blob: 4ff5d162ff9fd55381259ff8dd96f84064ecea72 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
Alexander van Heukeluma8c1be92008-07-02 01:29:44 +02003 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 */
8
9/*
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +020010 * Handle hardware traps and faults.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 */
Joe Perchesc767a542012-05-21 19:50:07 -070012
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
Frederic Weisbecker56dd9472013-02-24 00:23:25 +010015#include <linux/context_tracking.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/interrupt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/kallsyms.h>
Ingo Molnarb5964402008-02-26 11:15:50 +010018#include <linux/spinlock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/kprobes.h>
Andrew Morton1e2af922006-09-27 01:51:15 -070020#include <linux/uaccess.h>
Ingo Molnarb5964402008-02-26 11:15:50 +010021#include <linux/kdebug.h>
Jason Wesself503b5a2010-05-20 21:04:25 -050022#include <linux/kgdb.h>
Ingo Molnarb5964402008-02-26 11:15:50 +010023#include <linux/kernel.h>
24#include <linux/module.h>
25#include <linux/ptrace.h>
Oleg Nesterovb02ef202014-05-12 18:24:45 +020026#include <linux/uprobes.h>
Ingo Molnarb5964402008-02-26 11:15:50 +010027#include <linux/string.h>
Ingo Molnarb5964402008-02-26 11:15:50 +010028#include <linux/delay.h>
29#include <linux/errno.h>
30#include <linux/kexec.h>
31#include <linux/sched.h>
32#include <linux/timer.h>
33#include <linux/init.h>
Jeremy Fitzhardinge91768d62006-12-08 02:36:21 -080034#include <linux/bug.h>
Ingo Molnarb5964402008-02-26 11:15:50 +010035#include <linux/nmi.h>
36#include <linux/mm.h>
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +020037#include <linux/smp.h>
38#include <linux/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40#ifdef CONFIG_EISA
41#include <linux/ioport.h>
42#include <linux/eisa.h>
43#endif
44
Dave Jiangc0d12172007-07-19 01:49:46 -070045#if defined(CONFIG_EDAC)
46#include <linux/edac.h>
47#endif
48
Vegard Nossumf8561292008-04-04 00:53:23 +020049#include <asm/kmemcheck.h>
Ingo Molnarb5964402008-02-26 11:15:50 +010050#include <asm/stacktrace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <asm/processor.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <asm/debugreg.h>
Arun Sharma600634972011-07-26 16:09:06 -070053#include <linux/atomic.h>
Steven Rostedt08d636b2011-08-16 09:57:10 -040054#include <asm/ftrace.h>
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +020055#include <asm/traps.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include <asm/desc.h>
57#include <asm/i387.h>
Linus Torvalds1361b832012-02-21 13:19:22 -080058#include <asm/fpu-internal.h>
Hidetoshi Seto9e55e442009-06-15 17:22:15 +090059#include <asm/mce.h>
Kees Cook4eefbe72013-04-10 12:24:22 -070060#include <asm/fixmap.h>
Ingo Molnar1164dd02009-01-28 19:34:09 +010061#include <asm/mach_traps.h>
Jiri Kosina17f41572013-07-23 10:09:28 +020062#include <asm/alternative.h>
Dave Hansenfe3d1972014-11-14 07:18:29 -080063#include <asm/mpx.h>
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +020064
Alexander van Heukelum081f75b2008-10-03 22:00:39 +020065#ifdef CONFIG_X86_64
Thomas Gleixner428cf902009-08-20 10:35:46 +020066#include <asm/x86_init.h>
Alexander van Heukelum081f75b2008-10-03 22:00:39 +020067#include <asm/pgalloc.h>
68#include <asm/proto.h>
Kees Cook4df05f32013-07-16 11:34:41 -070069
70/* No need to be aligned, but done to keep all IDTs defined the same way. */
71gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
Alexander van Heukelum081f75b2008-10-03 22:00:39 +020072#else
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +020073#include <asm/processor-flags.h>
Ingo Molnar8e6dafd2009-02-23 00:34:39 +010074#include <asm/setup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Linus Torvalds1da177e2005-04-16 15:20:36 -070076asmlinkage int system_call(void);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +020077#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Kees Cook4df05f32013-07-16 11:34:41 -070079/* Must be page-aligned because the real IDT is used in a fixmap. */
80gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
81
Yinghai Lub77b8812008-12-19 15:23:44 -080082DECLARE_BITMAP(used_vectors, NR_VECTORS);
83EXPORT_SYMBOL_GPL(used_vectors);
84
Alexander van Heukelum762db432008-09-09 21:55:55 +020085static inline void conditional_sti(struct pt_regs *regs)
86{
87 if (regs->flags & X86_EFLAGS_IF)
88 local_irq_enable();
89}
90
Alexander van Heukelum3d2a71a2008-09-30 18:41:37 +020091static inline void preempt_conditional_sti(struct pt_regs *regs)
92{
Peter Zijlstrabdb43802013-09-10 12:15:23 +020093 preempt_count_inc();
Alexander van Heukelum3d2a71a2008-09-30 18:41:37 +020094 if (regs->flags & X86_EFLAGS_IF)
95 local_irq_enable();
96}
97
Thomas Gleixnerbe716612009-01-13 23:36:34 +010098static inline void conditional_cli(struct pt_regs *regs)
99{
100 if (regs->flags & X86_EFLAGS_IF)
101 local_irq_disable();
102}
103
Alexander van Heukelum3d2a71a2008-09-30 18:41:37 +0200104static inline void preempt_conditional_cli(struct pt_regs *regs)
105{
106 if (regs->flags & X86_EFLAGS_IF)
107 local_irq_disable();
Peter Zijlstrabdb43802013-09-10 12:15:23 +0200108 preempt_count_dec();
Alexander van Heukelum3d2a71a2008-09-30 18:41:37 +0200109}
110
Andy Lutomirski95927472014-11-19 17:41:09 -0800111enum ctx_state ist_enter(struct pt_regs *regs)
112{
Andy Lutomirskib926e6f2015-01-31 04:53:53 -0800113 enum ctx_state prev_state;
Andy Lutomirski95927472014-11-19 17:41:09 -0800114
115 if (user_mode_vm(regs)) {
116 /* Other than that, we're just an exception. */
Andy Lutomirskib926e6f2015-01-31 04:53:53 -0800117 prev_state = exception_enter();
Andy Lutomirski95927472014-11-19 17:41:09 -0800118 } else {
119 /*
120 * We might have interrupted pretty much anything. In
121 * fact, if we're a machine check, we can even interrupt
122 * NMI processing. We don't want in_nmi() to return true,
123 * but we need to notify RCU.
124 */
125 rcu_nmi_enter();
Andy Lutomirskib926e6f2015-01-31 04:53:53 -0800126 prev_state = IN_KERNEL; /* the value is irrelevant. */
Andy Lutomirski95927472014-11-19 17:41:09 -0800127 }
Andy Lutomirskib926e6f2015-01-31 04:53:53 -0800128
129 /*
130 * We are atomic because we're on the IST stack (or we're on x86_32,
131 * in which case we still shouldn't schedule).
132 *
133 * This must be after exception_enter(), because exception_enter()
134 * won't do anything if in_interrupt() returns true.
135 */
136 preempt_count_add(HARDIRQ_OFFSET);
137
138 /* This code is a bit fragile. Test it. */
139 rcu_lockdep_assert(rcu_is_watching(), "ist_enter didn't work");
140
141 return prev_state;
Andy Lutomirski95927472014-11-19 17:41:09 -0800142}
143
144void ist_exit(struct pt_regs *regs, enum ctx_state prev_state)
145{
Andy Lutomirskib926e6f2015-01-31 04:53:53 -0800146 /* Must be before exception_exit. */
Andy Lutomirski95927472014-11-19 17:41:09 -0800147 preempt_count_sub(HARDIRQ_OFFSET);
148
149 if (user_mode_vm(regs))
150 return exception_exit(prev_state);
151 else
152 rcu_nmi_exit();
153}
154
Andy Lutomirskibced35b2014-11-19 17:59:41 -0800155/**
156 * ist_begin_non_atomic() - begin a non-atomic section in an IST exception
157 * @regs: regs passed to the IST exception handler
158 *
159 * IST exception handlers normally cannot schedule. As a special
160 * exception, if the exception interrupted userspace code (i.e.
161 * user_mode_vm(regs) would return true) and the exception was not
162 * a double fault, it can be safe to schedule. ist_begin_non_atomic()
163 * begins a non-atomic section within an ist_enter()/ist_exit() region.
164 * Callers are responsible for enabling interrupts themselves inside
165 * the non-atomic section, and callers must call is_end_non_atomic()
166 * before ist_exit().
167 */
168void ist_begin_non_atomic(struct pt_regs *regs)
169{
170 BUG_ON(!user_mode_vm(regs));
171
172 /*
173 * Sanity check: we need to be on the normal thread stack. This
174 * will catch asm bugs and any attempt to use ist_preempt_enable
175 * from double_fault.
176 */
177 BUG_ON(((current_stack_pointer() ^ this_cpu_read_stable(kernel_stack))
178 & ~(THREAD_SIZE - 1)) != 0);
179
180 preempt_count_sub(HARDIRQ_OFFSET);
181}
182
183/**
184 * ist_end_non_atomic() - begin a non-atomic section in an IST exception
185 *
186 * Ends a non-atomic section started with ist_begin_non_atomic().
187 */
188void ist_end_non_atomic(void)
189{
190 preempt_count_add(HARDIRQ_OFFSET);
191}
192
Masami Hiramatsu93266382014-04-17 17:18:14 +0900193static nokprobe_inline int
Frederic Weisbeckerc416ddf2012-09-25 14:51:19 +0200194do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
195 struct pt_regs *regs, long error_code)
196{
197#ifdef CONFIG_X86_32
198 if (regs->flags & X86_VM_MASK) {
199 /*
200 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
201 * On nmi (interrupt 2), do_trap should not be called.
202 */
203 if (trapnr < X86_TRAP_UD) {
204 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
205 error_code, trapnr))
206 return 0;
207 }
208 return -1;
209 }
210#endif
211 if (!user_mode(regs)) {
212 if (!fixup_exception(regs)) {
213 tsk->thread.error_code = error_code;
214 tsk->thread.trap_nr = trapnr;
215 die(str, regs, error_code);
216 }
217 return 0;
218 }
219
220 return -1;
221}
222
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200223static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
224 siginfo_t *info)
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200225{
226 unsigned long siaddr;
227 int sicode;
228
229 switch (trapnr) {
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200230 default:
231 return SEND_SIG_PRIV;
232
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200233 case X86_TRAP_DE:
234 sicode = FPE_INTDIV;
Oleg Nesterovb02ef202014-05-12 18:24:45 +0200235 siaddr = uprobe_get_trap_addr(regs);
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200236 break;
237 case X86_TRAP_UD:
238 sicode = ILL_ILLOPN;
Oleg Nesterovb02ef202014-05-12 18:24:45 +0200239 siaddr = uprobe_get_trap_addr(regs);
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200240 break;
241 case X86_TRAP_AC:
242 sicode = BUS_ADRALN;
243 siaddr = 0;
244 break;
245 }
246
247 info->si_signo = signr;
248 info->si_errno = 0;
249 info->si_code = sicode;
250 info->si_addr = (void __user *)siaddr;
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200251 return info;
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200252}
253
Masami Hiramatsu93266382014-04-17 17:18:14 +0900254static void
Alexander van Heukelum3c1326f2008-09-26 14:03:08 +0200255do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
Ingo Molnarb5964402008-02-26 11:15:50 +0100256 long error_code, siginfo_t *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
Alexander Nyberg4f339ec2005-06-25 14:58:27 -0700258 struct task_struct *tsk = current;
Alexander Nyberg4f339ec2005-06-25 14:58:27 -0700259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Frederic Weisbeckerc416ddf2012-09-25 14:51:19 +0200261 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
262 return;
Ingo Molnarb5964402008-02-26 11:15:50 +0100263 /*
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530264 * We want error_code and trap_nr set for userspace faults and
Ingo Molnarb5964402008-02-26 11:15:50 +0100265 * kernelspace faults which result in die(), but not
266 * kernelspace faults which are fixed up. die() gives the
267 * process no chance to handle the signal and notice the
268 * kernel fault information, so that won't result in polluting
269 * the information about previously queued, but not yet
270 * delivered, faults. See also do_general_protection below.
271 */
272 tsk->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530273 tsk->thread.trap_nr = trapnr;
Ingo Molnarb5964402008-02-26 11:15:50 +0100274
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200275#ifdef CONFIG_X86_64
276 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
277 printk_ratelimit()) {
Joe Perchesc767a542012-05-21 19:50:07 -0700278 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
279 tsk->comm, tsk->pid, str,
280 regs->ip, regs->sp, error_code);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200281 print_vma_addr(" in ", regs->ip);
Joe Perchesc767a542012-05-21 19:50:07 -0700282 pr_cont("\n");
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200283 }
284#endif
285
Oleg Nesterov38cad572014-05-07 16:47:09 +0200286 force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900288NOKPROBE_SYMBOL(do_trap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
Oleg Nesterovdff07962014-05-07 17:21:34 +0200290static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200291 unsigned long trapnr, int signr)
Oleg Nesterovdff07962014-05-07 17:21:34 +0200292{
293 enum ctx_state prev_state = exception_enter();
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200294 siginfo_t info;
Oleg Nesterovdff07962014-05-07 17:21:34 +0200295
296 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
297 NOTIFY_STOP) {
298 conditional_sti(regs);
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200299 do_trap(trapnr, signr, str, regs, error_code,
300 fill_trap_info(regs, signr, trapnr, &info));
Oleg Nesterovdff07962014-05-07 17:21:34 +0200301 }
302
303 exception_exit(prev_state);
304}
305
Ingo Molnarb5964402008-02-26 11:15:50 +0100306#define DO_ERROR(trapnr, signr, str, name) \
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200307dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
Ingo Molnarb5964402008-02-26 11:15:50 +0100308{ \
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200309 do_error_trap(regs, error_code, str, trapnr, signr); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
311
Oleg Nesterov0eb14832014-05-08 20:12:24 +0200312DO_ERROR(X86_TRAP_DE, SIGFPE, "divide error", divide_error)
313DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
Oleg Nesterov0eb14832014-05-08 20:12:24 +0200314DO_ERROR(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op)
315DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",coprocessor_segment_overrun)
316DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
317DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
Oleg Nesterov0eb14832014-05-08 20:12:24 +0200318DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
Oleg Nesterov0eb14832014-05-08 20:12:24 +0200319DO_ERROR(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200321#ifdef CONFIG_X86_64
322/* Runs on IST stack */
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200323dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
324{
325 static const char str[] = "double fault";
326 struct task_struct *tsk = current;
327
Andy Lutomirskiaf726f22014-11-22 18:00:31 -0800328#ifdef CONFIG_X86_ESPFIX64
329 extern unsigned char native_irq_return_iret[];
330
331 /*
332 * If IRET takes a non-IST fault on the espfix64 stack, then we
333 * end up promoting it to a doublefault. In that case, modify
334 * the stack to make it look like we just entered the #GP
335 * handler from user space, similar to bad_iret.
Andy Lutomirski95927472014-11-19 17:41:09 -0800336 *
337 * No need for ist_enter here because we don't use RCU.
Andy Lutomirskiaf726f22014-11-22 18:00:31 -0800338 */
339 if (((long)regs->sp >> PGDIR_SHIFT) == ESPFIX_PGD_ENTRY &&
340 regs->cs == __KERNEL_CS &&
341 regs->ip == (unsigned long)native_irq_return_iret)
342 {
343 struct pt_regs *normal_regs = task_pt_regs(current);
344
345 /* Fake a #GP(0) from userspace. */
346 memmove(&normal_regs->ip, (void *)regs->sp, 5*8);
347 normal_regs->orig_ax = 0; /* Missing (lost) #GP error code */
348 regs->ip = (unsigned long)general_protection;
349 regs->sp = (unsigned long)&normal_regs->orig_ax;
Andy Lutomirski95927472014-11-19 17:41:09 -0800350
Andy Lutomirskiaf726f22014-11-22 18:00:31 -0800351 return;
352 }
353#endif
354
Andy Lutomirski95927472014-11-19 17:41:09 -0800355 ist_enter(regs); /* Discard prev_state because we won't return. */
Kees Cookc9408262012-03-09 16:07:10 -0800356 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200357
358 tsk->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530359 tsk->thread.trap_nr = X86_TRAP_DF;
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200360
Borislav Petkov4d067d82013-05-09 12:02:29 +0200361#ifdef CONFIG_DOUBLEFAULT
362 df_debug(regs, error_code);
363#endif
Ingo Molnarbd8b96d2008-12-26 09:20:22 +0100364 /*
365 * This is always a kernel trap and never fixable (and thus must
366 * never return).
367 */
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200368 for (;;)
369 die(str, regs, error_code);
370}
371#endif
372
Dave Hansenfe3d1972014-11-14 07:18:29 -0800373dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
374{
375 struct task_struct *tsk = current;
376 struct xsave_struct *xsave_buf;
377 enum ctx_state prev_state;
378 struct bndcsr *bndcsr;
379 siginfo_t *info;
380
381 prev_state = exception_enter();
382 if (notify_die(DIE_TRAP, "bounds", regs, error_code,
383 X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
384 goto exit;
385 conditional_sti(regs);
386
Andy Lutomirski394838c2015-03-09 17:42:31 -0700387 if (!user_mode_vm(regs))
Dave Hansenfe3d1972014-11-14 07:18:29 -0800388 die("bounds", regs, error_code);
389
390 if (!cpu_feature_enabled(X86_FEATURE_MPX)) {
391 /* The exception is not from Intel MPX */
392 goto exit_trap;
393 }
394
395 /*
396 * We need to look at BNDSTATUS to resolve this exception.
397 * It is not directly accessible, though, so we need to
398 * do an xsave and then pull it out of the xsave buffer.
399 */
400 fpu_save_init(&tsk->thread.fpu);
401 xsave_buf = &(tsk->thread.fpu.state->xsave);
402 bndcsr = get_xsave_addr(xsave_buf, XSTATE_BNDCSR);
403 if (!bndcsr)
404 goto exit_trap;
405
406 /*
407 * The error code field of the BNDSTATUS register communicates status
408 * information of a bound range exception #BR or operation involving
409 * bound directory.
410 */
411 switch (bndcsr->bndstatus & MPX_BNDSTA_ERROR_CODE) {
412 case 2: /* Bound directory has invalid entry. */
413 if (mpx_handle_bd_fault(xsave_buf))
414 goto exit_trap;
415 break; /* Success, it was handled */
416 case 1: /* Bound violation. */
417 info = mpx_generate_siginfo(regs, xsave_buf);
Dan Carpentere10abb22014-11-25 20:21:14 +0300418 if (IS_ERR(info)) {
Dave Hansenfe3d1972014-11-14 07:18:29 -0800419 /*
420 * We failed to decode the MPX instruction. Act as if
421 * the exception was not caused by MPX.
422 */
423 goto exit_trap;
424 }
425 /*
426 * Success, we decoded the instruction and retrieved
427 * an 'info' containing the address being accessed
428 * which caused the exception. This information
429 * allows and application to possibly handle the
430 * #BR exception itself.
431 */
432 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, info);
433 kfree(info);
434 break;
435 case 0: /* No exception caused by Intel MPX operations. */
436 goto exit_trap;
437 default:
438 die("bounds", regs, error_code);
439 }
440
441exit:
442 exception_exit(prev_state);
443 return;
444exit_trap:
445 /*
446 * This path out is for all the cases where we could not
447 * handle the exception in some way (like allocating a
448 * table or telling userspace about it. We will also end
449 * up here if the kernel has MPX turned off at compile
450 * time..
451 */
452 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, NULL);
453 exception_exit(prev_state);
454}
455
Masami Hiramatsu93266382014-04-17 17:18:14 +0900456dotraplinkage void
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200457do_general_protection(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200459 struct task_struct *tsk;
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100460 enum ctx_state prev_state;
Ingo Molnarb5964402008-02-26 11:15:50 +0100461
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100462 prev_state = exception_enter();
Alexander van Heukelumc6df0d72008-09-09 21:56:07 +0200463 conditional_sti(regs);
464
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200465#ifdef CONFIG_X86_32
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200466 if (regs->flags & X86_VM_MASK) {
467 local_irq_enable();
468 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200469 goto exit;
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200470 }
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200471#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200473 tsk = current;
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200474 if (!user_mode(regs)) {
475 if (fixup_exception(regs))
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200476 goto exit;
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200477
478 tsk->thread.error_code = error_code;
479 tsk->thread.trap_nr = X86_TRAP_GP;
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200480 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
481 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200482 die("general protection fault", regs, error_code);
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200483 goto exit;
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200486 tsk->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530487 tsk->thread.trap_nr = X86_TRAP_GP;
Ingo Molnarb5964402008-02-26 11:15:50 +0100488
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200489 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
490 printk_ratelimit()) {
Joe Perchesc767a542012-05-21 19:50:07 -0700491 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200492 tsk->comm, task_pid_nr(tsk),
493 regs->ip, regs->sp, error_code);
Andi Kleen03252912008-01-30 13:33:18 +0100494 print_vma_addr(" in ", regs->ip);
Joe Perchesc767a542012-05-21 19:50:07 -0700495 pr_cont("\n");
Andi Kleen03252912008-01-30 13:33:18 +0100496 }
Masoud Asgharifard Sharbianiabd4f752007-07-22 11:12:28 +0200497
Oleg Nesterov38cad572014-05-07 16:47:09 +0200498 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200499exit:
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100500 exception_exit(prev_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900502NOKPROBE_SYMBOL(do_general_protection);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +0200504/* May run on IST stack. */
Masami Hiramatsu93266382014-04-17 17:18:14 +0900505dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506{
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100507 enum ctx_state prev_state;
508
Steven Rostedt08d636b2011-08-16 09:57:10 -0400509#ifdef CONFIG_DYNAMIC_FTRACE
Steven Rostedta192cd02012-05-30 13:26:37 -0400510 /*
511 * ftrace must be first, everything else may cause a recursive crash.
512 * See note by declaration of modifying_ftrace_code in ftrace.c
513 */
514 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
515 ftrace_int3_handler(regs))
Steven Rostedt08d636b2011-08-16 09:57:10 -0400516 return;
517#endif
Jiri Kosina17f41572013-07-23 10:09:28 +0200518 if (poke_int3_handler(regs))
519 return;
520
Andy Lutomirski95927472014-11-19 17:41:09 -0800521 prev_state = ist_enter(regs);
Jason Wesself503b5a2010-05-20 21:04:25 -0500522#ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
Kees Cookc9408262012-03-09 16:07:10 -0800523 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
524 SIGTRAP) == NOTIFY_STOP)
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200525 goto exit;
Jason Wesself503b5a2010-05-20 21:04:25 -0500526#endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
Srikar Dronamrajucc3a1bf2011-10-25 19:51:59 +0530527
Masami Hiramatsu6f6343f2014-04-17 17:17:33 +0900528#ifdef CONFIG_KPROBES
529 if (kprobe_int3_handler(regs))
Masami Hiramatsu4cdf77a2014-06-14 06:47:12 +0000530 goto exit;
Masami Hiramatsu6f6343f2014-04-17 17:17:33 +0900531#endif
532
Kees Cookc9408262012-03-09 16:07:10 -0800533 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
534 SIGTRAP) == NOTIFY_STOP)
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200535 goto exit;
Ingo Molnarb5964402008-02-26 11:15:50 +0100536
Steven Rostedt42181182011-12-16 11:43:02 -0500537 /*
538 * Let others (NMI) know that the debug stack is in use
539 * as we may switch to the interrupt stack.
540 */
541 debug_stack_usage_inc();
Alexander van Heukelum4915a352008-10-03 22:00:34 +0200542 preempt_conditional_sti(regs);
Kees Cookc9408262012-03-09 16:07:10 -0800543 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
Alexander van Heukelum4915a352008-10-03 22:00:34 +0200544 preempt_conditional_cli(regs);
Steven Rostedt42181182011-12-16 11:43:02 -0500545 debug_stack_usage_dec();
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200546exit:
Andy Lutomirski95927472014-11-19 17:41:09 -0800547 ist_exit(regs, prev_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900549NOKPROBE_SYMBOL(do_int3);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200551#ifdef CONFIG_X86_64
Ingo Molnarbd8b96d2008-12-26 09:20:22 +0100552/*
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800553 * Help handler running on IST stack to switch off the IST stack if the
554 * interrupted code was in user mode. The actual stack switch is done in
555 * entry_64.S
Ingo Molnarbd8b96d2008-12-26 09:20:22 +0100556 */
Andy Lutomirski7ddc6a22014-11-24 17:39:06 -0800557asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200558{
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800559 struct pt_regs *regs = task_pt_regs(current);
560 *regs = *eregs;
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200561 return regs;
562}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900563NOKPROBE_SYMBOL(sync_regs);
Andy Lutomirskib645af22014-11-22 18:00:33 -0800564
565struct bad_iret_stack {
566 void *error_entry_ret;
567 struct pt_regs regs;
568};
569
Andy Lutomirski7ddc6a22014-11-24 17:39:06 -0800570asmlinkage __visible notrace
Andy Lutomirskib645af22014-11-22 18:00:33 -0800571struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
572{
573 /*
574 * This is called from entry_64.S early in handling a fault
575 * caused by a bad iret to user mode. To handle the fault
576 * correctly, we want move our stack frame to task_pt_regs
577 * and we want to pretend that the exception came from the
578 * iret target.
579 */
580 struct bad_iret_stack *new_stack =
581 container_of(task_pt_regs(current),
582 struct bad_iret_stack, regs);
583
584 /* Copy the IRET target to the new stack. */
585 memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
586
587 /* Copy the remainder of the stack from the current stack. */
588 memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
589
590 BUG_ON(!user_mode_vm(&new_stack->regs));
591 return new_stack;
592}
Andy Lutomirski7ddc6a22014-11-24 17:39:06 -0800593NOKPROBE_SYMBOL(fixup_bad_iret);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200594#endif
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596/*
597 * Our handling of the processor debug registers is non-trivial.
598 * We do not clear them on entry and exit from the kernel. Therefore
599 * it is possible to get a watchpoint trap here from inside the kernel.
600 * However, the code in ./ptrace.c has ensured that the user can
601 * only set watchpoints on userspace addresses. Therefore the in-kernel
602 * watchpoint trap can only occur in code which is reading/writing
603 * from user space. Such code must not hold kernel locks (since it
604 * can equally take a page fault), therefore it is safe to call
605 * force_sig_info even though that claims and releases locks.
Ingo Molnarb5964402008-02-26 11:15:50 +0100606 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 * Code in ./signal.c ensures that the debug control register
608 * is restored before we deliver any signal, and therefore that
609 * user code runs with the correct debug control register even though
610 * we clear it here.
611 *
612 * Being careful here means that we don't have to be as careful in a
613 * lot of more complicated places (task switching can be a bit lazy
614 * about restoring all the debug state, and ptrace doesn't have to
615 * find every occurrence of the TF bit that could be saved away even
616 * by user code)
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +0200617 *
618 * May run on IST stack.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 */
Masami Hiramatsu93266382014-04-17 17:18:14 +0900620dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 struct task_struct *tsk = current;
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100623 enum ctx_state prev_state;
Frederic Weisbeckera1e80fa2010-06-30 15:09:06 +0200624 int user_icebp = 0;
K.Prasad08d68322009-06-01 23:44:08 +0530625 unsigned long dr6;
Srinivasa Dsda654b72008-09-23 15:23:52 +0530626 int si_code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Andy Lutomirski95927472014-11-19 17:41:09 -0800628 prev_state = ist_enter(regs);
Masami Hiramatsu4cdf77a2014-06-14 06:47:12 +0000629
K.Prasad08d68322009-06-01 23:44:08 +0530630 get_debugreg(dr6, 6);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
K.Prasad40f92492010-01-28 16:44:01 +0530632 /* Filter out all the reserved bits which are preset to 1 */
633 dr6 &= ~DR6_RESERVED;
634
Frederic Weisbeckera1e80fa2010-06-30 15:09:06 +0200635 /*
636 * If dr6 has no reason to give us about the origin of this trap,
637 * then it's very likely the result of an icebp/int01 trap.
638 * User wants a sigtrap for that.
639 */
Andy Lutomirski394838c2015-03-09 17:42:31 -0700640 if (!dr6 && user_mode_vm(regs))
Frederic Weisbeckera1e80fa2010-06-30 15:09:06 +0200641 user_icebp = 1;
642
Vegard Nossumf8561292008-04-04 00:53:23 +0200643 /* Catch kmemcheck conditions first of all! */
Ingo Molnareadb8a02009-06-17 12:52:15 +0200644 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200645 goto exit;
Vegard Nossumf8561292008-04-04 00:53:23 +0200646
K.Prasad08d68322009-06-01 23:44:08 +0530647 /* DR6 may or may not be cleared by the CPU */
648 set_debugreg(0, 6);
Roland McGrath10faa812008-01-30 13:30:54 +0100649
Peter Zijlstraea8e61b2010-03-25 14:51:51 +0100650 /*
651 * The processor cleared BTF, so don't mark that we need it set.
652 */
653 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
654
K.Prasad08d68322009-06-01 23:44:08 +0530655 /* Store the virtualized DR6 value */
656 tsk->thread.debugreg6 = dr6;
657
Masami Hiramatsu6f6343f2014-04-17 17:17:33 +0900658#ifdef CONFIG_KPROBES
659 if (kprobe_debug_handler(regs))
660 goto exit;
661#endif
662
Rusty Russell5a802e12013-06-16 14:12:47 +0930663 if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
K.Prasad62edab92009-06-01 23:47:06 +0530664 SIGTRAP) == NOTIFY_STOP)
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200665 goto exit;
Alexander van Heukelum3d2a71a2008-09-30 18:41:37 +0200666
Steven Rostedt42181182011-12-16 11:43:02 -0500667 /*
668 * Let others (NMI) know that the debug stack is in use
669 * as we may switch to the interrupt stack.
670 */
671 debug_stack_usage_inc();
672
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 /* It's safe to allow irq's after DR6 has been saved */
Alexander van Heukelum3d2a71a2008-09-30 18:41:37 +0200674 preempt_conditional_sti(regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
K.Prasad08d68322009-06-01 23:44:08 +0530676 if (regs->flags & X86_VM_MASK) {
Kees Cookc9408262012-03-09 16:07:10 -0800677 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
678 X86_TRAP_DB);
Bart Oldeman65542872010-09-23 13:16:58 -0400679 preempt_conditional_cli(regs);
Steven Rostedt42181182011-12-16 11:43:02 -0500680 debug_stack_usage_dec();
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200681 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 }
683
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 /*
K.Prasad08d68322009-06-01 23:44:08 +0530685 * Single-stepping through system calls: ignore any exceptions in
686 * kernel space, but re-enable TF when returning to user mode.
687 *
688 * We already checked v86 mode above, so we can check for kernel mode
689 * by just checking the CPL of CS.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 */
K.Prasad08d68322009-06-01 23:44:08 +0530691 if ((dr6 & DR_STEP) && !user_mode(regs)) {
692 tsk->thread.debugreg6 &= ~DR_STEP;
693 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
694 regs->flags &= ~X86_EFLAGS_TF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 }
K.Prasad08d68322009-06-01 23:44:08 +0530696 si_code = get_si_code(tsk->thread.debugreg6);
Frederic Weisbeckera1e80fa2010-06-30 15:09:06 +0200697 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
K.Prasad08d68322009-06-01 23:44:08 +0530698 send_sigtrap(tsk, regs, error_code, si_code);
Alexander van Heukelum3d2a71a2008-09-30 18:41:37 +0200699 preempt_conditional_cli(regs);
Steven Rostedt42181182011-12-16 11:43:02 -0500700 debug_stack_usage_dec();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200702exit:
Andy Lutomirski95927472014-11-19 17:41:09 -0800703 ist_exit(regs, prev_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900705NOKPROBE_SYMBOL(do_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707/*
708 * Note that we play around with the 'TS' bit in an attempt to get
709 * the correct behaviour even in the presence of the asynchronous
710 * IRQ13 behaviour
711 */
Oleg Nesterov5e1b05b2014-05-08 20:34:00 +0200712static void math_error(struct pt_regs *regs, int error_code, int trapnr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
Brian Gerste2e75c92010-03-21 09:00:45 -0400714 struct task_struct *task = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 siginfo_t info;
Brian Gerst9b6dba92010-03-21 09:00:44 -0400716 unsigned short err;
Kees Cookc9408262012-03-09 16:07:10 -0800717 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
718 "simd exception";
Brian Gerste2e75c92010-03-21 09:00:45 -0400719
720 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
721 return;
722 conditional_sti(regs);
723
724 if (!user_mode_vm(regs))
725 {
726 if (!fixup_exception(regs)) {
727 task->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530728 task->thread.trap_nr = trapnr;
Brian Gerste2e75c92010-03-21 09:00:45 -0400729 die(str, regs, error_code);
730 }
731 return;
732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734 /*
735 * Save the info for the exception handler and clear the error.
736 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 save_init_fpu(task);
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530738 task->thread.trap_nr = trapnr;
Brian Gerst9b6dba92010-03-21 09:00:44 -0400739 task->thread.error_code = error_code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 info.si_signo = SIGFPE;
741 info.si_errno = 0;
Oleg Nesterovb02ef202014-05-12 18:24:45 +0200742 info.si_addr = (void __user *)uprobe_get_trap_addr(regs);
Kees Cookc9408262012-03-09 16:07:10 -0800743 if (trapnr == X86_TRAP_MF) {
Brian Gerst9b6dba92010-03-21 09:00:44 -0400744 unsigned short cwd, swd;
745 /*
746 * (~cwd & swd) will mask out exceptions that are not set to unmasked
747 * status. 0x3f is the exception bits in these regs, 0x200 is the
748 * C1 reg you need in case of a stack fault, 0x040 is the stack
749 * fault bit. We should only be taking one exception at a time,
750 * so if this combination doesn't produce any single exception,
751 * then we have a bad program that isn't synchronizing its FPU usage
752 * and it will suffer the consequences since we won't be able to
753 * fully reproduce the context of the exception
754 */
755 cwd = get_fpu_cwd(task);
756 swd = get_fpu_swd(task);
H. Peter Anvinadf77ba2008-12-22 17:56:05 -0800757
Brian Gerst9b6dba92010-03-21 09:00:44 -0400758 err = swd & ~cwd;
759 } else {
760 /*
761 * The SIMD FPU exceptions are handled a little differently, as there
762 * is only a single status/control register. Thus, to determine which
763 * unmasked exception was caught we must mask the exception mask bits
764 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
765 */
766 unsigned short mxcsr = get_fpu_mxcsr(task);
767 err = ~(mxcsr >> 7) & mxcsr;
768 }
H. Peter Anvinadf77ba2008-12-22 17:56:05 -0800769
770 if (err & 0x001) { /* Invalid op */
Ingo Molnarb5964402008-02-26 11:15:50 +0100771 /*
772 * swd & 0x240 == 0x040: Stack Underflow
773 * swd & 0x240 == 0x240: Stack Overflow
774 * User must clear the SF bit (0x40) if set
775 */
776 info.si_code = FPE_FLTINV;
H. Peter Anvinadf77ba2008-12-22 17:56:05 -0800777 } else if (err & 0x004) { /* Divide by Zero */
Ingo Molnarb5964402008-02-26 11:15:50 +0100778 info.si_code = FPE_FLTDIV;
H. Peter Anvinadf77ba2008-12-22 17:56:05 -0800779 } else if (err & 0x008) { /* Overflow */
Ingo Molnarb5964402008-02-26 11:15:50 +0100780 info.si_code = FPE_FLTOVF;
H. Peter Anvinadf77ba2008-12-22 17:56:05 -0800781 } else if (err & 0x012) { /* Denormal, Underflow */
782 info.si_code = FPE_FLTUND;
783 } else if (err & 0x020) { /* Precision */
Ingo Molnarb5964402008-02-26 11:15:50 +0100784 info.si_code = FPE_FLTRES;
H. Peter Anvinadf77ba2008-12-22 17:56:05 -0800785 } else {
Ingo Molnarbd8b96d2008-12-26 09:20:22 +0100786 /*
Kees Cookc9408262012-03-09 16:07:10 -0800787 * If we're using IRQ 13, or supposedly even some trap
788 * X86_TRAP_MF implementations, it's possible
789 * we get a spurious trap, which is not an error.
Ingo Molnarbd8b96d2008-12-26 09:20:22 +0100790 */
Kees Cookc9408262012-03-09 16:07:10 -0800791 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 }
793 force_sig_info(SIGFPE, &info, task);
794}
795
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200796dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100798 enum ctx_state prev_state;
799
800 prev_state = exception_enter();
Kees Cookc9408262012-03-09 16:07:10 -0800801 math_error(regs, error_code, X86_TRAP_MF);
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100802 exception_exit(prev_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803}
804
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200805dotraplinkage void
806do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100808 enum ctx_state prev_state;
809
810 prev_state = exception_enter();
Kees Cookc9408262012-03-09 16:07:10 -0800811 math_error(regs, error_code, X86_TRAP_XF);
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100812 exception_exit(prev_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813}
814
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200815dotraplinkage void
816do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817{
Alexander van Heukelumcf819782008-09-09 21:56:08 +0200818 conditional_sti(regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819#if 0
820 /* No need to warn about this any longer. */
Joe Perchesc767a542012-05-21 19:50:07 -0700821 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822#endif
823}
824
Andi Kleen2605fc22014-05-02 00:44:37 +0200825asmlinkage __visible void __attribute__((weak)) smp_thermal_interrupt(void)
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200826{
827}
828
Andi Kleen2605fc22014-05-02 00:44:37 +0200829asmlinkage __visible void __attribute__((weak)) smp_threshold_interrupt(void)
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200830{
831}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
833/*
Linus Torvalds34ddc812012-02-18 12:56:35 -0800834 * 'math_state_restore()' saves the current math information in the
835 * old math state array, and gets the new ones from the current task
836 *
837 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
838 * Don't touch unless you *really* know how it works.
839 *
840 * Must be called with kernel preemption disabled (eg with local
841 * local interrupts as in the case of do_device_not_available).
842 */
843void math_state_restore(void)
844{
845 struct task_struct *tsk = current;
846
847 if (!tsk_used_math(tsk)) {
848 local_irq_enable();
849 /*
850 * does a slab alloc which can sleep
851 */
852 if (init_fpu(tsk)) {
853 /*
854 * ran out of memory!
855 */
856 do_group_exit(SIGKILL);
857 return;
858 }
859 local_irq_disable();
860 }
861
Oleg Nesterov75756372015-01-15 20:20:28 +0100862 /* Avoid __kernel_fpu_begin() right after __thread_fpu_begin() */
863 kernel_fpu_disable();
Linus Torvalds34ddc812012-02-18 12:56:35 -0800864 __thread_fpu_begin(tsk);
Linus Torvalds80ab6f12012-02-19 11:48:44 -0800865 if (unlikely(restore_fpu_checking(tsk))) {
Suresh Siddha304bced2012-08-24 14:13:02 -0700866 drop_init_fpu(tsk);
Oleg Nesterov38cad572014-05-07 16:47:09 +0200867 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
Oleg Nesterov75756372015-01-15 20:20:28 +0100868 } else {
869 tsk->thread.fpu_counter++;
Linus Torvalds80ab6f12012-02-19 11:48:44 -0800870 }
Oleg Nesterov75756372015-01-15 20:20:28 +0100871 kernel_fpu_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872}
Rusty Russell5992b6d2007-07-19 01:49:21 -0700873EXPORT_SYMBOL_GPL(math_state_restore);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
Masami Hiramatsu93266382014-04-17 17:18:14 +0900875dotraplinkage void
Brian Gerstaa78bcf2009-02-10 09:51:45 -0500876do_device_not_available(struct pt_regs *regs, long error_code)
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200877{
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100878 enum ctx_state prev_state;
879
880 prev_state = exception_enter();
Suresh Siddha5d2bd702012-09-06 14:58:52 -0700881 BUG_ON(use_eager_fpu());
Suresh Siddha304bced2012-08-24 14:13:02 -0700882
Brian Gersta334fe42010-09-03 21:17:15 -0400883#ifdef CONFIG_MATH_EMULATION
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200884 if (read_cr0() & X86_CR0_EM) {
Tejun Heod3157602009-02-09 22:17:39 +0900885 struct math_emu_info info = { };
886
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200887 conditional_sti(regs);
Tejun Heod3157602009-02-09 22:17:39 +0900888
Brian Gerstaa78bcf2009-02-10 09:51:45 -0500889 info.regs = regs;
Tejun Heod3157602009-02-09 22:17:39 +0900890 math_emulate(&info);
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100891 exception_exit(prev_state);
Brian Gersta334fe42010-09-03 21:17:15 -0400892 return;
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200893 }
Brian Gersta334fe42010-09-03 21:17:15 -0400894#endif
895 math_state_restore(); /* interrupts still off */
896#ifdef CONFIG_X86_32
897 conditional_sti(regs);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200898#endif
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100899 exception_exit(prev_state);
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200900}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900901NOKPROBE_SYMBOL(do_device_not_available);
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200902
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200903#ifdef CONFIG_X86_32
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200904dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
Alexander van Heukelumf8e08702008-09-09 21:56:13 +0200905{
906 siginfo_t info;
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100907 enum ctx_state prev_state;
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200908
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100909 prev_state = exception_enter();
Alexander van Heukelumf8e08702008-09-09 21:56:13 +0200910 local_irq_enable();
911
912 info.si_signo = SIGILL;
913 info.si_errno = 0;
914 info.si_code = ILL_BADSTK;
Hannes Ederfc6fcdf2009-02-22 01:00:57 +0100915 info.si_addr = NULL;
Kees Cookc9408262012-03-09 16:07:10 -0800916 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200917 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
918 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
919 &info);
920 }
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +0100921 exception_exit(prev_state);
Alexander van Heukelumf8e08702008-09-09 21:56:13 +0200922}
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200923#endif
Alexander van Heukelumf8e08702008-09-09 21:56:13 +0200924
Jan Kiszka29c84392010-05-20 21:04:29 -0500925/* Set of traps needed for early debugging. */
926void __init early_trap_init(void)
927{
Kees Cookc9408262012-03-09 16:07:10 -0800928 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
Jan Kiszka29c84392010-05-20 21:04:29 -0500929 /* int3 can be called from all */
Kees Cookc9408262012-03-09 16:07:10 -0800930 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
H. Peter Anvin8170e6b2013-01-24 12:19:52 -0800931#ifdef CONFIG_X86_32
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400932 set_intr_gate(X86_TRAP_PF, page_fault);
H. Peter Anvin8170e6b2013-01-24 12:19:52 -0800933#endif
Jan Kiszka29c84392010-05-20 21:04:29 -0500934 load_idt(&idt_descr);
935}
936
H. Peter Anvin8170e6b2013-01-24 12:19:52 -0800937void __init early_trap_pf_init(void)
938{
939#ifdef CONFIG_X86_64
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400940 set_intr_gate(X86_TRAP_PF, page_fault);
H. Peter Anvin8170e6b2013-01-24 12:19:52 -0800941#endif
942}
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944void __init trap_init(void)
945{
Rusty Russelldbeb2be2007-10-19 20:35:03 +0200946 int i;
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948#ifdef CONFIG_EISA
Ingo Molnar927222b2008-01-30 13:33:49 +0100949 void __iomem *p = early_ioremap(0x0FFFD9, 4);
Ingo Molnarb5964402008-02-26 11:15:50 +0100950
951 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 EISA_bus = 1;
Ingo Molnar927222b2008-01-30 13:33:49 +0100953 early_iounmap(p, 4);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954#endif
955
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400956 set_intr_gate(X86_TRAP_DE, divide_error);
Kees Cookc9408262012-03-09 16:07:10 -0800957 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
Alexander van Heukelum699d2932008-10-03 22:00:32 +0200958 /* int4 can be called from all */
Kees Cookc9408262012-03-09 16:07:10 -0800959 set_system_intr_gate(X86_TRAP_OF, &overflow);
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400960 set_intr_gate(X86_TRAP_BR, bounds);
961 set_intr_gate(X86_TRAP_UD, invalid_op);
962 set_intr_gate(X86_TRAP_NM, device_not_available);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200963#ifdef CONFIG_X86_32
Kees Cookc9408262012-03-09 16:07:10 -0800964 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200965#else
Kees Cookc9408262012-03-09 16:07:10 -0800966 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200967#endif
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400968 set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
969 set_intr_gate(X86_TRAP_TS, invalid_TSS);
970 set_intr_gate(X86_TRAP_NP, segment_not_present);
Andy Lutomirski6f442be2014-11-22 18:00:32 -0800971 set_intr_gate(X86_TRAP_SS, stack_segment);
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400972 set_intr_gate(X86_TRAP_GP, general_protection);
973 set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
974 set_intr_gate(X86_TRAP_MF, coprocessor_error);
975 set_intr_gate(X86_TRAP_AC, alignment_check);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976#ifdef CONFIG_X86_MCE
Kees Cookc9408262012-03-09 16:07:10 -0800977 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978#endif
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400979 set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
Yinghai Lubb3f0b52009-01-25 02:38:09 -0800981 /* Reserve all the builtin and the syscall vector: */
982 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
983 set_bit(i, used_vectors);
984
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200985#ifdef CONFIG_IA32_EMULATION
986 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
Yinghai Lubb3f0b52009-01-25 02:38:09 -0800987 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200988#endif
989
990#ifdef CONFIG_X86_32
Alexander van Heukelum699d2932008-10-03 22:00:32 +0200991 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
Rusty Russelldbeb2be2007-10-19 20:35:03 +0200992 set_bit(SYSCALL_VECTOR, used_vectors);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200993#endif
Yinghai Lubb3f0b52009-01-25 02:38:09 -0800994
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 /*
Kees Cook4eefbe72013-04-10 12:24:22 -0700996 * Set the IDT descriptor to a fixed read-only location, so that the
997 * "sidt" instruction will not leak the location of the kernel, and
998 * to defend the IDT against arbitrary memory write vulnerabilities.
999 * It will be reloaded in cpu_init() */
1000 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
1001 idt_descr.address = fix_to_virt(FIX_RO_IDT);
1002
1003 /*
Ingo Molnarb5964402008-02-26 11:15:50 +01001004 * Should be a barrier for any external CPU state:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 */
1006 cpu_init();
1007
Thomas Gleixner428cf902009-08-20 10:35:46 +02001008 x86_init.irqs.trap_init();
Steven Rostedt228bdaa2011-12-09 03:02:19 -05001009
1010#ifdef CONFIG_X86_64
Seiji Aguchi629f4f92013-06-20 11:45:44 -04001011 memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
Kees Cookc9408262012-03-09 16:07:10 -08001012 set_nmi_gate(X86_TRAP_DB, &debug);
1013 set_nmi_gate(X86_TRAP_BP, &int3);
Steven Rostedt228bdaa2011-12-09 03:02:19 -05001014#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015}