blob: b70ca12dd389c57fd32222fa1279f0d46c39a6ac [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>
Paul Gortmaker186f4362016-07-13 20:18:56 -040024#include <linux/export.h>
Ingo Molnarb5964402008-02-26 11:15:50 +010025#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>
Andy Lutomirski35de5b02016-04-26 12:23:24 -070054#include <asm/text-patching.h>
Steven Rostedt08d636b2011-08-16 09:57:10 -040055#include <asm/ftrace.h>
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +020056#include <asm/traps.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <asm/desc.h>
Ingo Molnar78f7f1e2015-04-24 02:54:44 +020058#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 Hansena84eeaa2015-06-07 11:37:01 -070063#include <asm/fpu/xstate.h>
Dave Hansene7126cf2015-06-07 11:37:03 -070064#include <asm/trace/mpx.h>
Dave Hansenfe3d1972014-11-14 07:18:29 -080065#include <asm/mpx.h>
Brian Gerstba3e1272015-07-29 01:41:21 -040066#include <asm/vm86.h>
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +020067
Alexander van Heukelum081f75b2008-10-03 22:00:39 +020068#ifdef CONFIG_X86_64
Thomas Gleixner428cf902009-08-20 10:35:46 +020069#include <asm/x86_init.h>
Alexander van Heukelum081f75b2008-10-03 22:00:39 +020070#include <asm/pgalloc.h>
71#include <asm/proto.h>
Kees Cook4df05f32013-07-16 11:34:41 -070072
73/* No need to be aligned, but done to keep all IDTs defined the same way. */
74gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
Alexander van Heukelum081f75b2008-10-03 22:00:39 +020075#else
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +020076#include <asm/processor-flags.h>
Ingo Molnar8e6dafd2009-02-23 00:34:39 +010077#include <asm/setup.h>
Ingo Molnarb2502b42015-06-08 08:42:03 +020078#include <asm/proto.h>
Alexander van Heukelum081f75b2008-10-03 22:00:39 +020079#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Kees Cook4df05f32013-07-16 11:34:41 -070081/* Must be page-aligned because the real IDT is used in a fixmap. */
82gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
83
Yinghai Lub77b8812008-12-19 15:23:44 -080084DECLARE_BITMAP(used_vectors, NR_VECTORS);
85EXPORT_SYMBOL_GPL(used_vectors);
86
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +010087static inline void cond_local_irq_enable(struct pt_regs *regs)
Alexander van Heukelum762db432008-09-09 21:55:55 +020088{
89 if (regs->flags & X86_EFLAGS_IF)
90 local_irq_enable();
91}
92
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +010093static inline void cond_local_irq_disable(struct pt_regs *regs)
Thomas Gleixnerbe716612009-01-13 23:36:34 +010094{
95 if (regs->flags & X86_EFLAGS_IF)
96 local_irq_disable();
97}
98
Andy Lutomirskiaaee8c32016-05-24 15:54:04 -070099/*
100 * In IST context, we explicitly disable preemption. This serves two
101 * purposes: it makes it much less likely that we would accidentally
102 * schedule in IST context and it will force a warning if we somehow
103 * manage to schedule by accident.
104 */
Andy Lutomirski8c840142015-07-03 12:44:32 -0700105void ist_enter(struct pt_regs *regs)
Andy Lutomirski95927472014-11-19 17:41:09 -0800106{
Andy Lutomirskif39b6f02015-03-18 18:33:33 -0700107 if (user_mode(regs)) {
Linus Torvalds57780772015-09-01 08:40:25 -0700108 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
Andy Lutomirski95927472014-11-19 17:41:09 -0800109 } else {
110 /*
111 * We might have interrupted pretty much anything. In
112 * fact, if we're a machine check, we can even interrupt
113 * NMI processing. We don't want in_nmi() to return true,
114 * but we need to notify RCU.
115 */
116 rcu_nmi_enter();
Andy Lutomirski95927472014-11-19 17:41:09 -0800117 }
Andy Lutomirskib926e6f2015-01-31 04:53:53 -0800118
Andy Lutomirskiaaee8c32016-05-24 15:54:04 -0700119 preempt_disable();
Andy Lutomirskib926e6f2015-01-31 04:53:53 -0800120
121 /* This code is a bit fragile. Test it. */
Paul E. McKenneyf78f5b92015-06-18 15:50:02 -0700122 RCU_LOCKDEP_WARN(!rcu_is_watching(), "ist_enter didn't work");
Andy Lutomirski95927472014-11-19 17:41:09 -0800123}
124
Andy Lutomirski8c840142015-07-03 12:44:32 -0700125void ist_exit(struct pt_regs *regs)
Andy Lutomirski95927472014-11-19 17:41:09 -0800126{
Andy Lutomirskiaaee8c32016-05-24 15:54:04 -0700127 preempt_enable_no_resched();
Andy Lutomirski95927472014-11-19 17:41:09 -0800128
Andy Lutomirski8c840142015-07-03 12:44:32 -0700129 if (!user_mode(regs))
Andy Lutomirski95927472014-11-19 17:41:09 -0800130 rcu_nmi_exit();
131}
132
Andy Lutomirskibced35b2014-11-19 17:59:41 -0800133/**
134 * ist_begin_non_atomic() - begin a non-atomic section in an IST exception
135 * @regs: regs passed to the IST exception handler
136 *
137 * IST exception handlers normally cannot schedule. As a special
138 * exception, if the exception interrupted userspace code (i.e.
Andy Lutomirskif39b6f02015-03-18 18:33:33 -0700139 * user_mode(regs) would return true) and the exception was not
Andy Lutomirskibced35b2014-11-19 17:59:41 -0800140 * a double fault, it can be safe to schedule. ist_begin_non_atomic()
141 * begins a non-atomic section within an ist_enter()/ist_exit() region.
142 * Callers are responsible for enabling interrupts themselves inside
Andy Lutomirski8c840142015-07-03 12:44:32 -0700143 * the non-atomic section, and callers must call ist_end_non_atomic()
Andy Lutomirskibced35b2014-11-19 17:59:41 -0800144 * before ist_exit().
145 */
146void ist_begin_non_atomic(struct pt_regs *regs)
147{
Andy Lutomirskif39b6f02015-03-18 18:33:33 -0700148 BUG_ON(!user_mode(regs));
Andy Lutomirskibced35b2014-11-19 17:59:41 -0800149
150 /*
151 * Sanity check: we need to be on the normal thread stack. This
152 * will catch asm bugs and any attempt to use ist_preempt_enable
153 * from double_fault.
154 */
Andy Lutomirskia7fcf282015-03-06 17:50:19 -0800155 BUG_ON((unsigned long)(current_top_of_stack() -
156 current_stack_pointer()) >= THREAD_SIZE);
Andy Lutomirskibced35b2014-11-19 17:59:41 -0800157
Andy Lutomirskiaaee8c32016-05-24 15:54:04 -0700158 preempt_enable_no_resched();
Andy Lutomirskibced35b2014-11-19 17:59:41 -0800159}
160
161/**
162 * ist_end_non_atomic() - begin a non-atomic section in an IST exception
163 *
164 * Ends a non-atomic section started with ist_begin_non_atomic().
165 */
166void ist_end_non_atomic(void)
167{
Andy Lutomirskiaaee8c32016-05-24 15:54:04 -0700168 preempt_disable();
Andy Lutomirskibced35b2014-11-19 17:59:41 -0800169}
170
Masami Hiramatsu93266382014-04-17 17:18:14 +0900171static nokprobe_inline int
Frederic Weisbeckerc416ddf2012-09-25 14:51:19 +0200172do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
173 struct pt_regs *regs, long error_code)
174{
Andy Lutomirskid74ef112015-03-18 18:33:35 -0700175 if (v8086_mode(regs)) {
Frederic Weisbeckerc416ddf2012-09-25 14:51:19 +0200176 /*
177 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
178 * On nmi (interrupt 2), do_trap should not be called.
179 */
180 if (trapnr < X86_TRAP_UD) {
181 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
182 error_code, trapnr))
183 return 0;
184 }
185 return -1;
186 }
Andy Lutomirskid74ef112015-03-18 18:33:35 -0700187
Ingo Molnar55474c42015-03-29 11:02:34 +0200188 if (!user_mode(regs)) {
Tony Luck548acf12016-02-17 10:20:12 -0800189 if (!fixup_exception(regs, trapnr)) {
Frederic Weisbeckerc416ddf2012-09-25 14:51:19 +0200190 tsk->thread.error_code = error_code;
191 tsk->thread.trap_nr = trapnr;
192 die(str, regs, error_code);
193 }
194 return 0;
195 }
196
197 return -1;
198}
199
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200200static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
201 siginfo_t *info)
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200202{
203 unsigned long siaddr;
204 int sicode;
205
206 switch (trapnr) {
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200207 default:
208 return SEND_SIG_PRIV;
209
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200210 case X86_TRAP_DE:
211 sicode = FPE_INTDIV;
Oleg Nesterovb02ef202014-05-12 18:24:45 +0200212 siaddr = uprobe_get_trap_addr(regs);
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200213 break;
214 case X86_TRAP_UD:
215 sicode = ILL_ILLOPN;
Oleg Nesterovb02ef202014-05-12 18:24:45 +0200216 siaddr = uprobe_get_trap_addr(regs);
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200217 break;
218 case X86_TRAP_AC:
219 sicode = BUS_ADRALN;
220 siaddr = 0;
221 break;
222 }
223
224 info->si_signo = signr;
225 info->si_errno = 0;
226 info->si_code = sicode;
227 info->si_addr = (void __user *)siaddr;
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200228 return info;
Oleg Nesterov958d3d72014-05-07 17:59:39 +0200229}
230
Masami Hiramatsu93266382014-04-17 17:18:14 +0900231static void
Alexander van Heukelum3c1326f2008-09-26 14:03:08 +0200232do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
Ingo Molnarb5964402008-02-26 11:15:50 +0100233 long error_code, siginfo_t *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234{
Alexander Nyberg4f339ec2005-06-25 14:58:27 -0700235 struct task_struct *tsk = current;
Alexander Nyberg4f339ec2005-06-25 14:58:27 -0700236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Frederic Weisbeckerc416ddf2012-09-25 14:51:19 +0200238 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
239 return;
Ingo Molnarb5964402008-02-26 11:15:50 +0100240 /*
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530241 * We want error_code and trap_nr set for userspace faults and
Ingo Molnarb5964402008-02-26 11:15:50 +0100242 * kernelspace faults which result in die(), but not
243 * kernelspace faults which are fixed up. die() gives the
244 * process no chance to handle the signal and notice the
245 * kernel fault information, so that won't result in polluting
246 * the information about previously queued, but not yet
247 * delivered, faults. See also do_general_protection below.
248 */
249 tsk->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530250 tsk->thread.trap_nr = trapnr;
Ingo Molnarb5964402008-02-26 11:15:50 +0100251
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200252 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
253 printk_ratelimit()) {
Joe Perchesc767a542012-05-21 19:50:07 -0700254 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
255 tsk->comm, tsk->pid, str,
256 regs->ip, regs->sp, error_code);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200257 print_vma_addr(" in ", regs->ip);
Joe Perchesc767a542012-05-21 19:50:07 -0700258 pr_cont("\n");
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200259 }
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200260
Oleg Nesterov38cad572014-05-07 16:47:09 +0200261 force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900263NOKPROBE_SYMBOL(do_trap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Oleg Nesterovdff07962014-05-07 17:21:34 +0200265static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200266 unsigned long trapnr, int signr)
Oleg Nesterovdff07962014-05-07 17:21:34 +0200267{
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200268 siginfo_t info;
Oleg Nesterovdff07962014-05-07 17:21:34 +0200269
Linus Torvalds57780772015-09-01 08:40:25 -0700270 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
Andy Lutomirski02fdcd52015-07-03 12:44:24 -0700271
Oleg Nesterovdff07962014-05-07 17:21:34 +0200272 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
273 NOTIFY_STOP) {
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100274 cond_local_irq_enable(regs);
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200275 do_trap(trapnr, signr, str, regs, error_code,
276 fill_trap_info(regs, signr, trapnr, &info));
Oleg Nesterovdff07962014-05-07 17:21:34 +0200277 }
Oleg Nesterovdff07962014-05-07 17:21:34 +0200278}
279
Ingo Molnarb5964402008-02-26 11:15:50 +0100280#define DO_ERROR(trapnr, signr, str, name) \
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200281dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
Ingo Molnarb5964402008-02-26 11:15:50 +0100282{ \
Oleg Nesterov1c326c42014-05-08 20:04:11 +0200283 do_error_trap(regs, error_code, str, trapnr, signr); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284}
285
Oleg Nesterov0eb14832014-05-08 20:12:24 +0200286DO_ERROR(X86_TRAP_DE, SIGFPE, "divide error", divide_error)
287DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
Oleg Nesterov0eb14832014-05-08 20:12:24 +0200288DO_ERROR(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op)
289DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",coprocessor_segment_overrun)
290DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
291DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
Oleg Nesterov0eb14832014-05-08 20:12:24 +0200292DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
Oleg Nesterov0eb14832014-05-08 20:12:24 +0200293DO_ERROR(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200295#ifdef CONFIG_X86_64
296/* Runs on IST stack */
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200297dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
298{
299 static const char str[] = "double fault";
300 struct task_struct *tsk = current;
301
Andy Lutomirskiaf726f22014-11-22 18:00:31 -0800302#ifdef CONFIG_X86_ESPFIX64
303 extern unsigned char native_irq_return_iret[];
304
305 /*
306 * If IRET takes a non-IST fault on the espfix64 stack, then we
307 * end up promoting it to a doublefault. In that case, modify
308 * the stack to make it look like we just entered the #GP
309 * handler from user space, similar to bad_iret.
Andy Lutomirski95927472014-11-19 17:41:09 -0800310 *
311 * No need for ist_enter here because we don't use RCU.
Andy Lutomirskiaf726f22014-11-22 18:00:31 -0800312 */
313 if (((long)regs->sp >> PGDIR_SHIFT) == ESPFIX_PGD_ENTRY &&
314 regs->cs == __KERNEL_CS &&
315 regs->ip == (unsigned long)native_irq_return_iret)
316 {
317 struct pt_regs *normal_regs = task_pt_regs(current);
318
319 /* Fake a #GP(0) from userspace. */
320 memmove(&normal_regs->ip, (void *)regs->sp, 5*8);
321 normal_regs->orig_ax = 0; /* Missing (lost) #GP error code */
322 regs->ip = (unsigned long)general_protection;
323 regs->sp = (unsigned long)&normal_regs->orig_ax;
Andy Lutomirski95927472014-11-19 17:41:09 -0800324
Andy Lutomirskiaf726f22014-11-22 18:00:31 -0800325 return;
326 }
327#endif
328
Andy Lutomirski8c840142015-07-03 12:44:32 -0700329 ist_enter(regs);
Kees Cookc9408262012-03-09 16:07:10 -0800330 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200331
332 tsk->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530333 tsk->thread.trap_nr = X86_TRAP_DF;
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200334
Borislav Petkov4d067d82013-05-09 12:02:29 +0200335#ifdef CONFIG_DOUBLEFAULT
336 df_debug(regs, error_code);
337#endif
Ingo Molnarbd8b96d2008-12-26 09:20:22 +0100338 /*
339 * This is always a kernel trap and never fixable (and thus must
340 * never return).
341 */
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200342 for (;;)
343 die(str, regs, error_code);
344}
345#endif
346
Dave Hansenfe3d1972014-11-14 07:18:29 -0800347dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
348{
Dave Hansen1126cb452015-09-02 16:31:29 -0700349 const struct mpx_bndcsr *bndcsr;
Dave Hansenfe3d1972014-11-14 07:18:29 -0800350 siginfo_t *info;
351
Linus Torvalds57780772015-09-01 08:40:25 -0700352 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
Dave Hansenfe3d1972014-11-14 07:18:29 -0800353 if (notify_die(DIE_TRAP, "bounds", regs, error_code,
354 X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
Andy Lutomirski8c840142015-07-03 12:44:32 -0700355 return;
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100356 cond_local_irq_enable(regs);
Dave Hansenfe3d1972014-11-14 07:18:29 -0800357
Andy Lutomirskif39b6f02015-03-18 18:33:33 -0700358 if (!user_mode(regs))
Dave Hansenfe3d1972014-11-14 07:18:29 -0800359 die("bounds", regs, error_code);
360
361 if (!cpu_feature_enabled(X86_FEATURE_MPX)) {
362 /* The exception is not from Intel MPX */
363 goto exit_trap;
364 }
365
366 /*
367 * We need to look at BNDSTATUS to resolve this exception.
Dave Hansena84eeaa2015-06-07 11:37:01 -0700368 * A NULL here might mean that it is in its 'init state',
369 * which is all zeros which indicates MPX was not
370 * responsible for the exception.
Dave Hansenfe3d1972014-11-14 07:18:29 -0800371 */
Dave Hansend91cab72015-09-02 16:31:26 -0700372 bndcsr = get_xsave_field_ptr(XFEATURE_MASK_BNDCSR);
Dave Hansenfe3d1972014-11-14 07:18:29 -0800373 if (!bndcsr)
374 goto exit_trap;
375
Dave Hansene7126cf2015-06-07 11:37:03 -0700376 trace_bounds_exception_mpx(bndcsr);
Dave Hansenfe3d1972014-11-14 07:18:29 -0800377 /*
378 * The error code field of the BNDSTATUS register communicates status
379 * information of a bound range exception #BR or operation involving
380 * bound directory.
381 */
382 switch (bndcsr->bndstatus & MPX_BNDSTA_ERROR_CODE) {
383 case 2: /* Bound directory has invalid entry. */
Dave Hansen46a6e0c2015-06-07 11:37:02 -0700384 if (mpx_handle_bd_fault())
Dave Hansenfe3d1972014-11-14 07:18:29 -0800385 goto exit_trap;
386 break; /* Success, it was handled */
387 case 1: /* Bound violation. */
Dave Hansen46a6e0c2015-06-07 11:37:02 -0700388 info = mpx_generate_siginfo(regs);
Dan Carpentere10abb22014-11-25 20:21:14 +0300389 if (IS_ERR(info)) {
Dave Hansenfe3d1972014-11-14 07:18:29 -0800390 /*
391 * We failed to decode the MPX instruction. Act as if
392 * the exception was not caused by MPX.
393 */
394 goto exit_trap;
395 }
396 /*
397 * Success, we decoded the instruction and retrieved
398 * an 'info' containing the address being accessed
399 * which caused the exception. This information
400 * allows and application to possibly handle the
401 * #BR exception itself.
402 */
403 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, info);
404 kfree(info);
405 break;
406 case 0: /* No exception caused by Intel MPX operations. */
407 goto exit_trap;
408 default:
409 die("bounds", regs, error_code);
410 }
411
Dave Hansenfe3d1972014-11-14 07:18:29 -0800412 return;
Andy Lutomirski8c840142015-07-03 12:44:32 -0700413
Dave Hansenfe3d1972014-11-14 07:18:29 -0800414exit_trap:
415 /*
416 * This path out is for all the cases where we could not
417 * handle the exception in some way (like allocating a
418 * table or telling userspace about it. We will also end
419 * up here if the kernel has MPX turned off at compile
420 * time..
421 */
422 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, NULL);
Dave Hansenfe3d1972014-11-14 07:18:29 -0800423}
424
Masami Hiramatsu93266382014-04-17 17:18:14 +0900425dotraplinkage void
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200426do_general_protection(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427{
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200428 struct task_struct *tsk;
Ingo Molnarb5964402008-02-26 11:15:50 +0100429
Linus Torvalds57780772015-09-01 08:40:25 -0700430 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100431 cond_local_irq_enable(regs);
Alexander van Heukelumc6df0d72008-09-09 21:56:07 +0200432
Andy Lutomirskid74ef112015-03-18 18:33:35 -0700433 if (v8086_mode(regs)) {
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200434 local_irq_enable();
435 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
Andy Lutomirski8c840142015-07-03 12:44:32 -0700436 return;
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200439 tsk = current;
Ingo Molnar55474c42015-03-29 11:02:34 +0200440 if (!user_mode(regs)) {
Tony Luck548acf12016-02-17 10:20:12 -0800441 if (fixup_exception(regs, X86_TRAP_GP))
Andy Lutomirski8c840142015-07-03 12:44:32 -0700442 return;
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200443
444 tsk->thread.error_code = error_code;
445 tsk->thread.trap_nr = X86_TRAP_GP;
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200446 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
447 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200448 die("general protection fault", regs, error_code);
Andy Lutomirski8c840142015-07-03 12:44:32 -0700449 return;
Frederic Weisbeckeref3f6282012-09-24 21:05:52 +0200450 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200452 tsk->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530453 tsk->thread.trap_nr = X86_TRAP_GP;
Ingo Molnarb5964402008-02-26 11:15:50 +0100454
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200455 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
456 printk_ratelimit()) {
Joe Perchesc767a542012-05-21 19:50:07 -0700457 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
Alexander van Heukelum13485ab2008-07-02 01:32:04 +0200458 tsk->comm, task_pid_nr(tsk),
459 regs->ip, regs->sp, error_code);
Andi Kleen03252912008-01-30 13:33:18 +0100460 print_vma_addr(" in ", regs->ip);
Joe Perchesc767a542012-05-21 19:50:07 -0700461 pr_cont("\n");
Andi Kleen03252912008-01-30 13:33:18 +0100462 }
Masoud Asgharifard Sharbianiabd4f752007-07-22 11:12:28 +0200463
Oleg Nesterov38cad572014-05-07 16:47:09 +0200464 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900466NOKPROBE_SYMBOL(do_general_protection);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +0200468/* May run on IST stack. */
Masami Hiramatsu93266382014-04-17 17:18:14 +0900469dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470{
Steven Rostedt08d636b2011-08-16 09:57:10 -0400471#ifdef CONFIG_DYNAMIC_FTRACE
Steven Rostedta192cd02012-05-30 13:26:37 -0400472 /*
473 * ftrace must be first, everything else may cause a recursive crash.
474 * See note by declaration of modifying_ftrace_code in ftrace.c
475 */
476 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
477 ftrace_int3_handler(regs))
Steven Rostedt08d636b2011-08-16 09:57:10 -0400478 return;
479#endif
Jiri Kosina17f41572013-07-23 10:09:28 +0200480 if (poke_int3_handler(regs))
481 return;
482
Andy Lutomirski8c840142015-07-03 12:44:32 -0700483 ist_enter(regs);
Linus Torvalds57780772015-09-01 08:40:25 -0700484 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
Jason Wesself503b5a2010-05-20 21:04:25 -0500485#ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
Kees Cookc9408262012-03-09 16:07:10 -0800486 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
487 SIGTRAP) == NOTIFY_STOP)
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200488 goto exit;
Jason Wesself503b5a2010-05-20 21:04:25 -0500489#endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
Srikar Dronamrajucc3a1bf2011-10-25 19:51:59 +0530490
Masami Hiramatsu6f6343f2014-04-17 17:17:33 +0900491#ifdef CONFIG_KPROBES
492 if (kprobe_int3_handler(regs))
Masami Hiramatsu4cdf77a2014-06-14 06:47:12 +0000493 goto exit;
Masami Hiramatsu6f6343f2014-04-17 17:17:33 +0900494#endif
495
Kees Cookc9408262012-03-09 16:07:10 -0800496 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
497 SIGTRAP) == NOTIFY_STOP)
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200498 goto exit;
Ingo Molnarb5964402008-02-26 11:15:50 +0100499
Steven Rostedt42181182011-12-16 11:43:02 -0500500 /*
501 * Let others (NMI) know that the debug stack is in use
502 * as we may switch to the interrupt stack.
503 */
504 debug_stack_usage_inc();
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100505 preempt_disable();
506 cond_local_irq_enable(regs);
Kees Cookc9408262012-03-09 16:07:10 -0800507 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100508 cond_local_irq_disable(regs);
509 preempt_enable_no_resched();
Steven Rostedt42181182011-12-16 11:43:02 -0500510 debug_stack_usage_dec();
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200511exit:
Andy Lutomirski8c840142015-07-03 12:44:32 -0700512 ist_exit(regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900514NOKPROBE_SYMBOL(do_int3);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200516#ifdef CONFIG_X86_64
Ingo Molnarbd8b96d2008-12-26 09:20:22 +0100517/*
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800518 * Help handler running on IST stack to switch off the IST stack if the
519 * interrupted code was in user mode. The actual stack switch is done in
520 * entry_64.S
Ingo Molnarbd8b96d2008-12-26 09:20:22 +0100521 */
Andy Lutomirski7ddc6a22014-11-24 17:39:06 -0800522asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200523{
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800524 struct pt_regs *regs = task_pt_regs(current);
525 *regs = *eregs;
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200526 return regs;
527}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900528NOKPROBE_SYMBOL(sync_regs);
Andy Lutomirskib645af22014-11-22 18:00:33 -0800529
530struct bad_iret_stack {
531 void *error_entry_ret;
532 struct pt_regs regs;
533};
534
Andy Lutomirski7ddc6a22014-11-24 17:39:06 -0800535asmlinkage __visible notrace
Andy Lutomirskib645af22014-11-22 18:00:33 -0800536struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
537{
538 /*
539 * This is called from entry_64.S early in handling a fault
540 * caused by a bad iret to user mode. To handle the fault
541 * correctly, we want move our stack frame to task_pt_regs
542 * and we want to pretend that the exception came from the
543 * iret target.
544 */
545 struct bad_iret_stack *new_stack =
546 container_of(task_pt_regs(current),
547 struct bad_iret_stack, regs);
548
549 /* Copy the IRET target to the new stack. */
550 memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
551
552 /* Copy the remainder of the stack from the current stack. */
553 memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
554
Andy Lutomirskif39b6f02015-03-18 18:33:33 -0700555 BUG_ON(!user_mode(&new_stack->regs));
Andy Lutomirskib645af22014-11-22 18:00:33 -0800556 return new_stack;
557}
Andy Lutomirski7ddc6a22014-11-24 17:39:06 -0800558NOKPROBE_SYMBOL(fixup_bad_iret);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200559#endif
560
Andy Lutomirskif2b37572016-03-09 19:00:30 -0800561static bool is_sysenter_singlestep(struct pt_regs *regs)
562{
563 /*
564 * We don't try for precision here. If we're anywhere in the region of
565 * code that can be single-stepped in the SYSENTER entry path, then
566 * assume that this is a useless single-step trap due to SYSENTER
567 * being invoked with TF set. (We don't know in advance exactly
568 * which instructions will be hit because BTF could plausibly
569 * be set.)
570 */
571#ifdef CONFIG_X86_32
572 return (regs->ip - (unsigned long)__begin_SYSENTER_singlestep_region) <
573 (unsigned long)__end_SYSENTER_singlestep_region -
574 (unsigned long)__begin_SYSENTER_singlestep_region;
575#elif defined(CONFIG_IA32_EMULATION)
576 return (regs->ip - (unsigned long)entry_SYSENTER_compat) <
577 (unsigned long)__end_entry_SYSENTER_compat -
578 (unsigned long)entry_SYSENTER_compat;
579#else
580 return false;
581#endif
582}
583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584/*
585 * Our handling of the processor debug registers is non-trivial.
586 * We do not clear them on entry and exit from the kernel. Therefore
587 * it is possible to get a watchpoint trap here from inside the kernel.
588 * However, the code in ./ptrace.c has ensured that the user can
589 * only set watchpoints on userspace addresses. Therefore the in-kernel
590 * watchpoint trap can only occur in code which is reading/writing
591 * from user space. Such code must not hold kernel locks (since it
592 * can equally take a page fault), therefore it is safe to call
593 * force_sig_info even though that claims and releases locks.
Ingo Molnarb5964402008-02-26 11:15:50 +0100594 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 * Code in ./signal.c ensures that the debug control register
596 * is restored before we deliver any signal, and therefore that
597 * user code runs with the correct debug control register even though
598 * we clear it here.
599 *
600 * Being careful here means that we don't have to be as careful in a
601 * lot of more complicated places (task switching can be a bit lazy
602 * about restoring all the debug state, and ptrace doesn't have to
603 * find every occurrence of the TF bit that could be saved away even
604 * by user code)
Alexander van Heukelumc1d518c2008-10-03 23:17:11 +0200605 *
606 * May run on IST stack.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Masami Hiramatsu93266382014-04-17 17:18:14 +0900608dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 struct task_struct *tsk = current;
Frederic Weisbeckera1e80fa2010-06-30 15:09:06 +0200611 int user_icebp = 0;
K.Prasad08d68322009-06-01 23:44:08 +0530612 unsigned long dr6;
Srinivasa Dsda654b72008-09-23 15:23:52 +0530613 int si_code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Andy Lutomirski8c840142015-07-03 12:44:32 -0700615 ist_enter(regs);
Masami Hiramatsu4cdf77a2014-06-14 06:47:12 +0000616
K.Prasad08d68322009-06-01 23:44:08 +0530617 get_debugreg(dr6, 6);
Andy Lutomirski8bb56432016-03-09 19:00:29 -0800618 /*
619 * The Intel SDM says:
620 *
621 * Certain debug exceptions may clear bits 0-3. The remaining
622 * contents of the DR6 register are never cleared by the
623 * processor. To avoid confusion in identifying debug
624 * exceptions, debug handlers should clear the register before
625 * returning to the interrupted task.
626 *
627 * Keep it simple: clear DR6 immediately.
628 */
629 set_debugreg(0, 6);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
K.Prasad40f92492010-01-28 16:44:01 +0530631 /* Filter out all the reserved bits which are preset to 1 */
632 dr6 &= ~DR6_RESERVED;
633
Frederic Weisbeckera1e80fa2010-06-30 15:09:06 +0200634 /*
Andy Lutomirski81edd9f2016-03-09 19:00:28 -0800635 * The SDM says "The processor clears the BTF flag when it
636 * generates a debug exception." Clear TIF_BLOCKSTEP to keep
637 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
638 */
639 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
640
Andy Lutomirskif2b37572016-03-09 19:00:30 -0800641 if (unlikely(!user_mode(regs) && (dr6 & DR_STEP) &&
642 is_sysenter_singlestep(regs))) {
643 dr6 &= ~DR_STEP;
644 if (!dr6)
645 goto exit;
646 /*
647 * else we might have gotten a single-step trap and hit a
648 * watchpoint at the same time, in which case we should fall
649 * through and handle the watchpoint.
650 */
651 }
652
Andy Lutomirski81edd9f2016-03-09 19:00:28 -0800653 /*
Frederic Weisbeckera1e80fa2010-06-30 15:09:06 +0200654 * If dr6 has no reason to give us about the origin of this trap,
655 * then it's very likely the result of an icebp/int01 trap.
656 * User wants a sigtrap for that.
657 */
Andy Lutomirskif39b6f02015-03-18 18:33:33 -0700658 if (!dr6 && user_mode(regs))
Frederic Weisbeckera1e80fa2010-06-30 15:09:06 +0200659 user_icebp = 1;
660
Andy Lutomirskif2b37572016-03-09 19:00:30 -0800661 /* Catch kmemcheck conditions! */
Ingo Molnareadb8a02009-06-17 12:52:15 +0200662 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200663 goto exit;
Vegard Nossumf8561292008-04-04 00:53:23 +0200664
K.Prasad08d68322009-06-01 23:44:08 +0530665 /* Store the virtualized DR6 value */
666 tsk->thread.debugreg6 = dr6;
667
Masami Hiramatsu6f6343f2014-04-17 17:17:33 +0900668#ifdef CONFIG_KPROBES
669 if (kprobe_debug_handler(regs))
670 goto exit;
671#endif
672
Rusty Russell5a802e12013-06-16 14:12:47 +0930673 if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
K.Prasad62edab92009-06-01 23:47:06 +0530674 SIGTRAP) == NOTIFY_STOP)
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200675 goto exit;
Alexander van Heukelum3d2a71a2008-09-30 18:41:37 +0200676
Steven Rostedt42181182011-12-16 11:43:02 -0500677 /*
678 * Let others (NMI) know that the debug stack is in use
679 * as we may switch to the interrupt stack.
680 */
681 debug_stack_usage_inc();
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 /* It's safe to allow irq's after DR6 has been saved */
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100684 preempt_disable();
685 cond_local_irq_enable(regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
Andy Lutomirskid74ef112015-03-18 18:33:35 -0700687 if (v8086_mode(regs)) {
Kees Cookc9408262012-03-09 16:07:10 -0800688 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
689 X86_TRAP_DB);
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100690 cond_local_irq_disable(regs);
691 preempt_enable_no_resched();
Steven Rostedt42181182011-12-16 11:43:02 -0500692 debug_stack_usage_dec();
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200693 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
695
Andy Lutomirskif2b37572016-03-09 19:00:30 -0800696 if (WARN_ON_ONCE((dr6 & DR_STEP) && !user_mode(regs))) {
697 /*
698 * Historical junk that used to handle SYSENTER single-stepping.
699 * This should be unreachable now. If we survive for a while
700 * without anyone hitting this warning, we'll turn this into
701 * an oops.
702 */
K.Prasad08d68322009-06-01 23:44:08 +0530703 tsk->thread.debugreg6 &= ~DR_STEP;
704 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
705 regs->flags &= ~X86_EFLAGS_TF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
K.Prasad08d68322009-06-01 23:44:08 +0530707 si_code = get_si_code(tsk->thread.debugreg6);
Frederic Weisbeckera1e80fa2010-06-30 15:09:06 +0200708 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
K.Prasad08d68322009-06-01 23:44:08 +0530709 send_sigtrap(tsk, regs, error_code, si_code);
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100710 cond_local_irq_disable(regs);
711 preempt_enable_no_resched();
Steven Rostedt42181182011-12-16 11:43:02 -0500712 debug_stack_usage_dec();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200714exit:
Andy Lutomirski2a41aa42016-03-09 19:00:33 -0800715#if defined(CONFIG_X86_32)
716 /*
717 * This is the most likely code path that involves non-trivial use
718 * of the SYSENTER stack. Check that we haven't overrun it.
719 */
720 WARN(this_cpu_read(cpu_tss.SYSENTER_stack_canary) != STACK_END_MAGIC,
721 "Overran or corrupted SYSENTER stack\n");
722#endif
Andy Lutomirski8c840142015-07-03 12:44:32 -0700723 ist_exit(regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900725NOKPROBE_SYMBOL(do_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
727/*
728 * Note that we play around with the 'TS' bit in an attempt to get
729 * the correct behaviour even in the presence of the asynchronous
730 * IRQ13 behaviour
731 */
Oleg Nesterov5e1b05b2014-05-08 20:34:00 +0200732static void math_error(struct pt_regs *regs, int error_code, int trapnr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
Brian Gerste2e75c92010-03-21 09:00:45 -0400734 struct task_struct *task = current;
Ingo Molnare1cebad2015-04-30 09:29:38 +0200735 struct fpu *fpu = &task->thread.fpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 siginfo_t info;
Kees Cookc9408262012-03-09 16:07:10 -0800737 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
738 "simd exception";
Brian Gerste2e75c92010-03-21 09:00:45 -0400739
740 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
741 return;
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100742 cond_local_irq_enable(regs);
Brian Gerste2e75c92010-03-21 09:00:45 -0400743
Ingo Molnare1cebad2015-04-30 09:29:38 +0200744 if (!user_mode(regs)) {
Tony Luck548acf12016-02-17 10:20:12 -0800745 if (!fixup_exception(regs, trapnr)) {
Brian Gerste2e75c92010-03-21 09:00:45 -0400746 task->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530747 task->thread.trap_nr = trapnr;
Brian Gerste2e75c92010-03-21 09:00:45 -0400748 die(str, regs, error_code);
749 }
750 return;
751 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
753 /*
754 * Save the info for the exception handler and clear the error.
755 */
Ingo Molnare1cebad2015-04-30 09:29:38 +0200756 fpu__save(fpu);
757
758 task->thread.trap_nr = trapnr;
Brian Gerst9b6dba92010-03-21 09:00:44 -0400759 task->thread.error_code = error_code;
Ingo Molnare1cebad2015-04-30 09:29:38 +0200760 info.si_signo = SIGFPE;
761 info.si_errno = 0;
762 info.si_addr = (void __user *)uprobe_get_trap_addr(regs);
H. Peter Anvinadf77ba2008-12-22 17:56:05 -0800763
Ingo Molnare1cebad2015-04-30 09:29:38 +0200764 info.si_code = fpu__exception_code(fpu, trapnr);
H. Peter Anvinadf77ba2008-12-22 17:56:05 -0800765
Ingo Molnare1cebad2015-04-30 09:29:38 +0200766 /* Retry when we get spurious exceptions: */
767 if (!info.si_code)
Kees Cookc9408262012-03-09 16:07:10 -0800768 return;
Ingo Molnare1cebad2015-04-30 09:29:38 +0200769
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 force_sig_info(SIGFPE, &info, task);
771}
772
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200773dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774{
Linus Torvalds57780772015-09-01 08:40:25 -0700775 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
Kees Cookc9408262012-03-09 16:07:10 -0800776 math_error(regs, error_code, X86_TRAP_MF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777}
778
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200779dotraplinkage void
780do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781{
Linus Torvalds57780772015-09-01 08:40:25 -0700782 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
Kees Cookc9408262012-03-09 16:07:10 -0800783 math_error(regs, error_code, X86_TRAP_XF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
785
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200786dotraplinkage void
787do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100789 cond_local_irq_enable(regs);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200790}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Masami Hiramatsu93266382014-04-17 17:18:14 +0900792dotraplinkage void
Brian Gerstaa78bcf2009-02-10 09:51:45 -0500793do_device_not_available(struct pt_regs *regs, long error_code)
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200794{
Linus Torvalds57780772015-09-01 08:40:25 -0700795 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
Suresh Siddha304bced2012-08-24 14:13:02 -0700796
Brian Gersta334fe42010-09-03 21:17:15 -0400797#ifdef CONFIG_MATH_EMULATION
Andy Lutomirskic6ab1092016-01-24 14:38:09 -0800798 if (!boot_cpu_has(X86_FEATURE_FPU) && (read_cr0() & X86_CR0_EM)) {
Tejun Heod3157602009-02-09 22:17:39 +0900799 struct math_emu_info info = { };
800
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100801 cond_local_irq_enable(regs);
Tejun Heod3157602009-02-09 22:17:39 +0900802
Brian Gerstaa78bcf2009-02-10 09:51:45 -0500803 info.regs = regs;
Tejun Heod3157602009-02-09 22:17:39 +0900804 math_emulate(&info);
Brian Gersta334fe42010-09-03 21:17:15 -0400805 return;
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200806 }
Brian Gersta334fe42010-09-03 21:17:15 -0400807#endif
Ingo Molnare1884d62015-05-04 11:49:58 +0200808 fpu__restore(&current->thread.fpu); /* interrupts still off */
Brian Gersta334fe42010-09-03 21:17:15 -0400809#ifdef CONFIG_X86_32
Alexander Kuleshovd99e1bd2016-01-25 20:41:46 +0100810 cond_local_irq_enable(regs);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200811#endif
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200812}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900813NOKPROBE_SYMBOL(do_device_not_available);
Alexander van Heukelum7643e9b2008-09-09 21:56:02 +0200814
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200815#ifdef CONFIG_X86_32
Alexander van Heukelume407d6202008-09-30 18:41:36 +0200816dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
Alexander van Heukelumf8e08702008-09-09 21:56:13 +0200817{
818 siginfo_t info;
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200819
Linus Torvalds57780772015-09-01 08:40:25 -0700820 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
Alexander van Heukelumf8e08702008-09-09 21:56:13 +0200821 local_irq_enable();
822
823 info.si_signo = SIGILL;
824 info.si_errno = 0;
825 info.si_code = ILL_BADSTK;
Hannes Ederfc6fcdf2009-02-22 01:00:57 +0100826 info.si_addr = NULL;
Kees Cookc9408262012-03-09 16:07:10 -0800827 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +0200828 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
829 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
830 &info);
831 }
Alexander van Heukelumf8e08702008-09-09 21:56:13 +0200832}
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200833#endif
Alexander van Heukelumf8e08702008-09-09 21:56:13 +0200834
Jan Kiszka29c84392010-05-20 21:04:29 -0500835/* Set of traps needed for early debugging. */
836void __init early_trap_init(void)
837{
Wang Nanb4d83272015-02-26 13:49:39 +0800838 /*
Wang Nan5eca7452015-02-27 12:19:49 +0800839 * Don't use IST to set DEBUG_STACK as it doesn't work until TSS
840 * is ready in cpu_init() <-- trap_init(). Before trap_init(),
841 * CPU runs at ring 0 so it is impossible to hit an invalid
842 * stack. Using the original stack works well enough at this
843 * early stage. DEBUG_STACK will be equipped after cpu_init() in
Wang Nanb4d83272015-02-26 13:49:39 +0800844 * trap_init().
Wang Nan5eca7452015-02-27 12:19:49 +0800845 *
846 * We don't need to set trace_idt_table like set_intr_gate(),
847 * since we don't have trace_debug and it will be reset to
848 * 'debug' in trap_init() by set_intr_gate_ist().
Wang Nanb4d83272015-02-26 13:49:39 +0800849 */
Wang Nan5eca7452015-02-27 12:19:49 +0800850 set_intr_gate_notrace(X86_TRAP_DB, debug);
Jan Kiszka29c84392010-05-20 21:04:29 -0500851 /* int3 can be called from all */
Wang Nan5eca7452015-02-27 12:19:49 +0800852 set_system_intr_gate(X86_TRAP_BP, &int3);
H. Peter Anvin8170e6b2013-01-24 12:19:52 -0800853#ifdef CONFIG_X86_32
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400854 set_intr_gate(X86_TRAP_PF, page_fault);
H. Peter Anvin8170e6b2013-01-24 12:19:52 -0800855#endif
Jan Kiszka29c84392010-05-20 21:04:29 -0500856 load_idt(&idt_descr);
857}
858
H. Peter Anvin8170e6b2013-01-24 12:19:52 -0800859void __init early_trap_pf_init(void)
860{
861#ifdef CONFIG_X86_64
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400862 set_intr_gate(X86_TRAP_PF, page_fault);
H. Peter Anvin8170e6b2013-01-24 12:19:52 -0800863#endif
864}
865
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866void __init trap_init(void)
867{
Rusty Russelldbeb2be2007-10-19 20:35:03 +0200868 int i;
869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870#ifdef CONFIG_EISA
Ingo Molnar927222b2008-01-30 13:33:49 +0100871 void __iomem *p = early_ioremap(0x0FFFD9, 4);
Ingo Molnarb5964402008-02-26 11:15:50 +0100872
873 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 EISA_bus = 1;
Ingo Molnar927222b2008-01-30 13:33:49 +0100875 early_iounmap(p, 4);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876#endif
877
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400878 set_intr_gate(X86_TRAP_DE, divide_error);
Kees Cookc9408262012-03-09 16:07:10 -0800879 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
Alexander van Heukelum699d2932008-10-03 22:00:32 +0200880 /* int4 can be called from all */
Kees Cookc9408262012-03-09 16:07:10 -0800881 set_system_intr_gate(X86_TRAP_OF, &overflow);
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400882 set_intr_gate(X86_TRAP_BR, bounds);
883 set_intr_gate(X86_TRAP_UD, invalid_op);
884 set_intr_gate(X86_TRAP_NM, device_not_available);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200885#ifdef CONFIG_X86_32
Kees Cookc9408262012-03-09 16:07:10 -0800886 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200887#else
Kees Cookc9408262012-03-09 16:07:10 -0800888 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200889#endif
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400890 set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
891 set_intr_gate(X86_TRAP_TS, invalid_TSS);
892 set_intr_gate(X86_TRAP_NP, segment_not_present);
Andy Lutomirski6f442be2014-11-22 18:00:32 -0800893 set_intr_gate(X86_TRAP_SS, stack_segment);
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400894 set_intr_gate(X86_TRAP_GP, general_protection);
895 set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
896 set_intr_gate(X86_TRAP_MF, coprocessor_error);
897 set_intr_gate(X86_TRAP_AC, alignment_check);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898#ifdef CONFIG_X86_MCE
Kees Cookc9408262012-03-09 16:07:10 -0800899 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900#endif
Seiji Aguchi25c74b12013-10-30 16:37:00 -0400901 set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
Yinghai Lubb3f0b52009-01-25 02:38:09 -0800903 /* Reserve all the builtin and the syscall vector: */
904 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
905 set_bit(i, used_vectors);
906
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200907#ifdef CONFIG_IA32_EMULATION
Ingo Molnar2cd23552015-06-08 08:28:07 +0200908 set_system_intr_gate(IA32_SYSCALL_VECTOR, entry_INT80_compat);
Yinghai Lubb3f0b52009-01-25 02:38:09 -0800909 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200910#endif
911
912#ifdef CONFIG_X86_32
Andy Lutomirskia798f092016-03-09 13:24:32 -0800913 set_system_intr_gate(IA32_SYSCALL_VECTOR, entry_INT80_32);
Brian Gerst51bb9282015-05-09 11:36:52 -0400914 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
Alexander van Heukelum081f75b2008-10-03 22:00:39 +0200915#endif
Yinghai Lubb3f0b52009-01-25 02:38:09 -0800916
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 /*
Kees Cook4eefbe72013-04-10 12:24:22 -0700918 * Set the IDT descriptor to a fixed read-only location, so that the
919 * "sidt" instruction will not leak the location of the kernel, and
920 * to defend the IDT against arbitrary memory write vulnerabilities.
921 * It will be reloaded in cpu_init() */
922 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
923 idt_descr.address = fix_to_virt(FIX_RO_IDT);
924
925 /*
Ingo Molnarb5964402008-02-26 11:15:50 +0100926 * Should be a barrier for any external CPU state:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 */
928 cpu_init();
929
Wang Nanb4d83272015-02-26 13:49:39 +0800930 /*
931 * X86_TRAP_DB and X86_TRAP_BP have been set
Wang Nan5eca7452015-02-27 12:19:49 +0800932 * in early_trap_init(). However, ITS works only after
Wang Nanb4d83272015-02-26 13:49:39 +0800933 * cpu_init() loads TSS. See comments in early_trap_init().
934 */
935 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
936 /* int3 can be called from all */
937 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
938
Thomas Gleixner428cf902009-08-20 10:35:46 +0200939 x86_init.irqs.trap_init();
Steven Rostedt228bdaa2011-12-09 03:02:19 -0500940
941#ifdef CONFIG_X86_64
Seiji Aguchi629f4f92013-06-20 11:45:44 -0400942 memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
Kees Cookc9408262012-03-09 16:07:10 -0800943 set_nmi_gate(X86_TRAP_DB, &debug);
944 set_nmi_gate(X86_TRAP_BP, &int3);
Steven Rostedt228bdaa2011-12-09 03:02:19 -0500945#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946}