Avoid beyond bounds copy while caching ACL
[zen-stable.git] / arch / x86 / kernel / traps.c
blob4bbe04d967441196c3c0a3f3390f9e6826f156a0
1 /*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 */
9 /*
10 * Handle hardware traps and faults.
12 #include <linux/interrupt.h>
13 #include <linux/kallsyms.h>
14 #include <linux/spinlock.h>
15 #include <linux/kprobes.h>
16 #include <linux/uaccess.h>
17 #include <linux/kdebug.h>
18 #include <linux/kgdb.h>
19 #include <linux/kernel.h>
20 #include <linux/module.h>
21 #include <linux/ptrace.h>
22 #include <linux/string.h>
23 #include <linux/delay.h>
24 #include <linux/errno.h>
25 #include <linux/kexec.h>
26 #include <linux/sched.h>
27 #include <linux/timer.h>
28 #include <linux/init.h>
29 #include <linux/bug.h>
30 #include <linux/nmi.h>
31 #include <linux/mm.h>
32 #include <linux/smp.h>
33 #include <linux/io.h>
35 #ifdef CONFIG_EISA
36 #include <linux/ioport.h>
37 #include <linux/eisa.h>
38 #endif
40 #ifdef CONFIG_MCA
41 #include <linux/mca.h>
42 #endif
44 #if defined(CONFIG_EDAC)
45 #include <linux/edac.h>
46 #endif
48 #include <asm/kmemcheck.h>
49 #include <asm/stacktrace.h>
50 #include <asm/processor.h>
51 #include <asm/debugreg.h>
52 #include <linux/atomic.h>
53 #include <asm/system.h>
54 #include <asm/traps.h>
55 #include <asm/desc.h>
56 #include <asm/i387.h>
57 #include <asm/mce.h>
59 #include <asm/mach_traps.h>
61 #ifdef CONFIG_X86_64
62 #include <asm/x86_init.h>
63 #include <asm/pgalloc.h>
64 #include <asm/proto.h>
65 #else
66 #include <asm/processor-flags.h>
67 #include <asm/setup.h>
69 asmlinkage int system_call(void);
71 /* Do we ignore FPU interrupts ? */
72 char ignore_fpu_irq;
75 * The IDT has to be page-aligned to simplify the Pentium
76 * F0 0F bug workaround.
78 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
79 #endif
81 DECLARE_BITMAP(used_vectors, NR_VECTORS);
82 EXPORT_SYMBOL_GPL(used_vectors);
84 static inline void conditional_sti(struct pt_regs *regs)
86 if (regs->flags & X86_EFLAGS_IF)
87 local_irq_enable();
90 static inline void preempt_conditional_sti(struct pt_regs *regs)
92 inc_preempt_count();
93 if (regs->flags & X86_EFLAGS_IF)
94 local_irq_enable();
97 static inline void conditional_cli(struct pt_regs *regs)
99 if (regs->flags & X86_EFLAGS_IF)
100 local_irq_disable();
103 static inline void preempt_conditional_cli(struct pt_regs *regs)
105 if (regs->flags & X86_EFLAGS_IF)
106 local_irq_disable();
107 dec_preempt_count();
110 static void __kprobes
111 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
112 long error_code, siginfo_t *info)
114 struct task_struct *tsk = current;
116 #ifdef CONFIG_X86_32
117 if (regs->flags & X86_VM_MASK) {
119 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
120 * On nmi (interrupt 2), do_trap should not be called.
122 if (trapnr < 6)
123 goto vm86_trap;
124 goto trap_signal;
126 #endif
128 if (!user_mode(regs))
129 goto kernel_trap;
131 #ifdef CONFIG_X86_32
132 trap_signal:
133 #endif
135 * We want error_code and trap_no set for userspace faults and
136 * kernelspace faults which result in die(), but not
137 * kernelspace faults which are fixed up. die() gives the
138 * process no chance to handle the signal and notice the
139 * kernel fault information, so that won't result in polluting
140 * the information about previously queued, but not yet
141 * delivered, faults. See also do_general_protection below.
143 tsk->thread.error_code = error_code;
144 tsk->thread.trap_no = trapnr;
146 #ifdef CONFIG_X86_64
147 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
148 printk_ratelimit()) {
149 printk(KERN_INFO
150 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
151 tsk->comm, tsk->pid, str,
152 regs->ip, regs->sp, error_code);
153 print_vma_addr(" in ", regs->ip);
154 printk("\n");
156 #endif
158 if (info)
159 force_sig_info(signr, info, tsk);
160 else
161 force_sig(signr, tsk);
162 return;
164 kernel_trap:
165 if (!fixup_exception(regs)) {
166 tsk->thread.error_code = error_code;
167 tsk->thread.trap_no = trapnr;
168 die(str, regs, error_code);
170 return;
172 #ifdef CONFIG_X86_32
173 vm86_trap:
174 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
175 error_code, trapnr))
176 goto trap_signal;
177 return;
178 #endif
181 #define DO_ERROR(trapnr, signr, str, name) \
182 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
184 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
185 == NOTIFY_STOP) \
186 return; \
187 conditional_sti(regs); \
188 do_trap(trapnr, signr, str, regs, error_code, NULL); \
191 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
192 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
194 siginfo_t info; \
195 info.si_signo = signr; \
196 info.si_errno = 0; \
197 info.si_code = sicode; \
198 info.si_addr = (void __user *)siaddr; \
199 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
200 == NOTIFY_STOP) \
201 return; \
202 conditional_sti(regs); \
203 do_trap(trapnr, signr, str, regs, error_code, &info); \
206 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
207 DO_ERROR(4, SIGSEGV, "overflow", overflow)
208 DO_ERROR(5, SIGSEGV, "bounds", bounds)
209 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
210 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
211 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
212 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
213 #ifdef CONFIG_X86_32
214 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
215 #endif
216 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
218 #ifdef CONFIG_X86_64
219 /* Runs on IST stack */
220 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
222 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
223 12, SIGBUS) == NOTIFY_STOP)
224 return;
225 preempt_conditional_sti(regs);
226 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
227 preempt_conditional_cli(regs);
230 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
232 static const char str[] = "double fault";
233 struct task_struct *tsk = current;
235 /* Return not checked because double check cannot be ignored */
236 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
238 tsk->thread.error_code = error_code;
239 tsk->thread.trap_no = 8;
242 * This is always a kernel trap and never fixable (and thus must
243 * never return).
245 for (;;)
246 die(str, regs, error_code);
248 #endif
250 dotraplinkage void __kprobes
251 do_general_protection(struct pt_regs *regs, long error_code)
253 struct task_struct *tsk;
255 conditional_sti(regs);
257 #ifdef CONFIG_X86_32
258 if (regs->flags & X86_VM_MASK)
259 goto gp_in_vm86;
260 #endif
262 tsk = current;
263 if (!user_mode(regs))
264 goto gp_in_kernel;
266 tsk->thread.error_code = error_code;
267 tsk->thread.trap_no = 13;
269 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
270 printk_ratelimit()) {
271 printk(KERN_INFO
272 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
273 tsk->comm, task_pid_nr(tsk),
274 regs->ip, regs->sp, error_code);
275 print_vma_addr(" in ", regs->ip);
276 printk("\n");
279 force_sig(SIGSEGV, tsk);
280 return;
282 #ifdef CONFIG_X86_32
283 gp_in_vm86:
284 local_irq_enable();
285 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
286 return;
287 #endif
289 gp_in_kernel:
290 if (fixup_exception(regs))
291 return;
293 tsk->thread.error_code = error_code;
294 tsk->thread.trap_no = 13;
295 if (notify_die(DIE_GPF, "general protection fault", regs,
296 error_code, 13, SIGSEGV) == NOTIFY_STOP)
297 return;
298 die("general protection fault", regs, error_code);
301 /* May run on IST stack. */
302 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
304 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
305 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
306 == NOTIFY_STOP)
307 return;
308 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
310 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
311 == NOTIFY_STOP)
312 return;
315 * Let others (NMI) know that the debug stack is in use
316 * as we may switch to the interrupt stack.
318 debug_stack_usage_inc();
319 preempt_conditional_sti(regs);
320 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
321 preempt_conditional_cli(regs);
322 debug_stack_usage_dec();
325 #ifdef CONFIG_X86_64
327 * Help handler running on IST stack to switch back to user stack
328 * for scheduling or signal handling. The actual stack switch is done in
329 * entry.S
331 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
333 struct pt_regs *regs = eregs;
334 /* Did already sync */
335 if (eregs == (struct pt_regs *)eregs->sp)
337 /* Exception from user space */
338 else if (user_mode(eregs))
339 regs = task_pt_regs(current);
341 * Exception from kernel and interrupts are enabled. Move to
342 * kernel process stack.
344 else if (eregs->flags & X86_EFLAGS_IF)
345 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
346 if (eregs != regs)
347 *regs = *eregs;
348 return regs;
350 #endif
353 * Our handling of the processor debug registers is non-trivial.
354 * We do not clear them on entry and exit from the kernel. Therefore
355 * it is possible to get a watchpoint trap here from inside the kernel.
356 * However, the code in ./ptrace.c has ensured that the user can
357 * only set watchpoints on userspace addresses. Therefore the in-kernel
358 * watchpoint trap can only occur in code which is reading/writing
359 * from user space. Such code must not hold kernel locks (since it
360 * can equally take a page fault), therefore it is safe to call
361 * force_sig_info even though that claims and releases locks.
363 * Code in ./signal.c ensures that the debug control register
364 * is restored before we deliver any signal, and therefore that
365 * user code runs with the correct debug control register even though
366 * we clear it here.
368 * Being careful here means that we don't have to be as careful in a
369 * lot of more complicated places (task switching can be a bit lazy
370 * about restoring all the debug state, and ptrace doesn't have to
371 * find every occurrence of the TF bit that could be saved away even
372 * by user code)
374 * May run on IST stack.
376 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
378 struct task_struct *tsk = current;
379 int user_icebp = 0;
380 unsigned long dr6;
381 int si_code;
383 get_debugreg(dr6, 6);
385 /* Filter out all the reserved bits which are preset to 1 */
386 dr6 &= ~DR6_RESERVED;
389 * If dr6 has no reason to give us about the origin of this trap,
390 * then it's very likely the result of an icebp/int01 trap.
391 * User wants a sigtrap for that.
393 if (!dr6 && user_mode(regs))
394 user_icebp = 1;
396 /* Catch kmemcheck conditions first of all! */
397 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
398 return;
400 /* DR6 may or may not be cleared by the CPU */
401 set_debugreg(0, 6);
404 * The processor cleared BTF, so don't mark that we need it set.
406 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
408 /* Store the virtualized DR6 value */
409 tsk->thread.debugreg6 = dr6;
411 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
412 SIGTRAP) == NOTIFY_STOP)
413 return;
416 * Let others (NMI) know that the debug stack is in use
417 * as we may switch to the interrupt stack.
419 debug_stack_usage_inc();
421 /* It's safe to allow irq's after DR6 has been saved */
422 preempt_conditional_sti(regs);
424 if (regs->flags & X86_VM_MASK) {
425 handle_vm86_trap((struct kernel_vm86_regs *) regs,
426 error_code, 1);
427 preempt_conditional_cli(regs);
428 debug_stack_usage_dec();
429 return;
433 * Single-stepping through system calls: ignore any exceptions in
434 * kernel space, but re-enable TF when returning to user mode.
436 * We already checked v86 mode above, so we can check for kernel mode
437 * by just checking the CPL of CS.
439 if ((dr6 & DR_STEP) && !user_mode(regs)) {
440 tsk->thread.debugreg6 &= ~DR_STEP;
441 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
442 regs->flags &= ~X86_EFLAGS_TF;
444 si_code = get_si_code(tsk->thread.debugreg6);
445 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
446 send_sigtrap(tsk, regs, error_code, si_code);
447 preempt_conditional_cli(regs);
448 debug_stack_usage_dec();
450 return;
454 * Note that we play around with the 'TS' bit in an attempt to get
455 * the correct behaviour even in the presence of the asynchronous
456 * IRQ13 behaviour
458 void math_error(struct pt_regs *regs, int error_code, int trapnr)
460 struct task_struct *task = current;
461 siginfo_t info;
462 unsigned short err;
463 char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
465 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
466 return;
467 conditional_sti(regs);
469 if (!user_mode_vm(regs))
471 if (!fixup_exception(regs)) {
472 task->thread.error_code = error_code;
473 task->thread.trap_no = trapnr;
474 die(str, regs, error_code);
476 return;
480 * Save the info for the exception handler and clear the error.
482 save_init_fpu(task);
483 task->thread.trap_no = trapnr;
484 task->thread.error_code = error_code;
485 info.si_signo = SIGFPE;
486 info.si_errno = 0;
487 info.si_addr = (void __user *)regs->ip;
488 if (trapnr == 16) {
489 unsigned short cwd, swd;
491 * (~cwd & swd) will mask out exceptions that are not set to unmasked
492 * status. 0x3f is the exception bits in these regs, 0x200 is the
493 * C1 reg you need in case of a stack fault, 0x040 is the stack
494 * fault bit. We should only be taking one exception at a time,
495 * so if this combination doesn't produce any single exception,
496 * then we have a bad program that isn't synchronizing its FPU usage
497 * and it will suffer the consequences since we won't be able to
498 * fully reproduce the context of the exception
500 cwd = get_fpu_cwd(task);
501 swd = get_fpu_swd(task);
503 err = swd & ~cwd;
504 } else {
506 * The SIMD FPU exceptions are handled a little differently, as there
507 * is only a single status/control register. Thus, to determine which
508 * unmasked exception was caught we must mask the exception mask bits
509 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
511 unsigned short mxcsr = get_fpu_mxcsr(task);
512 err = ~(mxcsr >> 7) & mxcsr;
515 if (err & 0x001) { /* Invalid op */
517 * swd & 0x240 == 0x040: Stack Underflow
518 * swd & 0x240 == 0x240: Stack Overflow
519 * User must clear the SF bit (0x40) if set
521 info.si_code = FPE_FLTINV;
522 } else if (err & 0x004) { /* Divide by Zero */
523 info.si_code = FPE_FLTDIV;
524 } else if (err & 0x008) { /* Overflow */
525 info.si_code = FPE_FLTOVF;
526 } else if (err & 0x012) { /* Denormal, Underflow */
527 info.si_code = FPE_FLTUND;
528 } else if (err & 0x020) { /* Precision */
529 info.si_code = FPE_FLTRES;
530 } else {
532 * If we're using IRQ 13, or supposedly even some trap 16
533 * implementations, it's possible we get a spurious trap...
535 return; /* Spurious trap, no error */
537 force_sig_info(SIGFPE, &info, task);
540 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
542 #ifdef CONFIG_X86_32
543 ignore_fpu_irq = 1;
544 #endif
546 math_error(regs, error_code, 16);
549 dotraplinkage void
550 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
552 math_error(regs, error_code, 19);
555 dotraplinkage void
556 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
558 conditional_sti(regs);
559 #if 0
560 /* No need to warn about this any longer. */
561 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
562 #endif
565 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
569 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
574 * 'math_state_restore()' saves the current math information in the
575 * old math state array, and gets the new ones from the current task
577 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
578 * Don't touch unless you *really* know how it works.
580 * Must be called with kernel preemption disabled (eg with local
581 * local interrupts as in the case of do_device_not_available).
583 void math_state_restore(void)
585 struct task_struct *tsk = current;
587 if (!tsk_used_math(tsk)) {
588 local_irq_enable();
590 * does a slab alloc which can sleep
592 if (init_fpu(tsk)) {
594 * ran out of memory!
596 do_group_exit(SIGKILL);
597 return;
599 local_irq_disable();
602 __thread_fpu_begin(tsk);
604 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
606 if (unlikely(restore_fpu_checking(tsk))) {
607 __thread_fpu_end(tsk);
608 force_sig(SIGSEGV, tsk);
609 return;
612 tsk->fpu_counter++;
614 EXPORT_SYMBOL_GPL(math_state_restore);
616 dotraplinkage void __kprobes
617 do_device_not_available(struct pt_regs *regs, long error_code)
619 #ifdef CONFIG_MATH_EMULATION
620 if (read_cr0() & X86_CR0_EM) {
621 struct math_emu_info info = { };
623 conditional_sti(regs);
625 info.regs = regs;
626 math_emulate(&info);
627 return;
629 #endif
630 math_state_restore(); /* interrupts still off */
631 #ifdef CONFIG_X86_32
632 conditional_sti(regs);
633 #endif
636 #ifdef CONFIG_X86_32
637 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
639 siginfo_t info;
640 local_irq_enable();
642 info.si_signo = SIGILL;
643 info.si_errno = 0;
644 info.si_code = ILL_BADSTK;
645 info.si_addr = NULL;
646 if (notify_die(DIE_TRAP, "iret exception",
647 regs, error_code, 32, SIGILL) == NOTIFY_STOP)
648 return;
649 do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
651 #endif
653 /* Set of traps needed for early debugging. */
654 void __init early_trap_init(void)
656 set_intr_gate_ist(1, &debug, DEBUG_STACK);
657 /* int3 can be called from all */
658 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
659 set_intr_gate(14, &page_fault);
660 load_idt(&idt_descr);
663 void __init trap_init(void)
665 int i;
667 #ifdef CONFIG_EISA
668 void __iomem *p = early_ioremap(0x0FFFD9, 4);
670 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
671 EISA_bus = 1;
672 early_iounmap(p, 4);
673 #endif
675 set_intr_gate(0, &divide_error);
676 set_intr_gate_ist(2, &nmi, NMI_STACK);
677 /* int4 can be called from all */
678 set_system_intr_gate(4, &overflow);
679 set_intr_gate(5, &bounds);
680 set_intr_gate(6, &invalid_op);
681 set_intr_gate(7, &device_not_available);
682 #ifdef CONFIG_X86_32
683 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
684 #else
685 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
686 #endif
687 set_intr_gate(9, &coprocessor_segment_overrun);
688 set_intr_gate(10, &invalid_TSS);
689 set_intr_gate(11, &segment_not_present);
690 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
691 set_intr_gate(13, &general_protection);
692 set_intr_gate(15, &spurious_interrupt_bug);
693 set_intr_gate(16, &coprocessor_error);
694 set_intr_gate(17, &alignment_check);
695 #ifdef CONFIG_X86_MCE
696 set_intr_gate_ist(18, &machine_check, MCE_STACK);
697 #endif
698 set_intr_gate(19, &simd_coprocessor_error);
700 /* Reserve all the builtin and the syscall vector: */
701 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
702 set_bit(i, used_vectors);
704 #ifdef CONFIG_IA32_EMULATION
705 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
706 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
707 #endif
709 #ifdef CONFIG_X86_32
710 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
711 set_bit(SYSCALL_VECTOR, used_vectors);
712 #endif
715 * Should be a barrier for any external CPU state:
717 cpu_init();
719 x86_init.irqs.trap_init();
721 #ifdef CONFIG_X86_64
722 memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
723 set_nmi_gate(1, &debug);
724 set_nmi_gate(3, &int3);
725 #endif