x86_64, traps: Fix the espfix64 #DF fixup and rewrite it in C
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / x86 / kernel / traps.c
CommitLineData
1da177e4 1/*
1da177e4 2 * Copyright (C) 1991, 1992 Linus Torvalds
a8c1be9d 3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
1da177e4
LT
4 *
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 */
8
9/*
c1d518c8 10 * Handle hardware traps and faults.
1da177e4 11 */
c767a54b
JP
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
56dd9470 15#include <linux/context_tracking.h>
b5964405
IM
16#include <linux/interrupt.h>
17#include <linux/kallsyms.h>
18#include <linux/spinlock.h>
b5964405
IM
19#include <linux/kprobes.h>
20#include <linux/uaccess.h>
b5964405 21#include <linux/kdebug.h>
f503b5ae 22#include <linux/kgdb.h>
1da177e4 23#include <linux/kernel.h>
b5964405
IM
24#include <linux/module.h>
25#include <linux/ptrace.h>
1da177e4 26#include <linux/string.h>
b5964405 27#include <linux/delay.h>
1da177e4 28#include <linux/errno.h>
b5964405
IM
29#include <linux/kexec.h>
30#include <linux/sched.h>
1da177e4 31#include <linux/timer.h>
1da177e4 32#include <linux/init.h>
91768d6c 33#include <linux/bug.h>
b5964405
IM
34#include <linux/nmi.h>
35#include <linux/mm.h>
c1d518c8
AH
36#include <linux/smp.h>
37#include <linux/io.h>
1da177e4
LT
38
39#ifdef CONFIG_EISA
40#include <linux/ioport.h>
41#include <linux/eisa.h>
42#endif
43
c0d12172
DJ
44#if defined(CONFIG_EDAC)
45#include <linux/edac.h>
46#endif
47
f8561296 48#include <asm/kmemcheck.h>
b5964405 49#include <asm/stacktrace.h>
1da177e4 50#include <asm/processor.h>
1da177e4 51#include <asm/debugreg.h>
60063497 52#include <linux/atomic.h>
08d636b6 53#include <asm/ftrace.h>
c1d518c8 54#include <asm/traps.h>
1da177e4
LT
55#include <asm/desc.h>
56#include <asm/i387.h>
1361b83a 57#include <asm/fpu-internal.h>
9e55e44e 58#include <asm/mce.h>
4eefbe79 59#include <asm/fixmap.h>
1164dd00 60#include <asm/mach_traps.h>
c1d518c8 61
081f75bb 62#ifdef CONFIG_X86_64
428cf902 63#include <asm/x86_init.h>
081f75bb
AH
64#include <asm/pgalloc.h>
65#include <asm/proto.h>
081f75bb 66#else
c1d518c8 67#include <asm/processor-flags.h>
8e6dafd6 68#include <asm/setup.h>
1da177e4 69
1da177e4
LT
70asmlinkage int system_call(void);
71
1da177e4
LT
72/*
73 * The IDT has to be page-aligned to simplify the Pentium
07e81d61 74 * F0 0F bug workaround.
1da177e4 75 */
07e81d61 76gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
081f75bb 77#endif
1da177e4 78
b77b881f
YL
79DECLARE_BITMAP(used_vectors, NR_VECTORS);
80EXPORT_SYMBOL_GPL(used_vectors);
81
762db434
AH
82static inline void conditional_sti(struct pt_regs *regs)
83{
84 if (regs->flags & X86_EFLAGS_IF)
85 local_irq_enable();
86}
87
3d2a71a5
AH
88static inline void preempt_conditional_sti(struct pt_regs *regs)
89{
90 inc_preempt_count();
91 if (regs->flags & X86_EFLAGS_IF)
92 local_irq_enable();
93}
94
be716615
TG
95static inline void conditional_cli(struct pt_regs *regs)
96{
97 if (regs->flags & X86_EFLAGS_IF)
98 local_irq_disable();
99}
100
3d2a71a5
AH
101static inline void preempt_conditional_cli(struct pt_regs *regs)
102{
103 if (regs->flags & X86_EFLAGS_IF)
104 local_irq_disable();
105 dec_preempt_count();
106}
107
c416ddf5
FW
108static int __kprobes
109do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
110 struct pt_regs *regs, long error_code)
1da177e4 111{
081f75bb 112#ifdef CONFIG_X86_32
6b6891f9 113 if (regs->flags & X86_VM_MASK) {
3c1326f8 114 /*
c416ddf5 115 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
3c1326f8
AH
116 * On nmi (interrupt 2), do_trap should not be called.
117 */
c416ddf5
FW
118 if (trapnr < X86_TRAP_UD) {
119 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
120 error_code, trapnr))
121 return 0;
122 }
123 return -1;
1da177e4 124 }
081f75bb 125#endif
c416ddf5
FW
126 if (!user_mode(regs)) {
127 if (!fixup_exception(regs)) {
128 tsk->thread.error_code = error_code;
129 tsk->thread.trap_nr = trapnr;
130 die(str, regs, error_code);
131 }
132 return 0;
133 }
1da177e4 134
c416ddf5
FW
135 return -1;
136}
1da177e4 137
c416ddf5
FW
138static void __kprobes
139do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
140 long error_code, siginfo_t *info)
141{
142 struct task_struct *tsk = current;
143
144
145 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
146 return;
b5964405 147 /*
51e7dc70 148 * We want error_code and trap_nr set for userspace faults and
b5964405
IM
149 * kernelspace faults which result in die(), but not
150 * kernelspace faults which are fixed up. die() gives the
151 * process no chance to handle the signal and notice the
152 * kernel fault information, so that won't result in polluting
153 * the information about previously queued, but not yet
154 * delivered, faults. See also do_general_protection below.
155 */
156 tsk->thread.error_code = error_code;
51e7dc70 157 tsk->thread.trap_nr = trapnr;
d1895183 158
081f75bb
AH
159#ifdef CONFIG_X86_64
160 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
161 printk_ratelimit()) {
c767a54b
JP
162 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
163 tsk->comm, tsk->pid, str,
164 regs->ip, regs->sp, error_code);
081f75bb 165 print_vma_addr(" in ", regs->ip);
c767a54b 166 pr_cont("\n");
081f75bb
AH
167 }
168#endif
169
b5964405
IM
170 if (info)
171 force_sig_info(signr, info, tsk);
172 else
173 force_sig(signr, tsk);
1da177e4
LT
174}
175
b5964405 176#define DO_ERROR(trapnr, signr, str, name) \
e407d620 177dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
b5964405 178{ \
6c1e0256
FW
179 enum ctx_state prev_state; \
180 \
181 prev_state = exception_enter(); \
6ba3c97a
FW
182 if (notify_die(DIE_TRAP, str, regs, error_code, \
183 trapnr, signr) == NOTIFY_STOP) { \
6c1e0256 184 exception_exit(prev_state); \
b5964405 185 return; \
6ba3c97a 186 } \
61aef7d2 187 conditional_sti(regs); \
3c1326f8 188 do_trap(trapnr, signr, str, regs, error_code, NULL); \
6c1e0256 189 exception_exit(prev_state); \
1da177e4
LT
190}
191
3c1326f8 192#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
e407d620 193dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
b5964405
IM
194{ \
195 siginfo_t info; \
6c1e0256
FW
196 enum ctx_state prev_state; \
197 \
b5964405
IM
198 info.si_signo = signr; \
199 info.si_errno = 0; \
200 info.si_code = sicode; \
201 info.si_addr = (void __user *)siaddr; \
6c1e0256 202 prev_state = exception_enter(); \
6ba3c97a
FW
203 if (notify_die(DIE_TRAP, str, regs, error_code, \
204 trapnr, signr) == NOTIFY_STOP) { \
6c1e0256 205 exception_exit(prev_state); \
b5964405 206 return; \
6ba3c97a 207 } \
61aef7d2 208 conditional_sti(regs); \
3c1326f8 209 do_trap(trapnr, signr, str, regs, error_code, &info); \
6c1e0256 210 exception_exit(prev_state); \
1da177e4
LT
211}
212
c9408265
KC
213DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
214 regs->ip)
215DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
216DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
217DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
218 regs->ip)
219DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
220 coprocessor_segment_overrun)
221DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
222DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
081f75bb 223#ifdef CONFIG_X86_32
c9408265 224DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
081f75bb 225#endif
c9408265
KC
226DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
227 BUS_ADRALN, 0)
1da177e4 228
081f75bb
AH
229#ifdef CONFIG_X86_64
230/* Runs on IST stack */
231dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
232{
6c1e0256
FW
233 enum ctx_state prev_state;
234
235 prev_state = exception_enter();
081f75bb 236 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
6ba3c97a
FW
237 X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
238 preempt_conditional_sti(regs);
239 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
240 preempt_conditional_cli(regs);
241 }
6c1e0256 242 exception_exit(prev_state);
081f75bb
AH
243}
244
245dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
246{
247 static const char str[] = "double fault";
248 struct task_struct *tsk = current;
249
ea21cb1f
AL
250#ifdef CONFIG_X86_ESPFIX64
251 extern unsigned char native_irq_return_iret[];
252
253 /*
254 * If IRET takes a non-IST fault on the espfix64 stack, then we
255 * end up promoting it to a doublefault. In that case, modify
256 * the stack to make it look like we just entered the #GP
257 * handler from user space, similar to bad_iret.
258 */
259 if (((long)regs->sp >> PGDIR_SHIFT) == ESPFIX_PGD_ENTRY &&
260 regs->cs == __KERNEL_CS &&
261 regs->ip == (unsigned long)native_irq_return_iret)
262 {
263 struct pt_regs *normal_regs = task_pt_regs(current);
264
265 /* Fake a #GP(0) from userspace. */
266 memmove(&normal_regs->ip, (void *)regs->sp, 5*8);
267 normal_regs->orig_ax = 0; /* Missing (lost) #GP error code */
268 regs->ip = (unsigned long)general_protection;
269 regs->sp = (unsigned long)&normal_regs->orig_ax;
270 return;
271 }
272#endif
273
6c1e0256 274 exception_enter();
081f75bb 275 /* Return not checked because double check cannot be ignored */
c9408265 276 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
081f75bb
AH
277
278 tsk->thread.error_code = error_code;
51e7dc70 279 tsk->thread.trap_nr = X86_TRAP_DF;
081f75bb 280
bd8b96df
IM
281 /*
282 * This is always a kernel trap and never fixable (and thus must
283 * never return).
284 */
081f75bb
AH
285 for (;;)
286 die(str, regs, error_code);
287}
288#endif
289
e407d620 290dotraplinkage void __kprobes
13485ab5 291do_general_protection(struct pt_regs *regs, long error_code)
1da177e4 292{
13485ab5 293 struct task_struct *tsk;
6c1e0256 294 enum ctx_state prev_state;
b5964405 295
6c1e0256 296 prev_state = exception_enter();
c6df0d71
AH
297 conditional_sti(regs);
298
081f75bb 299#ifdef CONFIG_X86_32
ef3f6288
FW
300 if (regs->flags & X86_VM_MASK) {
301 local_irq_enable();
302 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
6ba3c97a 303 goto exit;
ef3f6288 304 }
081f75bb 305#endif
1da177e4 306
13485ab5 307 tsk = current;
ef3f6288
FW
308 if (!user_mode(regs)) {
309 if (fixup_exception(regs))
6ba3c97a 310 goto exit;
ef3f6288
FW
311
312 tsk->thread.error_code = error_code;
313 tsk->thread.trap_nr = X86_TRAP_GP;
6ba3c97a
FW
314 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
315 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
ef3f6288 316 die("general protection fault", regs, error_code);
6ba3c97a 317 goto exit;
ef3f6288 318 }
1da177e4 319
13485ab5 320 tsk->thread.error_code = error_code;
51e7dc70 321 tsk->thread.trap_nr = X86_TRAP_GP;
b5964405 322
13485ab5
AH
323 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
324 printk_ratelimit()) {
c767a54b 325 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
13485ab5
AH
326 tsk->comm, task_pid_nr(tsk),
327 regs->ip, regs->sp, error_code);
03252919 328 print_vma_addr(" in ", regs->ip);
c767a54b 329 pr_cont("\n");
03252919 330 }
abd4f750 331
13485ab5 332 force_sig(SIGSEGV, tsk);
6ba3c97a 333exit:
6c1e0256 334 exception_exit(prev_state);
1da177e4
LT
335}
336
c1d518c8 337/* May run on IST stack. */
08d636b6 338dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
1da177e4 339{
6c1e0256
FW
340 enum ctx_state prev_state;
341
08d636b6 342#ifdef CONFIG_DYNAMIC_FTRACE
a192cd04
SR
343 /*
344 * ftrace must be first, everything else may cause a recursive crash.
345 * See note by declaration of modifying_ftrace_code in ftrace.c
346 */
347 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
348 ftrace_int3_handler(regs))
08d636b6
SR
349 return;
350#endif
6c1e0256 351 prev_state = exception_enter();
f503b5ae 352#ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
c9408265
KC
353 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
354 SIGTRAP) == NOTIFY_STOP)
6ba3c97a 355 goto exit;
f503b5ae 356#endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
cc3a1bf5 357
c9408265
KC
358 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
359 SIGTRAP) == NOTIFY_STOP)
6ba3c97a 360 goto exit;
b5964405 361
42181186
SR
362 /*
363 * Let others (NMI) know that the debug stack is in use
364 * as we may switch to the interrupt stack.
365 */
366 debug_stack_usage_inc();
4915a35e 367 preempt_conditional_sti(regs);
c9408265 368 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
4915a35e 369 preempt_conditional_cli(regs);
42181186 370 debug_stack_usage_dec();
6ba3c97a 371exit:
6c1e0256 372 exception_exit(prev_state);
1da177e4 373}
1da177e4 374
081f75bb 375#ifdef CONFIG_X86_64
bd8b96df
IM
376/*
377 * Help handler running on IST stack to switch back to user stack
378 * for scheduling or signal handling. The actual stack switch is done in
379 * entry.S
380 */
081f75bb
AH
381asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
382{
383 struct pt_regs *regs = eregs;
384 /* Did already sync */
385 if (eregs == (struct pt_regs *)eregs->sp)
386 ;
387 /* Exception from user space */
388 else if (user_mode(eregs))
389 regs = task_pt_regs(current);
bd8b96df
IM
390 /*
391 * Exception from kernel and interrupts are enabled. Move to
392 * kernel process stack.
393 */
081f75bb
AH
394 else if (eregs->flags & X86_EFLAGS_IF)
395 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
396 if (eregs != regs)
397 *regs = *eregs;
398 return regs;
399}
400#endif
401
1da177e4
LT
402/*
403 * Our handling of the processor debug registers is non-trivial.
404 * We do not clear them on entry and exit from the kernel. Therefore
405 * it is possible to get a watchpoint trap here from inside the kernel.
406 * However, the code in ./ptrace.c has ensured that the user can
407 * only set watchpoints on userspace addresses. Therefore the in-kernel
408 * watchpoint trap can only occur in code which is reading/writing
409 * from user space. Such code must not hold kernel locks (since it
410 * can equally take a page fault), therefore it is safe to call
411 * force_sig_info even though that claims and releases locks.
b5964405 412 *
1da177e4
LT
413 * Code in ./signal.c ensures that the debug control register
414 * is restored before we deliver any signal, and therefore that
415 * user code runs with the correct debug control register even though
416 * we clear it here.
417 *
418 * Being careful here means that we don't have to be as careful in a
419 * lot of more complicated places (task switching can be a bit lazy
420 * about restoring all the debug state, and ptrace doesn't have to
421 * find every occurrence of the TF bit that could be saved away even
422 * by user code)
c1d518c8
AH
423 *
424 * May run on IST stack.
1da177e4 425 */
e407d620 426dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
1da177e4 427{
1da177e4 428 struct task_struct *tsk = current;
6c1e0256 429 enum ctx_state prev_state;
a1e80faf 430 int user_icebp = 0;
08d68323 431 unsigned long dr6;
da654b74 432 int si_code;
1da177e4 433
6c1e0256 434 prev_state = exception_enter();
6ba3c97a 435
08d68323 436 get_debugreg(dr6, 6);
1da177e4 437
40f9249a
P
438 /* Filter out all the reserved bits which are preset to 1 */
439 dr6 &= ~DR6_RESERVED;
440
a1e80faf
FW
441 /*
442 * If dr6 has no reason to give us about the origin of this trap,
443 * then it's very likely the result of an icebp/int01 trap.
444 * User wants a sigtrap for that.
445 */
446 if (!dr6 && user_mode(regs))
447 user_icebp = 1;
448
f8561296 449 /* Catch kmemcheck conditions first of all! */
eadb8a09 450 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
6ba3c97a 451 goto exit;
f8561296 452
08d68323
P
453 /* DR6 may or may not be cleared by the CPU */
454 set_debugreg(0, 6);
10faa81e 455
ea8e61b7
PZ
456 /*
457 * The processor cleared BTF, so don't mark that we need it set.
458 */
459 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
460
08d68323
P
461 /* Store the virtualized DR6 value */
462 tsk->thread.debugreg6 = dr6;
463
62edab90
P
464 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
465 SIGTRAP) == NOTIFY_STOP)
6ba3c97a 466 goto exit;
3d2a71a5 467
42181186
SR
468 /*
469 * Let others (NMI) know that the debug stack is in use
470 * as we may switch to the interrupt stack.
471 */
472 debug_stack_usage_inc();
473
1da177e4 474 /* It's safe to allow irq's after DR6 has been saved */
3d2a71a5 475 preempt_conditional_sti(regs);
1da177e4 476
08d68323 477 if (regs->flags & X86_VM_MASK) {
c9408265
KC
478 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
479 X86_TRAP_DB);
6554287b 480 preempt_conditional_cli(regs);
42181186 481 debug_stack_usage_dec();
6ba3c97a 482 goto exit;
1da177e4
LT
483 }
484
1da177e4 485 /*
08d68323
P
486 * Single-stepping through system calls: ignore any exceptions in
487 * kernel space, but re-enable TF when returning to user mode.
488 *
489 * We already checked v86 mode above, so we can check for kernel mode
490 * by just checking the CPL of CS.
1da177e4 491 */
08d68323
P
492 if ((dr6 & DR_STEP) && !user_mode(regs)) {
493 tsk->thread.debugreg6 &= ~DR_STEP;
494 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
495 regs->flags &= ~X86_EFLAGS_TF;
1da177e4 496 }
08d68323 497 si_code = get_si_code(tsk->thread.debugreg6);
a1e80faf 498 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
08d68323 499 send_sigtrap(tsk, regs, error_code, si_code);
3d2a71a5 500 preempt_conditional_cli(regs);
42181186 501 debug_stack_usage_dec();
1da177e4 502
6ba3c97a 503exit:
6c1e0256 504 exception_exit(prev_state);
1da177e4
LT
505}
506
507/*
508 * Note that we play around with the 'TS' bit in an attempt to get
509 * the correct behaviour even in the presence of the asynchronous
510 * IRQ13 behaviour
511 */
9b6dba9e 512void math_error(struct pt_regs *regs, int error_code, int trapnr)
1da177e4 513{
e2e75c91 514 struct task_struct *task = current;
1da177e4 515 siginfo_t info;
9b6dba9e 516 unsigned short err;
c9408265
KC
517 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
518 "simd exception";
e2e75c91
BG
519
520 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
521 return;
522 conditional_sti(regs);
523
524 if (!user_mode_vm(regs))
525 {
526 if (!fixup_exception(regs)) {
527 task->thread.error_code = error_code;
51e7dc70 528 task->thread.trap_nr = trapnr;
e2e75c91
BG
529 die(str, regs, error_code);
530 }
531 return;
532 }
1da177e4
LT
533
534 /*
535 * Save the info for the exception handler and clear the error.
536 */
1da177e4 537 save_init_fpu(task);
51e7dc70 538 task->thread.trap_nr = trapnr;
9b6dba9e 539 task->thread.error_code = error_code;
1da177e4
LT
540 info.si_signo = SIGFPE;
541 info.si_errno = 0;
9b6dba9e 542 info.si_addr = (void __user *)regs->ip;
c9408265 543 if (trapnr == X86_TRAP_MF) {
9b6dba9e
BG
544 unsigned short cwd, swd;
545 /*
546 * (~cwd & swd) will mask out exceptions that are not set to unmasked
547 * status. 0x3f is the exception bits in these regs, 0x200 is the
548 * C1 reg you need in case of a stack fault, 0x040 is the stack
549 * fault bit. We should only be taking one exception at a time,
550 * so if this combination doesn't produce any single exception,
551 * then we have a bad program that isn't synchronizing its FPU usage
552 * and it will suffer the consequences since we won't be able to
553 * fully reproduce the context of the exception
554 */
555 cwd = get_fpu_cwd(task);
556 swd = get_fpu_swd(task);
adf77bac 557
9b6dba9e
BG
558 err = swd & ~cwd;
559 } else {
560 /*
561 * The SIMD FPU exceptions are handled a little differently, as there
562 * is only a single status/control register. Thus, to determine which
563 * unmasked exception was caught we must mask the exception mask bits
564 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
565 */
566 unsigned short mxcsr = get_fpu_mxcsr(task);
567 err = ~(mxcsr >> 7) & mxcsr;
568 }
adf77bac
PA
569
570 if (err & 0x001) { /* Invalid op */
b5964405
IM
571 /*
572 * swd & 0x240 == 0x040: Stack Underflow
573 * swd & 0x240 == 0x240: Stack Overflow
574 * User must clear the SF bit (0x40) if set
575 */
576 info.si_code = FPE_FLTINV;
adf77bac 577 } else if (err & 0x004) { /* Divide by Zero */
b5964405 578 info.si_code = FPE_FLTDIV;
adf77bac 579 } else if (err & 0x008) { /* Overflow */
b5964405 580 info.si_code = FPE_FLTOVF;
adf77bac
PA
581 } else if (err & 0x012) { /* Denormal, Underflow */
582 info.si_code = FPE_FLTUND;
583 } else if (err & 0x020) { /* Precision */
b5964405 584 info.si_code = FPE_FLTRES;
adf77bac 585 } else {
bd8b96df 586 /*
c9408265
KC
587 * If we're using IRQ 13, or supposedly even some trap
588 * X86_TRAP_MF implementations, it's possible
589 * we get a spurious trap, which is not an error.
bd8b96df 590 */
c9408265 591 return;
1da177e4
LT
592 }
593 force_sig_info(SIGFPE, &info, task);
594}
595
e407d620 596dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
1da177e4 597{
6c1e0256
FW
598 enum ctx_state prev_state;
599
600 prev_state = exception_enter();
c9408265 601 math_error(regs, error_code, X86_TRAP_MF);
6c1e0256 602 exception_exit(prev_state);
1da177e4
LT
603}
604
e407d620
AH
605dotraplinkage void
606do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
1da177e4 607{
6c1e0256
FW
608 enum ctx_state prev_state;
609
610 prev_state = exception_enter();
c9408265 611 math_error(regs, error_code, X86_TRAP_XF);
6c1e0256 612 exception_exit(prev_state);
1da177e4
LT
613}
614
e407d620
AH
615dotraplinkage void
616do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
1da177e4 617{
cf81978d 618 conditional_sti(regs);
1da177e4
LT
619#if 0
620 /* No need to warn about this any longer. */
c767a54b 621 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1da177e4
LT
622#endif
623}
624
081f75bb 625asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1da177e4 626{
1da177e4 627}
4efc0670 628
7856f6cc 629asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
081f75bb
AH
630{
631}
632
1da177e4 633/*
b5964405 634 * 'math_state_restore()' saves the current math information in the
1da177e4
LT
635 * old math state array, and gets the new ones from the current task
636 *
637 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
638 * Don't touch unless you *really* know how it works.
639 *
be98c2cd
LT
640 * Must be called with kernel preemption disabled (eg with local
641 * local interrupts as in the case of do_device_not_available).
1da177e4 642 */
be98c2cd 643void math_state_restore(void)
1da177e4 644{
f94edacf 645 struct task_struct *tsk = current;
1da177e4 646
aa283f49
SS
647 if (!tsk_used_math(tsk)) {
648 local_irq_enable();
649 /*
650 * does a slab alloc which can sleep
651 */
652 if (init_fpu(tsk)) {
653 /*
654 * ran out of memory!
655 */
656 do_group_exit(SIGKILL);
657 return;
658 }
659 local_irq_disable();
660 }
661
f94edacf 662 __thread_fpu_begin(tsk);
304bceda 663
80ab6f1e
LT
664 /*
665 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
666 */
667 if (unlikely(restore_fpu_checking(tsk))) {
304bceda 668 drop_init_fpu(tsk);
80ab6f1e
LT
669 force_sig(SIGSEGV, tsk);
670 return;
671 }
b3b0870e
LT
672
673 tsk->fpu_counter++;
1da177e4 674}
5992b6da 675EXPORT_SYMBOL_GPL(math_state_restore);
1da177e4 676
e407d620 677dotraplinkage void __kprobes
aa78bcfa 678do_device_not_available(struct pt_regs *regs, long error_code)
7643e9b9 679{
6c1e0256
FW
680 enum ctx_state prev_state;
681
682 prev_state = exception_enter();
5d2bd700 683 BUG_ON(use_eager_fpu());
304bceda 684
a334fe43 685#ifdef CONFIG_MATH_EMULATION
7643e9b9 686 if (read_cr0() & X86_CR0_EM) {
d315760f
TH
687 struct math_emu_info info = { };
688
7643e9b9 689 conditional_sti(regs);
d315760f 690
aa78bcfa 691 info.regs = regs;
d315760f 692 math_emulate(&info);
6c1e0256 693 exception_exit(prev_state);
a334fe43 694 return;
7643e9b9 695 }
a334fe43
BG
696#endif
697 math_state_restore(); /* interrupts still off */
698#ifdef CONFIG_X86_32
699 conditional_sti(regs);
081f75bb 700#endif
6c1e0256 701 exception_exit(prev_state);
7643e9b9
AH
702}
703
081f75bb 704#ifdef CONFIG_X86_32
e407d620 705dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
f8e0870f
AH
706{
707 siginfo_t info;
6c1e0256 708 enum ctx_state prev_state;
6ba3c97a 709
6c1e0256 710 prev_state = exception_enter();
f8e0870f
AH
711 local_irq_enable();
712
713 info.si_signo = SIGILL;
714 info.si_errno = 0;
715 info.si_code = ILL_BADSTK;
fc6fcdfb 716 info.si_addr = NULL;
c9408265 717 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
6ba3c97a
FW
718 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
719 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
720 &info);
721 }
6c1e0256 722 exception_exit(prev_state);
f8e0870f 723}
081f75bb 724#endif
f8e0870f 725
29c84391
JK
726/* Set of traps needed for early debugging. */
727void __init early_trap_init(void)
728{
c9408265 729 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
29c84391 730 /* int3 can be called from all */
c9408265 731 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
8170e6be 732#ifdef CONFIG_X86_32
c9408265 733 set_intr_gate(X86_TRAP_PF, &page_fault);
8170e6be 734#endif
29c84391
JK
735 load_idt(&idt_descr);
736}
737
8170e6be
PA
738void __init early_trap_pf_init(void)
739{
740#ifdef CONFIG_X86_64
741 set_intr_gate(X86_TRAP_PF, &page_fault);
742#endif
743}
744
1da177e4
LT
745void __init trap_init(void)
746{
dbeb2be2
RR
747 int i;
748
1da177e4 749#ifdef CONFIG_EISA
927222b1 750 void __iomem *p = early_ioremap(0x0FFFD9, 4);
b5964405
IM
751
752 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
1da177e4 753 EISA_bus = 1;
927222b1 754 early_iounmap(p, 4);
1da177e4
LT
755#endif
756
c9408265
KC
757 set_intr_gate(X86_TRAP_DE, &divide_error);
758 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
699d2937 759 /* int4 can be called from all */
c9408265
KC
760 set_system_intr_gate(X86_TRAP_OF, &overflow);
761 set_intr_gate(X86_TRAP_BR, &bounds);
762 set_intr_gate(X86_TRAP_UD, &invalid_op);
763 set_intr_gate(X86_TRAP_NM, &device_not_available);
081f75bb 764#ifdef CONFIG_X86_32
c9408265 765 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
081f75bb 766#else
c9408265 767 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
081f75bb 768#endif
c9408265
KC
769 set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
770 set_intr_gate(X86_TRAP_TS, &invalid_TSS);
771 set_intr_gate(X86_TRAP_NP, &segment_not_present);
772 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
773 set_intr_gate(X86_TRAP_GP, &general_protection);
774 set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
775 set_intr_gate(X86_TRAP_MF, &coprocessor_error);
776 set_intr_gate(X86_TRAP_AC, &alignment_check);
1da177e4 777#ifdef CONFIG_X86_MCE
c9408265 778 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
1da177e4 779#endif
c9408265 780 set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
1da177e4 781
bb3f0b59
YL
782 /* Reserve all the builtin and the syscall vector: */
783 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
784 set_bit(i, used_vectors);
785
081f75bb
AH
786#ifdef CONFIG_IA32_EMULATION
787 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
bb3f0b59 788 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
081f75bb
AH
789#endif
790
791#ifdef CONFIG_X86_32
699d2937 792 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
dbeb2be2 793 set_bit(SYSCALL_VECTOR, used_vectors);
081f75bb 794#endif
bb3f0b59 795
4eefbe79
KC
796 /*
797 * Set the IDT descriptor to a fixed read-only location, so that the
798 * "sidt" instruction will not leak the location of the kernel, and
799 * to defend the IDT against arbitrary memory write vulnerabilities.
800 * It will be reloaded in cpu_init() */
801 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
802 idt_descr.address = fix_to_virt(FIX_RO_IDT);
803
1da177e4 804 /*
b5964405 805 * Should be a barrier for any external CPU state:
1da177e4
LT
806 */
807 cpu_init();
808
428cf902 809 x86_init.irqs.trap_init();
228bdaa9
SR
810
811#ifdef CONFIG_X86_64
812 memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
c9408265
KC
813 set_nmi_gate(X86_TRAP_DB, &debug);
814 set_nmi_gate(X86_TRAP_BP, &int3);
228bdaa9 815#endif
1da177e4 816}