_dump(KMSG_DUMP_OOPS);
473 /*
474 * A system reset is a request to dump, so we always send
475 * it through the crashdump code (if fadump or kdump are
476 * registered).
477 */
478 crash_fadump(regs, "System Reset");
479
480 crash_kexec(regs);
481
482 /*
483 * We aren't the primary crash CPU. We need to send it
484 * to a holding pattern to avoid it ending up in the panic
485 * code.
486 */
487 crash_kexec_secondary(regs);
488
489 /*
490 * No debugger or crash dump registered, print logs then
491 * panic.
492 */
493 die("System Reset", regs, SIGABRT);
494
495 mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
496 add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
497 nmi_panic(regs, "System Reset");
498
499 out:
500 #ifdef CONFIG_PPC_BOOK3S_64
501 BUG_ON(get_paca()->in_nmi == 0);
502 if (get_paca()->in_nmi > 1)
503 die("Unrecoverable nested System Reset", regs, SIGABRT);
504 #endif
505 /* Must die if the interrupt is not recoverable */
506 if (regs_is_unrecoverable(regs)) {
507 /* For the reason explained in die_mce, nmi_exit before die */
508 nmi_exit();
509 die("Unrecoverable System Reset", regs, SIGABRT);
510 }
511
512 if (saved_hsrrs) {
513 mtspr(SPRN_HSRR0, hsrr0);
514 mtspr(SPRN_HSRR1, hsrr1);
515 }
516
517 /* What should we do here? We could issue a shutdown or hard reset. */
518
519 return 0;
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 * -- paulus.
528 */
check_io_access(struct pt_regs * regs)529 static inline int check_io_access(struct pt_regs *regs)
530 {
531 #ifdef CONFIG_PPC32
532 unsigned long msr = regs->msr;
533 const struct exception_table_entry *entry;
534 unsigned int *nip = (unsigned int *)regs->nip;
535
536 if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
537 && (entry = search_exception_tables(regs->nip)) != NULL) {
538 /*
539 * Check that it's a sync instruction, or somewhere
540 * in the twi; isync; nop sequence that inb/inw/inl uses.
541 * As the address is in the exception table
542 * we should be able to read the instr there.
543 * For the debug message, we look at the preceding
544 * load or store.
545 */
546 if (*nip == PPC_RAW_NOP())
547 nip -= 2;
548 else if (*nip == PPC_RAW_ISYNC())
549 --nip;
550 if (*nip == PPC_RAW_SYNC() || get_op(*nip) == OP_TRAP) {
551 unsigned int rb;
552
553 --nip;
554 rb = (*nip >> 11) & 0x1f;
555 printk(KERN_DEBUG "%s bad port %lx at %p\n",
556 (*nip & 0x100)? "OUT to": "IN from",
557 regs->gpr[rb] - _IO_BASE, nip);
558 regs_set_recoverable(regs);
559 regs_set_return_ip(regs, extable_fixup(entry));
560 return 1;
561 }
562 }
563 #endif /* CONFIG_PPC32 */
564 return 0;
565 }
566
567 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
568 /* On 4xx, the reason for the machine check or program exception
569 is in the ESR. */
570 #define get_reason(regs) ((regs)->esr)
571 #define REASON_FP ESR_FP
572 #define REASON_ILLEGAL (ESR_PIL | ESR_PUO)
573 #define REASON_PRIVILEGED ESR_PPR
574 #define REASON_TRAP ESR_PTR
575 #define REASON_PREFIXED 0
576 #define REASON_BOUNDARY 0
577
578 /* single-step stuff */
579 #define single_stepping(regs) (current->thread.debug.dbcr0 & DBCR0_IC)
580 #define clear_single_step(regs) (current->thread.debug.dbcr0 &= ~DBCR0_IC)
581 #define clear_br_trace(regs) do {} while(0)
582 #else
583 /* On non-4xx, the reason for the machine check or program
584 exception is in the MSR. */
585 #define get_reason(regs) ((regs)->msr)
586 #define REASON_TM SRR1_PROGTM
587 #define REASON_FP SRR1_PROGFPE
588 #define REASON_ILLEGAL SRR1_PROGILL
589 #define REASON_PRIVILEGED SRR1_PROGPRIV
590 #define REASON_TRAP SRR1_PROGTRAP
591 #define REASON_PREFIXED SRR1_PREFIXED
592 #define REASON_BOUNDARY SRR1_BOUNDARY
593
594 #define single_stepping(regs) ((regs)->msr & MSR_SE)
595 #define clear_single_step(regs) (regs_set_return_msr((regs), (regs)->msr & ~MSR_SE))
596 #define clear_br_trace(regs) (regs_set_return_msr((regs), (regs)->msr & ~MSR_BE))
597 #endif
598
599 #define inst_length(reason) (((reason) & REASON_PREFIXED) ? 8 : 4)
600
601 #if defined(CONFIG_E500)
machine_check_e500mc(struct pt_regs * regs)602 int machine_check_e500mc(struct pt_regs *regs)
603 {
604 unsigned long mcsr = mfspr(SPRN_MCSR);
605 unsigned long pvr = mfspr(SPRN_PVR);
606 unsigned long reason = mcsr;
607 int recoverable = 1;
608
609 if (reason & MCSR_LD) {
610 recoverable = fsl_rio_mcheck_exception(regs);
611 if (recoverable == 1)
612 goto silent_out;
613 }
614
615 printk("Machine check in kernel mode.\n");
616 printk("Caused by (from MCSR=%lx): ", reason);
617
618 if (reason & MCSR_MCP)
619 pr_cont("Machine Check Signal\n");
620
621 if (reason & MCSR_ICPERR) {
622 pr_cont("Instruction Cache Parity Error\n");
623
624 /*
625 * This is recoverable by invalidating the i-cache.
626 */
627 mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
628 while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
629 ;
630
631 /*
632 * This will generally be accompanied by an instruction
633 * fetch error report -- only treat MCSR_IF as fatal
634 * if it wasn't due to an L1 parity error.
635 */
636 reason &= ~MCSR_IF;
637 }
638
639 if (reason & MCSR_DCPERR_MC) {
640 pr_cont("Data Cache Parity Error\n");
641
642 /*
643 * In write shadow mode we auto-recover from the error, but it
644 * may still get logged and cause a machine check. We should
645 * only treat the non-write shadow case as non-recoverable.
646 */
647 /* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
648 * is not implemented but L1 data cache always runs in write
649 * shadow mode. Hence on data cache parity errors HW will
650 * automatically invalidate the L1 Data Cache.
651 */
652 if (PVR_VER(pvr) != PVR_VER_E6500) {
653 if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
654 recoverable = 0;
655 }
656 }
657
658 if (reason & MCSR_L2MMU_MHIT) {
659 pr_cont("Hit on multiple TLB entries\n");
660 recoverable = 0;
661 }
662
663 if (reason & MCSR_NMI)
664 pr_cont("Non-maskable interrupt\n");
665
666 if (reason & MCSR_IF) {
667 pr_cont("Instruction Fetch Error Report\n");
668 recoverable = 0;
669 }
670
671 if (reason & MCSR_LD) {
672 pr_cont("Load Error Report\n");
673 recoverable = 0;
674 }
675
676 if (reason & MCSR_ST) {
677 pr_cont("Store Error Report\n");
678 recoverable = 0;
679 }
680
681 if (reason & MCSR_LDG) {
682 pr_cont("Guarded Load Error Report\n");
683 recoverable = 0;
684 }
685
686 if (reason & MCSR_TLBSYNC)
687 pr_cont("Simultaneous tlbsync operations\n");
688
689 if (reason & MCSR_BSL2_ERR) {
690 pr_cont("Level 2 Cache Error\n");
691 recoverable = 0;
692 }
693
694 if (reason & MCSR_MAV) {
695 u64 addr;
696
697 addr = mfspr(SPRN_MCAR);
698 addr |= (u64)mfspr(SPRN_MCARU) << 32;
699
700 pr_cont("Machine Check %s Address: %#llx\n",
701 reason & MCSR_MEA ? "Effective" : "Physical", addr);
702 }
703
704 silent_out:
705 mtspr(SPRN_MCSR, mcsr);
706 return mfspr(SPRN_MCSR) == 0 && recoverable;
707 }
708
machine_check_e500(struct pt_regs * regs)709 int machine_check_e500(struct pt_regs *regs)
710 {
711 unsigned long reason = mfspr(SPRN_MCSR);
712
713 if (reason & MCSR_BUS_RBERR) {
714 if (fsl_rio_mcheck_exception(regs))
715 return 1;
716 if (fsl_pci_mcheck_exception(regs))
717 return 1;
718 }
719
720 printk("Machine check in kernel mode.\n");
721 printk("Caused by (from MCSR=%lx): ", reason);
722
723 if (reason & MCSR_MCP)
724 pr_cont("Machine Check Signal\n");
725 if (reason & MCSR_ICPERR)
726 pr_cont("Instruction Cache Parity Error\n");
727 if (reason & MCSR_DCP_PERR)
728 pr_cont("Data Cache Push Parity Error\n");
729 if (reason & MCSR_DCPERR)
730 pr_cont("Data Cache Parity Error\n");
731 if (reason & MCSR_BUS_IAERR)
732 pr_cont("Bus - Instruction Address Error\n");
733 if (reason & MCSR_BUS_RAERR)
734 pr_cont("Bus - Read Address Error\n");
735 if (reason & MCSR_BUS_WAERR)
736 pr_cont("Bus - Write Address Error\n");
737 if (reason & MCSR_BUS_IBERR)
738 pr_cont("Bus - Instruction Data Error\n");
739 if (reason & MCSR_BUS_RBERR)
740 pr_cont("Bus - Read Data Bus Error\n");
741 if (reason & MCSR_BUS_WBERR)
742 pr_cont("Bus - Write Data Bus Error\n");
743 if (reason & MCSR_BUS_IPERR)
744 pr_cont("Bus - Instruction Parity Error\n");
745 if (reason & MCSR_BUS_RPERR)
746 pr_cont("Bus - Read Parity Error\n");
747
748 return 0;
749 }
750
machine_check_generic(struct pt_regs * regs)751 int machine_check_generic(struct pt_regs *regs)
752 {
753 return 0;
754 }
755 #elif defined(CONFIG_PPC32)
machine_check_generic(struct pt_regs * regs)756 int machine_check_generic(struct pt_regs *regs)
757 {
758 unsigned long reason = regs->msr;
759
760 printk("Machine check in kernel mode.\n");
761 printk("Caused by (from SRR1=%lx): ", reason);
762 switch (reason & 0x601F0000) {
763 case 0x80000:
764 pr_cont("Machine check signal\n");
765 break;
766 case 0x40000:
767 case 0x140000: /* 7450 MSS error and TEA */
768 pr_cont("Transfer error ack signal\n");
769 break;
770 case 0x20000:
771 pr_cont("Data parity error signal\n");
772 break;
773 case 0x10000:
774 pr_cont("Address parity error signal\n");
775 break;
776 case 0x20000000:
777 pr_cont("L1 Data Cache error\n");
778 break;
779 case 0x40000000:
780 pr_cont("L1 Instruction Cache error\n");
781 break;
782 case 0x00100000:
783 pr_cont("L2 data cache parity error\n");
784 break;
785 default:
786 pr_cont("Unknown values in msr\n");
787 }
788 return 0;
789 }
790 #endif /* everything else */
791
die_mce(const char * str,struct pt_regs * regs,long err)792 void die_mce(const char *str, struct pt_regs *regs, long err)
793 {
794 /*
795 * The machine check wants to kill the interrupted context,
796 * but make_task_dead() checks for in_interrupt() and panics
797 * in that case, so exit the irq/nmi before calling die.
798 */
799 if (in_nmi())
800 nmi_exit();
801 else
802 irq_exit();
803 die(str, regs, err);
804 }
805
806 /*
807 * BOOK3S_64 does not usually call this handler as a non-maskable interrupt
808 * (it uses its own early real-mode handler to handle the MCE proper
809 * and then raises irq_work to call this handler when interrupts are
810 * enabled). The only time when this is not true is if the early handler
811 * is unrecoverable, then it does call this directly to try to get a
812 * message out.
813 */
__machine_check_exception(struct pt_regs * regs)814 static void __machine_check_exception(struct pt_regs *regs)
815 {
816 int recover = 0;
817
818 __this_cpu_inc(irq_stat.mce_exceptions);
819
820 add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
821
822 /* See if any machine dependent calls. In theory, we would want
823 * to call the CPU first, and call the ppc_md. one if the CPU
824 * one returns a positive number. However there is existing code
825 * that assumes the board gets a first chance, so let's keep it
826 * that way for now and fix things later. --BenH.
827 */
828 if (ppc_md.machine_check_exception)
829 recover = ppc_md.machine_check_exception(regs);
830 else if (cur_cpu_spec->machine_check)
831 recover = cur_cpu_spec->machine_check(regs);
832
833 if (recover > 0)
834 goto bail;
835
836 if (debugger_fault_handler(regs))
837 goto bail;
838
839 if (check_io_access(regs))
840 goto bail;
841
842 die_mce("Machine check", regs, SIGBUS);
843
844 bail:
845 /* Must die if the interrupt is not recoverable */
846 if (regs_is_unrecoverable(regs))
847 die_mce("Unrecoverable Machine check", regs, SIGBUS);
848 }
849
850 #ifdef CONFIG_PPC_BOOK3S_64
DEFINE_INTERRUPT_HANDLER_ASYNC(machine_check_exception_async)851 DEFINE_INTERRUPT_HANDLER_ASYNC(machine_check_exception_async)
852 {
853 __machine_check_exception(regs);
854 }
855 #endif
DEFINE_INTERRUPT_HANDLER_NMI(machine_check_exception)856 DEFINE_INTERRUPT_HANDLER_NMI(machine_check_exception)
857 {
858 __machine_check_exception(regs);
859
860 return 0;
861 }
862
DEFINE_INTERRUPT_HANDLER(SMIException)863 DEFINE_INTERRUPT_HANDLER(SMIException) /* async? */
864 {
865 die("System Management Interrupt", regs, SIGABRT);
866 }
867
868 #ifdef CONFIG_VSX
p9_hmi_special_emu(struct pt_regs * regs)869 static void p9_hmi_special_emu(struct pt_regs *regs)
870 {
871 unsigned int ra, rb, t, i, sel, instr, rc;
872 const void __user *addr;
873 u8 vbuf[16] __aligned(16), *vdst;
874 unsigned long ea, msr, msr_mask;
875 bool swap;
876
877 if (__get_user(instr, (unsigned int __user *)regs->nip))
878 return;
879
880 /*
881 * lxvb16x opcode: 0x7c0006d8
882 * lxvd2x opcode: 0x7c000698
883 * lxvh8x opcode: 0x7c000658
884 * lxvw4x opcode: 0x7c000618
885 */
886 if ((instr & 0xfc00073e) != 0x7c000618) {
887 pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
888 " instr=%08x\n",
889 smp_processor_id(), current->comm, current->pid,
890 regs->nip, instr);
891 return;
892 }
893
894 /* Grab vector registers into the task struct */
895 msr = regs->msr; /* Grab msr before we flush the bits */
896 flush_vsx_to_thread(current);
897 enable_kernel_altivec();
898
899 /*
900 * Is userspace running with a different endian (this is rare but
901 * not impossible)
902 */
903 swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
904
905 /* Decode the instruction */
906 ra = (instr >> 16) & 0x1f;
907 rb = (instr >> 11) & 0x1f;
908 t = (instr >> 21) & 0x1f;
909 if (instr & 1)
910 vdst = (u8 *)¤t->thread.vr_state.vr[t];
911 else
912 vdst = (u8 *)¤t->thread.fp_state.fpr[t][0];
913
914 /* Grab the vector address */
915 ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
916 if (is_32bit_task())
917 ea &= 0xfffffffful;
918 addr = (__force const void __user *)ea;
919
920 /* Check it */
921 if (!access_ok(addr, 16)) {
922 pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
923 " instr=%08x addr=%016lx\n",
924 smp_processor_id(), current->comm, current->pid,
925 regs->nip, instr, (unsigned long)addr);
926 return;
927 }
928
929 /* Read the vector */
930 rc = 0;
931 if ((unsigned long)addr & 0xfUL)
932 /* unaligned case */
933 rc = __copy_from_user_inatomic(vbuf, addr, 16);
934 else
935 __get_user_atomic_128_aligned(vbuf, addr, rc);
936 if (rc) {
937 pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
938 " instr=%08x addr=%016lx\n",
939 smp_processor_id(), current->comm, current->pid,
940 regs->nip, instr, (unsigned long)addr);
941 return;
942 }
943
944 pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
945 " instr=%08x addr=%016lx\n",
946 smp_processor_id(), current->comm, current->pid, regs->nip,
947 instr, (unsigned long) addr);
948
949 /* Grab instruction "selector" */
950 sel = (instr >> 6) & 3;
951
952 /*
953 * Check to make sure the facility is actually enabled. This
954 * could happen if we get a false positive hit.
955 *
956 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
957 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
958 */
959 msr_mask = MSR_VSX;
960 if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
961 msr_mask = MSR_VEC;
962 if (!(msr & msr_mask)) {
963 pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
964 " instr=%08x msr:%016lx\n",
965 smp_processor_id(), current->comm, current->pid,
966 regs->nip, instr, msr);
967 return;
968 }
969
970 /* Do logging here before we modify sel based on endian */
971 switch (sel) {
972 case 0: /* lxvw4x */
973 PPC_WARN_EMULATED(lxvw4x, regs);
974 break;
975 case 1: /* lxvh8x */
976 PPC_WARN_EMULATED(lxvh8x, regs);
977 break;
978 case 2: /* lxvd2x */
979 PPC_WARN_EMULATED(lxvd2x, regs);
980 break;
981 case 3: /* lxvb16x */
982 PPC_WARN_EMULATED(lxvb16x, regs);
983 break;
984 }
985
986 #ifdef __LITTLE_ENDIAN__
987 /*
988 * An LE kernel stores the vector in the task struct as an LE
989 * byte array (effectively swapping both the components and
990 * the content of the components). Those instructions expect
991 * the components to remain in ascending address order, so we
992 * swap them back.
993 *
994 * If we are running a BE user space, the expectation is that
995 * of a simple memcpy, so forcing the emulation to look like
996 * a lxvb16x should do the trick.
997 */
998 if (swap)
999 sel = 3;
1000
1001 switch (sel) {
1002 case 0: /* lxvw4x */
1003 for (i = 0; i < 4; i++)
1004 ((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1005 break;
1006 case 1: /* lxvh8x */
1007 for (i = 0; i < 8; i++)
1008 ((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1009 break;
1010 case 2: /* lxvd2x */
1011 for (i = 0; i < 2; i++)
1012 ((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1013 break;
1014 case 3: /* lxvb16x */
1015 for (i = 0; i < 16; i++)
1016 vdst[i] = vbuf[15-i];
1017 break;
1018 }
1019 #else /* __LITTLE_ENDIAN__ */
1020 /* On a big endian kernel, a BE userspace only needs a memcpy */
1021 if (!swap)
1022 sel = 3;
1023
1024 /* Otherwise, we need to swap the content of the components */
1025 switch (sel) {
1026 case 0: /* lxvw4x */
1027 for (i = 0; i < 4; i++)
1028 ((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1029 break;
1030 case 1: /* lxvh8x */
1031 for (i = 0; i < 8; i++)
1032 ((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1033 break;
1034 case 2: /* lxvd2x */
1035 for (i = 0; i < 2; i++)
1036 ((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1037 break;
1038 case 3: /* lxvb16x */
1039 memcpy(vdst, vbuf, 16);
1040 break;
1041 }
1042 #endif /* !__LITTLE_ENDIAN__ */
1043
1044 /* Go to next instruction */
1045 regs_add_return_ip(regs, 4);
1046 }
1047 #endif /* CONFIG_VSX */
1048
DEFINE_INTERRUPT_HANDLER_ASYNC(handle_hmi_exception)1049 DEFINE_INTERRUPT_HANDLER_ASYNC(handle_hmi_exception)
1050 {
1051 struct pt_regs *old_regs;
1052
1053 old_regs = set_irq_regs(regs);
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 set_irq_regs(old_regs);
1074 }
1075
DEFINE_INTERRUPT_HANDLER(unknown_exception)1076 DEFINE_INTERRUPT_HANDLER(unknown_exception)
1077 {
1078 printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1079 regs->nip, regs->msr, regs->trap);
1080
1081 _exception(SIGTRAP, regs, TRAP_UNK, 0);
1082 }
1083
DEFINE_INTERRUPT_HANDLER_ASYNC(unknown_async_exception)1084 DEFINE_INTERRUPT_HANDLER_ASYNC(unknown_async_exception)
1085 {
1086 printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1087 regs->nip, regs->msr, regs->trap);
1088
1089 _exception(SIGTRAP, regs, TRAP_UNK, 0);
1090 }
1091
DEFINE_INTERRUPT_HANDLER_NMI(unknown_nmi_exception)1092 DEFINE_INTERRUPT_HANDLER_NMI(unknown_nmi_exception)
1093 {
1094 printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1095 regs->nip, regs->msr, regs->trap);
1096
1097 _exception(SIGTRAP, regs, TRAP_UNK, 0);
1098
1099 return 0;
1100 }
1101
DEFINE_INTERRUPT_HANDLER(instruction_breakpoint_exception)1102 DEFINE_INTERRUPT_HANDLER(instruction_breakpoint_exception)
1103 {
1104 if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1105 5, SIGTRAP) == NOTIFY_STOP)
1106 return;
1107 if (debugger_iabr_match(regs))
1108 return;
1109 _exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1110 }
1111
DEFINE_INTERRUPT_HANDLER(RunModeException)1112 DEFINE_INTERRUPT_HANDLER(RunModeException)
1113 {
1114 _exception(SIGTRAP, regs, TRAP_UNK, 0);
1115 }
1116
__single_step_exception(struct pt_regs * regs)1117 static void __single_step_exception(struct pt_regs *regs)
1118 {
1119 clear_single_step(regs);
1120 clear_br_trace(regs);
1121
1122 if (kprobe_post_handler(regs))
1123 return;
1124
1125 if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1126 5, SIGTRAP) == NOTIFY_STOP)
1127 return;
1128 if (debugger_sstep(regs))
1129 return;
1130
1131 _exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1132 }
1133
DEFINE_INTERRUPT_HANDLER(single_step_exception)1134 DEFINE_INTERRUPT_HANDLER(single_step_exception)
1135 {
1136 __single_step_exception(regs);
1137 }
1138
1139 /*
1140 * After we have successfully emulated an instruction, we have to
1141 * check if the instruction was being single-stepped, and if so,
1142 * pretend we got a single-step exception. This was pointed out
1143 * by Kumar Gala. -- paulus
1144 */
emulate_single_step(struct pt_regs * regs)1145 static void emulate_single_step(struct pt_regs *regs)
1146 {
1147 if (single_stepping(regs))
1148 __single_step_exception(regs);
1149 }
1150
1151 #ifdef CONFIG_PPC_FPU_REGS
__parse_fpscr(unsigned long fpscr)1152 static inline int __parse_fpscr(unsigned long fpscr)
1153 {
1154 int ret = FPE_FLTUNK;
1155
1156 /* Invalid operation */
1157 if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1158 ret = FPE_FLTINV;
1159
1160 /* Overflow */
1161 else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1162 ret = FPE_FLTOVF;
1163
1164 /* Underflow */
1165 else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1166 ret = FPE_FLTUND;
1167
1168 /* Divide by zero */
1169 else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1170 ret = FPE_FLTDIV;
1171
1172 /* Inexact result */
1173 else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1174 ret = FPE_FLTRES;
1175
1176 return ret;
1177 }
1178 #endif
1179
parse_fpe(struct pt_regs * regs)1180 static void parse_fpe(struct pt_regs *regs)
1181 {
1182 int code = 0;
1183
1184 flush_fp_to_thread(current);
1185
1186 #ifdef CONFIG_PPC_FPU_REGS
1187 code = __parse_fpscr(current->thread.fp_state.fpscr);
1188 #endif
1189
1190 _exception(SIGFPE, regs, code, regs->nip);
1191 }
1192
1193 /*
1194 * Illegal instruction emulation support. Originally written to
1195 * provide the PVR to user applications using the mfspr rd, PVR.
1196 * Return non-zero if we can't emulate, or -EFAULT if the associated
1197 * memory access caused an access fault. Return zero on success.
1198 *
1199 * There are a couple of ways to do this, either "decode" the instruction
1200 * or directly match lots of bits. In this case, matching lots of
1201 * bits is faster and easier.
1202 *
1203 */
emulate_string_inst(struct pt_regs * regs,u32 instword)1204 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1205 {
1206 u8 rT = (instword >> 21) & 0x1f;
1207 u8 rA = (instword >> 16) & 0x1f;
1208 u8 NB_RB = (instword >> 11) & 0x1f;
1209 u32 num_bytes;
1210 unsigned long EA;
1211 int pos = 0;
1212
1213 /* Early out if we are an invalid form of lswx */
1214 if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1215 if ((rT == rA) || (rT == NB_RB))
1216 return -EINVAL;
1217
1218 EA = (rA == 0) ? 0 : regs->gpr[rA];
1219
1220 switch (instword & PPC_INST_STRING_MASK) {
1221 case PPC_INST_LSWX:
1222 case PPC_INST_STSWX:
1223 EA += NB_RB;
1224 num_bytes = regs->xer & 0x7f;
1225 break;
1226 case PPC_INST_LSWI:
1227 case PPC_INST_STSWI:
1228 num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1229 break;
1230 default:
1231 return -EINVAL;
1232 }
1233
1234 while (num_bytes != 0)
1235 {
1236 u8 val;
1237 u32 shift = 8 * (3 - (pos & 0x3));
1238
1239 /* if process is 32-bit, clear upper 32 bits of EA */
1240 if ((regs->msr & MSR_64BIT) == 0)
1241 EA &= 0xFFFFFFFF;
1242
1243 switch ((instword & PPC_INST_STRING_MASK)) {
1244 case PPC_INST_LSWX:
1245 case PPC_INST_LSWI:
1246 if (get_user(val, (u8 __user *)EA))
1247 return -EFAULT;
1248 /* first time updating this reg,
1249 * zero it out */
1250 if (pos == 0)
1251 regs->gpr[rT] = 0;
1252 regs->gpr[rT] |= val << shift;
1253 break;
1254 case PPC_INST_STSWI:
1255 case PPC_INST_STSWX:
1256 val = regs->gpr[rT] >> shift;
1257 if (put_user(val, (u8 __user *)EA))
1258 return -EFAULT;
1259 break;
1260 }
1261 /* move EA to next address */
1262 EA += 1;
1263 num_bytes--;
1264
1265 /* manage our position within the register */
1266 if (++pos == 4) {
1267 pos = 0;
1268 if (++rT == 32)
1269 rT = 0;
1270 }
1271 }
1272
1273 return 0;
1274 }
1275
emulate_popcntb_inst(struct pt_regs * regs,u32 instword)1276 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1277 {
1278 u32 ra,rs;
1279 unsigned long tmp;
1280
1281 ra = (instword >> 16) & 0x1f;
1282 rs = (instword >> 21) & 0x1f;
1283
1284 tmp = regs->gpr[rs];
1285 tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1286 tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1287 tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1288 regs->gpr[ra] = tmp;
1289
1290 return 0;
1291 }
1292
emulate_isel(struct pt_regs * regs,u32 instword)1293 static int emulate_isel(struct pt_regs *regs, u32 instword)
1294 {
1295 u8 rT = (instword >> 21) & 0x1f;
1296 u8 rA = (instword >> 16) & 0x1f;
1297 u8 rB = (instword >> 11) & 0x1f;
1298 u8 BC = (instword >> 6) & 0x1f;
1299 u8 bit;
1300 unsigned long tmp;
1301
1302 tmp = (rA == 0) ? 0 : regs->gpr[rA];
1303 bit = (regs->ccr >> (31 - BC)) & 0x1;
1304
1305 regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1306
1307 return 0;
1308 }
1309
1310 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
tm_abort_check(struct pt_regs * regs,int cause)1311 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1312 {
1313 /* If we're emulating a load/store in an active transaction, we cannot
1314 * emulate it as the kernel operates in transaction suspended context.
1315 * We need to abort the transaction. This creates a persistent TM
1316 * abort so tell the user what caused it with a new code.
1317 */
1318 if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1319 tm_enable();
1320 tm_abort(cause);
1321 return true;
1322 }
1323 return false;
1324 }
1325 #else
tm_abort_check(struct pt_regs * regs,int reason)1326 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1327 {
1328 return false;
1329 }
1330 #endif
1331
emulate_instruction(struct pt_regs * regs)1332 static int emulate_instruction(struct pt_regs *regs)
1333 {
1334 u32 instword;
1335 u32 rd;
1336
1337 if (!user_mode(regs))
1338 return -EINVAL;
1339
1340 if (get_user(instword, (u32 __user *)(regs->nip)))
1341 return -EFAULT;
1342
1343 /* Emulate the mfspr rD, PVR. */
1344 if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1345 PPC_WARN_EMULATED(mfpvr, regs);
1346 rd = (instword >> 21) & 0x1f;
1347 regs->gpr[rd] = mfspr(SPRN_PVR);
1348 return 0;
1349 }
1350
1351 /* Emulating the dcba insn is just a no-op. */
1352 if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1353 PPC_WARN_EMULATED(dcba, regs);
1354 return 0;
1355 }
1356
1357 /* Emulate the mcrxr insn. */
1358 if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1359 int shift = (instword >> 21) & 0x1c;
1360 unsigned long msk = 0xf0000000UL >> shift;
1361
1362 PPC_WARN_EMULATED(mcrxr, regs);
1363 regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1364 regs->xer &= ~0xf0000000UL;
1365 return 0;
1366 }
1367
1368 /* Emulate load/store string insn. */
1369 if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1370 if (tm_abort_check(regs,
1371 TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1372 return -EINVAL;
1373 PPC_WARN_EMULATED(string, regs);
1374 return emulate_string_inst(regs, instword);
1375 }
1376
1377 /* Emulate the popcntb (Population Count Bytes) instruction. */
1378 if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1379 PPC_WARN_EMULATED(popcntb, regs);
1380 return emulate_popcntb_inst(regs, instword);
1381 }
1382
1383 /* Emulate isel (Integer Select) instruction */
1384 if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1385 PPC_WARN_EMULATED(isel, regs);
1386 return emulate_isel(regs, instword);
1387 }
1388
1389 /* Emulate sync instruction variants */
1390 if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1391 PPC_WARN_EMULATED(sync, regs);
1392 asm volatile("sync");
1393 return 0;
1394 }
1395
1396 #ifdef CONFIG_PPC64
1397 /* Emulate the mfspr rD, DSCR. */
1398 if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1399 PPC_INST_MFSPR_DSCR_USER) ||
1400 ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1401 PPC_INST_MFSPR_DSCR)) &&
1402 cpu_has_feature(CPU_FTR_DSCR)) {
1403 PPC_WARN_EMULATED(mfdscr, regs);
1404 rd = (instword >> 21) & 0x1f;
1405 regs->gpr[rd] = mfspr(SPRN_DSCR);
1406 return 0;
1407 }
1408 /* Emulate the mtspr DSCR, rD. */
1409 if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1410 PPC_INST_MTSPR_DSCR_USER) ||
1411 ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1412 PPC_INST_MTSPR_DSCR)) &&
1413 cpu_has_feature(CPU_FTR_DSCR)) {
1414 PPC_WARN_EMULATED(mtdscr, regs);
1415 rd = (instword >> 21) & 0x1f;
1416 current->thread.dscr = regs->gpr[rd];
1417 current->thread.dscr_inherit = 1;
1418 mtspr(SPRN_DSCR, current->thread.dscr);
1419 return 0;
1420 }
1421 #endif
1422
1423 return -EINVAL;
1424 }
1425
1426 #ifdef CONFIG_GENERIC_BUG
is_valid_bugaddr(unsigned long addr)1427 int is_valid_bugaddr(unsigned long addr)
1428 {
1429 return is_kernel_addr(addr);
1430 }
1431 #endif
1432
1433 #ifdef CONFIG_MATH_EMULATION
emulate_math(struct pt_regs * regs)1434 static int emulate_math(struct pt_regs *regs)
1435 {
1436 int ret;
1437
1438 ret = do_mathemu(regs);
1439 if (ret >= 0)
1440 PPC_WARN_EMULATED(math, regs);
1441
1442 switch (ret) {
1443 case 0:
1444 emulate_single_step(regs);
1445 return 0;
1446 case 1: {
1447 int code = 0;
1448 code = __parse_fpscr(current->thread.fp_state.fpscr);
1449 _exception(SIGFPE, regs, code, regs->nip);
1450 return 0;
1451 }
1452 case -EFAULT:
1453 _exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1454 return 0;
1455 }
1456
1457 return -1;
1458 }
1459 #else
emulate_math(struct pt_regs * regs)1460 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1461 #endif
1462
do_program_check(struct pt_regs * regs)1463 static void do_program_check(struct pt_regs *regs)
1464 {
1465 unsigned int reason = get_reason(regs);
1466
1467 /* We can now get here via a FP Unavailable exception if the core
1468 * has no FPU, in that case the reason flags will be 0 */
1469
1470 if (reason & REASON_FP) {
1471 /* IEEE FP exception */
1472 parse_fpe(regs);
1473 return;
1474 }
1475 if (reason & REASON_TRAP) {
1476 unsigned long bugaddr;
1477 /* Debugger is first in line to stop recursive faults in
1478 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1479 if (debugger_bpt(regs))
1480 return;
1481
1482 if (kprobe_handler(regs))
1483 return;
1484
1485 /* trap exception */
1486 if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1487 == NOTIFY_STOP)
1488 return;
1489
1490 bugaddr = regs->nip;
1491 /*
1492 * Fixup bugaddr for BUG_ON() in real mode
1493 */
1494 if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1495 bugaddr += PAGE_OFFSET;
1496
1497 if (!(regs->msr & MSR_PR) && /* not user-mode */
1498 report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1499 const struct exception_table_entry *entry;
1500
1501 entry = search_exception_tables(bugaddr);
1502 if (entry) {
1503 regs_set_return_ip(regs, extable_fixup(entry) + regs->nip - bugaddr);
1504 return;
1505 }
1506 }
1507 _exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1508 return;
1509 }
1510 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1511 if (reason & REASON_TM) {
1512 /* This is a TM "Bad Thing Exception" program check.
1513 * This occurs when:
1514 * - An rfid/hrfid/mtmsrd attempts to cause an illegal
1515 * transition in TM states.
1516 * - A trechkpt is attempted when transactional.
1517 * - A treclaim is attempted when non transactional.
1518 * - A tend is illegally attempted.
1519 * - writing a TM SPR when transactional.
1520 *
1521 * If usermode caused this, it's done something illegal and
1522 * gets a SIGILL slap on the wrist. We call it an illegal
1523 * operand to distinguish from the instruction just being bad
1524 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1525 * illegal /placement/ of a valid instruction.
1526 */
1527 if (user_mode(regs)) {
1528 _exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1529 return;
1530 } else {
1531 printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1532 "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1533 regs->nip, regs->msr, get_paca()->tm_scratch);
1534 die("Unrecoverable exception", regs, SIGABRT);
1535 }
1536 }
1537 #endif
1538
1539 /*
1540 * If we took the program check in the kernel skip down to sending a
1541 * SIGILL. The subsequent cases all relate to emulating instructions
1542 * which we should only do for userspace. We also do not want to enable
1543 * interrupts for kernel faults because that might lead to further
1544 * faults, and loose the context of the original exception.
1545 */
1546 if (!user_mode(regs))
1547 goto sigill;
1548
1549 interrupt_cond_local_irq_enable(regs);
1550
1551 /* (reason & REASON_ILLEGAL) would be the obvious thing here,
1552 * but there seems to be a hardware bug on the 405GP (RevD)
1553 * that means ESR is sometimes set incorrectly - either to
1554 * ESR_DST (!?) or 0. In the process of chasing this with the
1555 * hardware people - not sure if it can happen on any illegal
1556 * instruction or only on FP instructions, whether there is a
1557 * pattern to occurrences etc. -dgibson 31/Mar/2003
1558 */
1559 if (!emulate_math(regs))
1560 return;
1561
1562 /* Try to emulate it if we should. */
1563 if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1564 switch (emulate_instruction(regs)) {
1565 case 0:
1566 regs_add_return_ip(regs, 4);
1567 emulate_single_step(regs);
1568 return;
1569 case -EFAULT:
1570 _exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1571 return;
1572 }
1573 }
1574
1575 sigill:
1576 if (reason & REASON_PRIVILEGED)
1577 _exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1578 else
1579 _exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1580
1581 }
1582
DEFINE_INTERRUPT_HANDLER(program_check_exception)1583 DEFINE_INTERRUPT_HANDLER(program_check_exception)
1584 {
1585 do_program_check(regs);
1586 }
1587
1588 /*
1589 * This occurs when running in hypervisor mode on POWER6 or later
1590 * and an illegal instruction is encountered.
1591 */
DEFINE_INTERRUPT_HANDLER(emulation_assist_interrupt)1592 DEFINE_INTERRUPT_HANDLER(emulation_assist_interrupt)
1593 {
1594 regs_set_return_msr(regs, regs->msr | REASON_ILLEGAL);
1595 do_program_check(regs);
1596 }
1597
DEFINE_INTERRUPT_HANDLER(alignment_exception)1598 DEFINE_INTERRUPT_HANDLER(alignment_exception)
1599 {
1600 int sig, code, fixed = 0;
1601 unsigned long reason;
1602
1603 interrupt_cond_local_irq_enable(regs);
1604
1605 reason = get_reason(regs);
1606 if (reason & REASON_BOUNDARY) {
1607 sig = SIGBUS;
1608 code = BUS_ADRALN;
1609 goto bad;
1610 }
1611
1612 if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1613 return;
1614
1615 /* we don't implement logging of alignment exceptions */
1616 if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1617 fixed = fix_alignment(regs);
1618
1619 if (fixed == 1) {
1620 /* skip over emulated instruction */
1621 regs_add_return_ip(regs, inst_length(reason));
1622 emulate_single_step(regs);
1623 return;
1624 }
1625
1626 /* Operand address was bad */
1627 if (fixed == -EFAULT) {
1628 sig = SIGSEGV;
1629 code = SEGV_ACCERR;
1630 } else {
1631 sig = SIGBUS;
1632 code = BUS_ADRALN;
1633 }
1634 bad:
1635 if (user_mode(regs))
1636 _exception(sig, regs, code, regs->dar);
1637 else
1638 bad_page_fault(regs, sig);
1639 }
1640
DEFINE_INTERRUPT_HANDLER(stack_overflow_exception)1641 DEFINE_INTERRUPT_HANDLER(stack_overflow_exception)
1642 {
1643 die("Kernel stack overflow", regs, SIGSEGV);
1644 }
1645
DEFINE_INTERRUPT_HANDLER(kernel_fp_unavailable_exception)1646 DEFINE_INTERRUPT_HANDLER(kernel_fp_unavailable_exception)
1647 {
1648 printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1649 "%lx at %lx\n", regs->trap, regs->nip);
1650 die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1651 }
1652
DEFINE_INTERRUPT_HANDLER(altivec_unavailable_exception)1653 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_exception)
1654 {
1655 if (user_mode(regs)) {
1656 /* A user program has executed an altivec instruction,
1657 but this kernel doesn't support altivec. */
1658 _exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1659 return;
1660 }
1661
1662 printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1663 "%lx at %lx\n", regs->trap, regs->nip);
1664 die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1665 }
1666
DEFINE_INTERRUPT_HANDLER(vsx_unavailable_exception)1667 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_exception)
1668 {
1669 if (user_mode(regs)) {
1670 /* A user program has executed an vsx instruction,
1671 but this kernel doesn't support vsx. */
1672 _exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1673 return;
1674 }
1675
1676 printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1677 "%lx at %lx\n", regs->trap, regs->nip);
1678 die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1679 }
1680
1681 #ifdef CONFIG_PPC64
tm_unavailable(struct pt_regs * regs)1682 static void tm_unavailable(struct pt_regs *regs)
1683 {
1684 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1685 if (user_mode(regs)) {
1686 current->thread.load_tm++;
1687 regs_set_return_msr(regs, regs->msr | MSR_TM);
1688 tm_enable();
1689 tm_restore_sprs(¤t->thread);
1690 return;
1691 }
1692 #endif
1693 pr_emerg("Unrecoverable TM Unavailable Exception "
1694 "%lx at %lx\n", regs->trap, regs->nip);
1695 die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1696 }
1697
DEFINE_INTERRUPT_HANDLER(facility_unavailable_exception)1698 DEFINE_INTERRUPT_HANDLER(facility_unavailable_exception)
1699 {
1700 static char *facility_strings[] = {
1701 [FSCR_FP_LG] = "FPU",
1702 [FSCR_VECVSX_LG] = "VMX/VSX",
1703 [FSCR_DSCR_LG] = "DSCR",
1704 [FSCR_PM_LG] = "PMU SPRs",
1705 [FSCR_BHRB_LG] = "BHRB",
1706 [FSCR_TM_LG] = "TM",
1707 [FSCR_EBB_LG] = "EBB",
1708 [FSCR_TAR_LG] = "TAR",
1709 [FSCR_MSGP_LG] = "MSGP",
1710 [FSCR_SCV_LG] = "SCV",
1711 [FSCR_PREFIX_LG] = "PREFIX",
1712 };
1713 char *facility = "unknown";
1714 u64 value;
1715 u32 instword, rd;
1716 u8 status;
1717 bool hv;
1718
1719 hv = (TRAP(regs) == INTERRUPT_H_FAC_UNAVAIL);
1720 if (hv)
1721 value = mfspr(SPRN_HFSCR);
1722 else
1723 value = mfspr(SPRN_FSCR);
1724
1725 status = value >> 56;
1726 if ((hv || status >= 2) &&
1727 (status < ARRAY_SIZE(facility_strings)) &&
1728 facility_strings[status])
1729 facility = facility_strings[status];
1730
1731 /* We should not have taken this interrupt in kernel */
1732 if (!user_mode(regs)) {
1733 pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1734 facility, status, regs->nip);
1735 die("Unexpected facility unavailable exception", regs, SIGABRT);
1736 }
1737
1738 interrupt_cond_local_irq_enable(regs);
1739
1740 if (status == FSCR_DSCR_LG) {
1741 /*
1742 * User is accessing the DSCR register using the problem
1743 * state only SPR number (0x03) either through a mfspr or
1744 * a mtspr instruction. If it is a write attempt through
1745 * a mtspr, then we set the inherit bit. This also allows
1746 * the user to write or read the register directly in the
1747 * future by setting via the FSCR DSCR bit. But in case it
1748 * is a read DSCR attempt through a mfspr instruction, we
1749 * just emulate the instruction instead. This code path will
1750 * always emulate all the mfspr instructions till the user
1751 * has attempted at least one mtspr instruction. This way it
1752 * preserves the same behaviour when the user is accessing
1753 * the DSCR through privilege level only SPR number (0x11)
1754 * which is emulated through illegal instruction exception.
1755 * We always leave HFSCR DSCR set.
1756 */
1757 if (get_user(instword, (u32 __user *)(regs->nip))) {
1758 pr_err("Failed to fetch the user instruction\n");
1759 return;
1760 }
1761
1762 /* Write into DSCR (mtspr 0x03, RS) */
1763 if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1764 == PPC_INST_MTSPR_DSCR_USER) {
1765 rd = (instword >> 21) & 0x1f;
1766 current->thread.dscr = regs->gpr[rd];
1767 current->thread.dscr_inherit = 1;
1768 current->thread.fscr |= FSCR_DSCR;
1769 mtspr(SPRN_FSCR, current->thread.fscr);
1770 }
1771
1772 /* Read from DSCR (mfspr RT, 0x03) */
1773 if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1774 == PPC_INST_MFSPR_DSCR_USER) {
1775 if (emulate_instruction(regs)) {
1776 pr_err("DSCR based mfspr emulation failed\n");
1777 return;
1778 }
1779 regs_add_return_ip(regs, 4);
1780 emulate_single_step(regs);
1781 }
1782 return;
1783 }
1784
1785 if (status == FSCR_TM_LG) {
1786 /*
1787 * If we're here then the hardware is TM aware because it
1788 * generated an exception with FSRM_TM set.
1789 *
1790 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1791 * told us not to do TM, or the kernel is not built with TM
1792 * support.
1793 *
1794 * If both of those things are true, then userspace can spam the
1795 * console by triggering the printk() below just by continually
1796 * doing tbegin (or any TM instruction). So in that case just
1797 * send the process a SIGILL immediately.
1798 */
1799 if (!cpu_has_feature(CPU_FTR_TM))
1800 goto out;
1801
1802 tm_unavailable(regs);
1803 return;
1804 }
1805
1806 pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1807 hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1808
1809 out:
1810 _exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1811 }
1812 #endif
1813
1814 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1815
DEFINE_INTERRUPT_HANDLER(fp_unavailable_tm)1816 DEFINE_INTERRUPT_HANDLER(fp_unavailable_tm)
1817 {
1818 /* Note: This does not handle any kind of FP laziness. */
1819
1820 TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1821 regs->nip, regs->msr);
1822
1823 /* We can only have got here if the task started using FP after
1824 * beginning the transaction. So, the transactional regs are just a
1825 * copy of the checkpointed ones. But, we still need to recheckpoint
1826 * as we're enabling FP for the process; it will return, abort the
1827 * transaction, and probably retry but now with FP enabled. So the
1828 * checkpointed FP registers need to be loaded.
1829 */
1830 tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1831
1832 /*
1833 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1834 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1835 *
1836 * At this point, ck{fp,vr}_state contains the exact values we want to
1837 * recheckpoint.
1838 */
1839
1840 /* Enable FP for the task: */
1841 current->thread.load_fp = 1;
1842
1843 /*
1844 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1845 */
1846 tm_recheckpoint(¤t->thread);
1847 }
1848
DEFINE_INTERRUPT_HANDLER(altivec_unavailable_tm)1849 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_tm)
1850 {
1851 /* See the comments in fp_unavailable_tm(). This function operates
1852 * the same way.
1853 */
1854
1855 TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1856 "MSR=%lx\n",
1857 regs->nip, regs->msr);
1858 tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1859 current->thread.load_vec = 1;
1860 tm_recheckpoint(¤t->thread);
1861 current->thread.used_vr = 1;
1862 }
1863
DEFINE_INTERRUPT_HANDLER(vsx_unavailable_tm)1864 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_tm)
1865 {
1866 /* See the comments in fp_unavailable_tm(). This works similarly,
1867 * though we're loading both FP and VEC registers in here.
1868 *
1869 * If FP isn't in use, load FP regs. If VEC isn't in use, load VEC
1870 * regs. Either way, set MSR_VSX.
1871 */
1872
1873 TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1874 "MSR=%lx\n",
1875 regs->nip, regs->msr);
1876
1877 current->thread.used_vsr = 1;
1878
1879 /* This reclaims FP and/or VR regs if they're already enabled */
1880 tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1881
1882 current->thread.load_vec = 1;
1883 current->thread.load_fp = 1;
1884
1885 tm_recheckpoint(¤t->thread);
1886 }
1887 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1888
1889 #ifdef CONFIG_PPC64
1890 DECLARE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi);
DEFINE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi)1891 DEFINE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi)
1892 {
1893 __this_cpu_inc(irq_stat.pmu_irqs);
1894
1895 perf_irq(regs);
1896
1897 return 0;
1898 }
1899 #endif
1900
1901 DECLARE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async);
DEFINE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async)1902 DEFINE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async)
1903 {
1904 __this_cpu_inc(irq_stat.pmu_irqs);
1905
1906 perf_irq(regs);
1907 }
1908
DEFINE_INTERRUPT_HANDLER_RAW(performance_monitor_exception)1909 DEFINE_INTERRUPT_HANDLER_RAW(performance_monitor_exception)
1910 {
1911 /*
1912 * On 64-bit, if perf interrupts hit in a local_irq_disable
1913 * (soft-masked) region, we consider them as NMIs. This is required to
1914 * prevent hash faults on user addresses when reading callchains (and
1915 * looks better from an irq tracing perspective).
1916 */
1917 if (IS_ENABLED(CONFIG_PPC64) && unlikely(arch_irq_disabled_regs(regs)))
1918 performance_monitor_exception_nmi(regs);
1919 else
1920 performance_monitor_exception_async(regs);
1921
1922 return 0;
1923 }
1924
1925 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
handle_debug(struct pt_regs * regs,unsigned long debug_status)1926 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1927 {
1928 int changed = 0;
1929 /*
1930 * Determine the cause of the debug event, clear the
1931 * event flags and send a trap to the handler. Torez
1932 */
1933 if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1934 dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1935 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1936 current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1937 #endif
1938 do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1939 5);
1940 changed |= 0x01;
1941 } else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1942 dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1943 do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1944 6);
1945 changed |= 0x01;
1946 } else if (debug_status & DBSR_IAC1) {
1947 current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1948 dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1949 do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1950 1);
1951 changed |= 0x01;
1952 } else if (debug_status & DBSR_IAC2) {
1953 current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1954 do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1955 2);
1956 changed |= 0x01;
1957 } else if (debug_status & DBSR_IAC3) {
1958 current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1959 dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1960 do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1961 3);
1962 changed |= 0x01;
1963 } else if (debug_status & DBSR_IAC4) {
1964 current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1965 do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1966 4);
1967 changed |= 0x01;
1968 }
1969 /*
1970 * At the point this routine was called, the MSR(DE) was turned off.
1971 * Check all other debug flags and see if that bit needs to be turned
1972 * back on or not.
1973 */
1974 if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1975 current->thread.debug.dbcr1))
1976 regs_set_return_msr(regs, regs->msr | MSR_DE);
1977 else
1978 /* Make sure the IDM flag is off */
1979 current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1980
1981 if (changed & 0x01)
1982 mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1983 }
1984
DEFINE_INTERRUPT_HANDLER(DebugException)1985 DEFINE_INTERRUPT_HANDLER(DebugException)
1986 {
1987 unsigned long debug_status = regs->dsisr;
1988
1989 current->thread.debug.dbsr = debug_status;
1990
1991 /* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1992 * on server, it stops on the target of the branch. In order to simulate
1993 * the server behaviour, we thus restart right away with a single step
1994 * instead of stopping here when hitting a BT
1995 */
1996 if (debug_status & DBSR_BT) {
1997 regs_set_return_msr(regs, regs->msr & ~MSR_DE);
1998
1999 /* Disable BT */
2000 mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
2001 /* Clear the BT event */
2002 mtspr(SPRN_DBSR, DBSR_BT);
2003
2004 /* Do the single step trick only when coming from userspace */
2005 if (user_mode(regs)) {
2006 current->thread.debug.dbcr0 &= ~DBCR0_BT;
2007 current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
2008 regs_set_return_msr(regs, regs->msr | MSR_DE);
2009 return;
2010 }
2011
2012 if (kprobe_post_handler(regs))
2013 return;
2014
2015 if (notify_die(DIE_SSTEP, "block_step", regs, 5,
2016 5, SIGTRAP) == NOTIFY_STOP) {
2017 return;
2018 }
2019 if (debugger_sstep(regs))
2020 return;
2021 } else if (debug_status & DBSR_IC) { /* Instruction complete */
2022 regs_set_return_msr(regs, regs->msr & ~MSR_DE);
2023
2024 /* Disable instruction completion */
2025 mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
2026 /* Clear the instruction completion event */
2027 mtspr(SPRN_DBSR, DBSR_IC);
2028
2029 if (kprobe_post_handler(regs))
2030 return;
2031
2032 if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2033 5, SIGTRAP) == NOTIFY_STOP) {
2034 return;
2035 }
2036
2037 if (debugger_sstep(regs))
2038 return;
2039
2040 if (user_mode(regs)) {
2041 current->thread.debug.dbcr0 &= ~DBCR0_IC;
2042 if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2043 current->thread.debug.dbcr1))
2044 regs_set_return_msr(regs, regs->msr | MSR_DE);
2045 else
2046 /* Make sure the IDM bit is off */
2047 current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2048 }
2049
2050 _exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2051 } else
2052 handle_debug(regs, debug_status);
2053 }
2054 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2055
2056 #ifdef CONFIG_ALTIVEC
DEFINE_INTERRUPT_HANDLER(altivec_assist_exception)2057 DEFINE_INTERRUPT_HANDLER(altivec_assist_exception)
2058 {
2059 int err;
2060
2061 if (!user_mode(regs)) {
2062 printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2063 " at %lx\n", regs->nip);
2064 die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2065 }
2066
2067 flush_altivec_to_thread(current);
2068
2069 PPC_WARN_EMULATED(altivec, regs);
2070 err = emulate_altivec(regs);
2071 if (err == 0) {
2072 regs_add_return_ip(regs, 4); /* skip emulated instruction */
2073 emulate_single_step(regs);
2074 return;
2075 }
2076
2077 if (err == -EFAULT) {
2078 /* got an error reading the instruction */
2079 _exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2080 } else {
2081 /* didn't recognize the instruction */
2082 /* XXX quick hack for now: set the non-Java bit in the VSCR */
2083 printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2084 "in %s at %lx\n", current->comm, regs->nip);
2085 current->thread.vr_state.vscr.u[3] |= 0x10000;
2086 }
2087 }
2088 #endif /* CONFIG_ALTIVEC */
2089
2090 #ifdef CONFIG_FSL_BOOKE
DEFINE_INTERRUPT_HANDLER(CacheLockingException)2091 DEFINE_INTERRUPT_HANDLER(CacheLockingException)
2092 {
2093 unsigned long error_code = regs->dsisr;
2094
2095 /* We treat cache locking instructions from the user
2096 * as priv ops, in the future we could try to do
2097 * something smarter
2098 */
2099 if (error_code & (ESR_DLK|ESR_ILK))
2100 _exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2101 return;
2102 }
2103 #endif /* CONFIG_FSL_BOOKE */
2104
2105 #ifdef CONFIG_SPE
DEFINE_INTERRUPT_HANDLER(SPEFloatingPointException)2106 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointException)
2107 {
2108 extern int do_spe_mathemu(struct pt_regs *regs);
2109 unsigned long spefscr;
2110 int fpexc_mode;
2111 int code = FPE_FLTUNK;
2112 int err;
2113
2114 interrupt_cond_local_irq_enable(regs);
2115
2116 flush_spe_to_thread(current);
2117
2118 spefscr = current->thread.spefscr;
2119 fpexc_mode = current->thread.fpexc_mode;
2120
2121 if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2122 code = FPE_FLTOVF;
2123 }
2124 else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2125 code = FPE_FLTUND;
2126 }
2127 else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2128 code = FPE_FLTDIV;
2129 else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2130 code = FPE_FLTINV;
2131 }
2132 else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2133 code = FPE_FLTRES;
2134
2135 err = do_spe_mathemu(regs);
2136 if (err == 0) {
2137 regs_add_return_ip(regs, 4); /* skip emulated instruction */
2138 emulate_single_step(regs);
2139 return;
2140 }
2141
2142 if (err == -EFAULT) {
2143 /* got an error reading the instruction */
2144 _exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2145 } else if (err == -EINVAL) {
2146 /* didn't recognize the instruction */
2147 printk(KERN_ERR "unrecognized spe instruction "
2148 "in %s at %lx\n", current->comm, regs->nip);
2149 } else {
2150 _exception(SIGFPE, regs, code, regs->nip);
2151 }
2152
2153 return;
2154 }
2155
DEFINE_INTERRUPT_HANDLER(SPEFloatingPointRoundException)2156 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointRoundException)
2157 {
2158 extern int speround_handler(struct pt_regs *regs);
2159 int err;
2160
2161 interrupt_cond_local_irq_enable(regs);
2162
2163 preempt_disable();
2164 if (regs->msr & MSR_SPE)
2165 giveup_spe(current);
2166 preempt_enable();
2167
2168 regs_add_return_ip(regs, -4);
2169 err = speround_handler(regs);
2170 if (err == 0) {
2171 regs_add_return_ip(regs, 4); /* skip emulated instruction */
2172 emulate_single_step(regs);
2173 return;
2174 }
2175
2176 if (err == -EFAULT) {
2177 /* got an error reading the instruction */
2178 _exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2179 } else if (err == -EINVAL) {
2180 /* didn't recognize the instruction */
2181 printk(KERN_ERR "unrecognized spe instruction "
2182 "in %s at %lx\n", current->comm, regs->nip);
2183 } else {
2184 _exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2185 return;
2186 }
2187 }
2188 #endif
2189
2190 /*
2191 * We enter here if we get an unrecoverable exception, that is, one
2192 * that happened at a point where the RI (recoverable interrupt) bit
2193 * in the MSR is 0. This indicates that SRR0/1 are live, and that
2194 * we therefore lost state by taking this exception.
2195 */
unrecoverable_exception(struct pt_regs * regs)2196 void __noreturn unrecoverable_exception(struct pt_regs *regs)
2197 {
2198 pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2199 regs->trap, regs->nip, regs->msr);
2200 die("Unrecoverable exception", regs, SIGABRT);
2201 /* die() should not return */
2202 for (;;)
2203 ;
2204 }
2205
2206 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2207 /*
2208 * Default handler for a Watchdog exception,
2209 * spins until a reboot occurs
2210 */
WatchdogHandler(struct pt_regs * regs)2211 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2212 {
2213 /* Generic WatchdogHandler, implement your own */
2214 mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2215 return;
2216 }
2217
DEFINE_INTERRUPT_HANDLER_NMI(WatchdogException)2218 DEFINE_INTERRUPT_HANDLER_NMI(WatchdogException)
2219 {
2220 printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2221 WatchdogHandler(regs);
2222 return 0;
2223 }
2224 #endif
2225
2226 /*
2227 * We enter here if we discover during exception entry that we are
2228 * running in supervisor mode with a userspace value in the stack pointer.
2229 */
DEFINE_INTERRUPT_HANDLER(kernel_bad_stack)2230 DEFINE_INTERRUPT_HANDLER(kernel_bad_stack)
2231 {
2232 printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2233 regs->gpr[1], regs->nip);
2234 die("Bad kernel stack pointer", regs, SIGABRT);
2235 }
2236
2237 #ifdef CONFIG_PPC_EMULATED_STATS
2238
2239 #define WARN_EMULATED_SETUP(type) .type = { .name = #type }
2240
2241 struct ppc_emulated ppc_emulated = {
2242 #ifdef CONFIG_ALTIVEC
2243 WARN_EMULATED_SETUP(altivec),
2244 #endif
2245 WARN_EMULATED_SETUP(dcba),
2246 WARN_EMULATED_SETUP(dcbz),
2247 WARN_EMULATED_SETUP(fp_pair),
2248 WARN_EMULATED_SETUP(isel),
2249 WARN_EMULATED_SETUP(mcrxr),
2250 WARN_EMULATED_SETUP(mfpvr),
2251 WARN_EMULATED_SETUP(multiple),
2252 WARN_EMULATED_SETUP(popcntb),
2253 WARN_EMULATED_SETUP(spe),
2254 WARN_EMULATED_SETUP(string),
2255 WARN_EMULATED_SETUP(sync),
2256 WARN_EMULATED_SETUP(unaligned),
2257 #ifdef CONFIG_MATH_EMULATION
2258 WARN_EMULATED_SETUP(math),
2259 #endif
2260 #ifdef CONFIG_VSX
2261 WARN_EMULATED_SETUP(vsx),
2262 #endif
2263 #ifdef CONFIG_PPC64
2264 WARN_EMULATED_SETUP(mfdscr),
2265 WARN_EMULATED_SETUP(mtdscr),
2266 WARN_EMULATED_SETUP(lq_stq),
2267 WARN_EMULATED_SETUP(lxvw4x),
2268 WARN_EMULATED_SETUP(lxvh8x),
2269 WARN_EMULATED_SETUP(lxvd2x),
2270 WARN_EMULATED_SETUP(lxvb16x),
2271 #endif
2272 };
2273
2274 u32 ppc_warn_emulated;
2275
ppc_warn_emulated_print(const char * type)2276 void ppc_warn_emulated_print(const char *type)
2277 {
2278 pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2279 type);
2280 }
2281
ppc_warn_emulated_init(void)2282 static int __init ppc_warn_emulated_init(void)
2283 {
2284 struct dentry *dir;
2285 unsigned int i;
2286 struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2287
2288 dir = debugfs_create_dir("emulated_instructions",
2289 arch_debugfs_dir);
2290
2291 debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated);
2292
2293 for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++)
2294 debugfs_create_u32(entries[i].name, 0644, dir,
2295 (u32 *)&entries[i].val.counter);
2296
2297 return 0;
2298 }
2299
2300 device_initcall(ppc_warn_emulated_init);
2301
2302 #endif /* CONFIG_PPC_EMULATED_STATS */
2303