]>
Commit | Line | Data |
---|---|---|
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 | */ |
b5964405 IM |
12 | #include <linux/interrupt.h> |
13 | #include <linux/kallsyms.h> | |
14 | #include <linux/spinlock.h> | |
b5964405 IM |
15 | #include <linux/kprobes.h> |
16 | #include <linux/uaccess.h> | |
b5964405 | 17 | #include <linux/kdebug.h> |
1da177e4 | 18 | #include <linux/kernel.h> |
b5964405 IM |
19 | #include <linux/module.h> |
20 | #include <linux/ptrace.h> | |
1da177e4 | 21 | #include <linux/string.h> |
b5964405 | 22 | #include <linux/delay.h> |
1da177e4 | 23 | #include <linux/errno.h> |
b5964405 IM |
24 | #include <linux/kexec.h> |
25 | #include <linux/sched.h> | |
1da177e4 | 26 | #include <linux/timer.h> |
1da177e4 | 27 | #include <linux/init.h> |
91768d6c | 28 | #include <linux/bug.h> |
b5964405 IM |
29 | #include <linux/nmi.h> |
30 | #include <linux/mm.h> | |
c1d518c8 AH |
31 | #include <linux/smp.h> |
32 | #include <linux/io.h> | |
1da177e4 LT |
33 | |
34 | #ifdef CONFIG_EISA | |
35 | #include <linux/ioport.h> | |
36 | #include <linux/eisa.h> | |
37 | #endif | |
38 | ||
39 | #ifdef CONFIG_MCA | |
40 | #include <linux/mca.h> | |
41 | #endif | |
42 | ||
c0d12172 DJ |
43 | #if defined(CONFIG_EDAC) |
44 | #include <linux/edac.h> | |
45 | #endif | |
46 | ||
f8561296 | 47 | #include <asm/kmemcheck.h> |
b5964405 | 48 | #include <asm/stacktrace.h> |
1da177e4 | 49 | #include <asm/processor.h> |
1da177e4 | 50 | #include <asm/debugreg.h> |
b5964405 IM |
51 | #include <asm/atomic.h> |
52 | #include <asm/system.h> | |
c1d518c8 | 53 | #include <asm/traps.h> |
1da177e4 LT |
54 | #include <asm/desc.h> |
55 | #include <asm/i387.h> | |
9e55e44e | 56 | #include <asm/mce.h> |
c1d518c8 | 57 | |
1164dd00 | 58 | #include <asm/mach_traps.h> |
c1d518c8 | 59 | |
081f75bb | 60 | #ifdef CONFIG_X86_64 |
428cf902 | 61 | #include <asm/x86_init.h> |
081f75bb AH |
62 | #include <asm/pgalloc.h> |
63 | #include <asm/proto.h> | |
081f75bb | 64 | #else |
c1d518c8 | 65 | #include <asm/processor-flags.h> |
8e6dafd6 | 66 | #include <asm/setup.h> |
1da177e4 | 67 | |
1da177e4 LT |
68 | asmlinkage int system_call(void); |
69 | ||
1da177e4 | 70 | /* Do we ignore FPU interrupts ? */ |
b5964405 | 71 | char ignore_fpu_irq; |
1da177e4 LT |
72 | |
73 | /* | |
74 | * The IDT has to be page-aligned to simplify the Pentium | |
07e81d61 | 75 | * F0 0F bug workaround. |
1da177e4 | 76 | */ |
07e81d61 | 77 | gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, }; |
081f75bb | 78 | #endif |
1da177e4 | 79 | |
b77b881f YL |
80 | DECLARE_BITMAP(used_vectors, NR_VECTORS); |
81 | EXPORT_SYMBOL_GPL(used_vectors); | |
82 | ||
badc7652 | 83 | static int ignore_nmis; |
e041c683 | 84 | |
762db434 AH |
85 | static inline void conditional_sti(struct pt_regs *regs) |
86 | { | |
87 | if (regs->flags & X86_EFLAGS_IF) | |
88 | local_irq_enable(); | |
89 | } | |
90 | ||
3d2a71a5 AH |
91 | static inline void preempt_conditional_sti(struct pt_regs *regs) |
92 | { | |
93 | inc_preempt_count(); | |
94 | if (regs->flags & X86_EFLAGS_IF) | |
95 | local_irq_enable(); | |
96 | } | |
97 | ||
be716615 TG |
98 | static inline void conditional_cli(struct pt_regs *regs) |
99 | { | |
100 | if (regs->flags & X86_EFLAGS_IF) | |
101 | local_irq_disable(); | |
102 | } | |
103 | ||
3d2a71a5 AH |
104 | static inline void preempt_conditional_cli(struct pt_regs *regs) |
105 | { | |
106 | if (regs->flags & X86_EFLAGS_IF) | |
107 | local_irq_disable(); | |
108 | dec_preempt_count(); | |
109 | } | |
110 | ||
b5964405 | 111 | static void __kprobes |
3c1326f8 | 112 | do_trap(int trapnr, int signr, char *str, struct pt_regs *regs, |
b5964405 | 113 | long error_code, siginfo_t *info) |
1da177e4 | 114 | { |
4f339ecb | 115 | struct task_struct *tsk = current; |
4f339ecb | 116 | |
081f75bb | 117 | #ifdef CONFIG_X86_32 |
6b6891f9 | 118 | if (regs->flags & X86_VM_MASK) { |
3c1326f8 AH |
119 | /* |
120 | * traps 0, 1, 3, 4, and 5 should be forwarded to vm86. | |
121 | * On nmi (interrupt 2), do_trap should not be called. | |
122 | */ | |
123 | if (trapnr < 6) | |
1da177e4 LT |
124 | goto vm86_trap; |
125 | goto trap_signal; | |
126 | } | |
081f75bb | 127 | #endif |
1da177e4 | 128 | |
717b594a | 129 | if (!user_mode(regs)) |
1da177e4 LT |
130 | goto kernel_trap; |
131 | ||
081f75bb | 132 | #ifdef CONFIG_X86_32 |
b5964405 | 133 | trap_signal: |
081f75bb | 134 | #endif |
b5964405 IM |
135 | /* |
136 | * We want error_code and trap_no set for userspace faults and | |
137 | * kernelspace faults which result in die(), but not | |
138 | * kernelspace faults which are fixed up. die() gives the | |
139 | * process no chance to handle the signal and notice the | |
140 | * kernel fault information, so that won't result in polluting | |
141 | * the information about previously queued, but not yet | |
142 | * delivered, faults. See also do_general_protection below. | |
143 | */ | |
144 | tsk->thread.error_code = error_code; | |
145 | tsk->thread.trap_no = trapnr; | |
d1895183 | 146 | |
081f75bb AH |
147 | #ifdef CONFIG_X86_64 |
148 | if (show_unhandled_signals && unhandled_signal(tsk, signr) && | |
149 | printk_ratelimit()) { | |
150 | printk(KERN_INFO | |
151 | "%s[%d] trap %s ip:%lx sp:%lx error:%lx", | |
152 | tsk->comm, tsk->pid, str, | |
153 | regs->ip, regs->sp, error_code); | |
154 | print_vma_addr(" in ", regs->ip); | |
155 | printk("\n"); | |
156 | } | |
157 | #endif | |
158 | ||
b5964405 IM |
159 | if (info) |
160 | force_sig_info(signr, info, tsk); | |
161 | else | |
162 | force_sig(signr, tsk); | |
163 | return; | |
1da177e4 | 164 | |
b5964405 IM |
165 | kernel_trap: |
166 | if (!fixup_exception(regs)) { | |
167 | tsk->thread.error_code = error_code; | |
168 | tsk->thread.trap_no = trapnr; | |
169 | die(str, regs, error_code); | |
1da177e4 | 170 | } |
b5964405 | 171 | return; |
1da177e4 | 172 | |
081f75bb | 173 | #ifdef CONFIG_X86_32 |
b5964405 IM |
174 | vm86_trap: |
175 | if (handle_vm86_trap((struct kernel_vm86_regs *) regs, | |
176 | error_code, trapnr)) | |
177 | goto trap_signal; | |
178 | return; | |
081f75bb | 179 | #endif |
1da177e4 LT |
180 | } |
181 | ||
b5964405 | 182 | #define DO_ERROR(trapnr, signr, str, name) \ |
e407d620 | 183 | dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \ |
b5964405 IM |
184 | { \ |
185 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
a8c1be9d | 186 | == NOTIFY_STOP) \ |
b5964405 | 187 | return; \ |
61aef7d2 | 188 | conditional_sti(regs); \ |
3c1326f8 | 189 | do_trap(trapnr, signr, str, regs, error_code, NULL); \ |
1da177e4 LT |
190 | } |
191 | ||
3c1326f8 | 192 | #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \ |
e407d620 | 193 | dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \ |
b5964405 IM |
194 | { \ |
195 | siginfo_t info; \ | |
196 | info.si_signo = signr; \ | |
197 | info.si_errno = 0; \ | |
198 | info.si_code = sicode; \ | |
199 | info.si_addr = (void __user *)siaddr; \ | |
b5964405 | 200 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ |
a8c1be9d | 201 | == NOTIFY_STOP) \ |
b5964405 | 202 | return; \ |
61aef7d2 | 203 | conditional_sti(regs); \ |
3c1326f8 | 204 | do_trap(trapnr, signr, str, regs, error_code, &info); \ |
1da177e4 LT |
205 | } |
206 | ||
3c1326f8 AH |
207 | DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip) |
208 | DO_ERROR(4, SIGSEGV, "overflow", overflow) | |
209 | DO_ERROR(5, SIGSEGV, "bounds", bounds) | |
210 | DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip) | |
51bc1ed6 | 211 | DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun) |
6bf77bf9 | 212 | DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS) |
36d936c7 | 213 | DO_ERROR(11, SIGBUS, "segment not present", segment_not_present) |
081f75bb | 214 | #ifdef CONFIG_X86_32 |
f5ca8187 | 215 | DO_ERROR(12, SIGBUS, "stack segment", stack_segment) |
081f75bb | 216 | #endif |
3c1326f8 | 217 | DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0) |
1da177e4 | 218 | |
081f75bb AH |
219 | #ifdef CONFIG_X86_64 |
220 | /* Runs on IST stack */ | |
221 | dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code) | |
222 | { | |
223 | if (notify_die(DIE_TRAP, "stack segment", regs, error_code, | |
224 | 12, SIGBUS) == NOTIFY_STOP) | |
225 | return; | |
226 | preempt_conditional_sti(regs); | |
227 | do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL); | |
228 | preempt_conditional_cli(regs); | |
229 | } | |
230 | ||
231 | dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code) | |
232 | { | |
233 | static const char str[] = "double fault"; | |
234 | struct task_struct *tsk = current; | |
235 | ||
236 | /* Return not checked because double check cannot be ignored */ | |
237 | notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV); | |
238 | ||
239 | tsk->thread.error_code = error_code; | |
240 | tsk->thread.trap_no = 8; | |
241 | ||
bd8b96df IM |
242 | /* |
243 | * This is always a kernel trap and never fixable (and thus must | |
244 | * never return). | |
245 | */ | |
081f75bb AH |
246 | for (;;) |
247 | die(str, regs, error_code); | |
248 | } | |
249 | #endif | |
250 | ||
e407d620 | 251 | dotraplinkage void __kprobes |
13485ab5 | 252 | do_general_protection(struct pt_regs *regs, long error_code) |
1da177e4 | 253 | { |
13485ab5 | 254 | struct task_struct *tsk; |
b5964405 | 255 | |
c6df0d71 AH |
256 | conditional_sti(regs); |
257 | ||
081f75bb | 258 | #ifdef CONFIG_X86_32 |
6b6891f9 | 259 | if (regs->flags & X86_VM_MASK) |
1da177e4 | 260 | goto gp_in_vm86; |
081f75bb | 261 | #endif |
1da177e4 | 262 | |
13485ab5 | 263 | tsk = current; |
717b594a | 264 | if (!user_mode(regs)) |
1da177e4 LT |
265 | goto gp_in_kernel; |
266 | ||
13485ab5 AH |
267 | tsk->thread.error_code = error_code; |
268 | tsk->thread.trap_no = 13; | |
b5964405 | 269 | |
13485ab5 AH |
270 | if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) && |
271 | printk_ratelimit()) { | |
abd4f750 | 272 | printk(KERN_INFO |
13485ab5 AH |
273 | "%s[%d] general protection ip:%lx sp:%lx error:%lx", |
274 | tsk->comm, task_pid_nr(tsk), | |
275 | regs->ip, regs->sp, error_code); | |
03252919 AK |
276 | print_vma_addr(" in ", regs->ip); |
277 | printk("\n"); | |
278 | } | |
abd4f750 | 279 | |
13485ab5 | 280 | force_sig(SIGSEGV, tsk); |
1da177e4 LT |
281 | return; |
282 | ||
081f75bb | 283 | #ifdef CONFIG_X86_32 |
1da177e4 LT |
284 | gp_in_vm86: |
285 | local_irq_enable(); | |
286 | handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code); | |
287 | return; | |
081f75bb | 288 | #endif |
1da177e4 LT |
289 | |
290 | gp_in_kernel: | |
13485ab5 AH |
291 | if (fixup_exception(regs)) |
292 | return; | |
293 | ||
294 | tsk->thread.error_code = error_code; | |
295 | tsk->thread.trap_no = 13; | |
296 | if (notify_die(DIE_GPF, "general protection fault", regs, | |
1da177e4 | 297 | error_code, 13, SIGSEGV) == NOTIFY_STOP) |
13485ab5 AH |
298 | return; |
299 | die("general protection fault", regs, error_code); | |
1da177e4 LT |
300 | } |
301 | ||
5deb45e3 | 302 | static notrace __kprobes void |
b5964405 | 303 | mem_parity_error(unsigned char reason, struct pt_regs *regs) |
1da177e4 | 304 | { |
b5964405 IM |
305 | printk(KERN_EMERG |
306 | "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", | |
307 | reason, smp_processor_id()); | |
308 | ||
309 | printk(KERN_EMERG | |
310 | "You have some hardware problem, likely on the PCI bus.\n"); | |
c0d12172 DJ |
311 | |
312 | #if defined(CONFIG_EDAC) | |
b5964405 | 313 | if (edac_handler_set()) { |
c0d12172 DJ |
314 | edac_atomic_assert_error(); |
315 | return; | |
316 | } | |
317 | #endif | |
318 | ||
8da5adda | 319 | if (panic_on_unrecovered_nmi) |
b5964405 | 320 | panic("NMI: Not continuing"); |
1da177e4 | 321 | |
c41c5cd3 | 322 | printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); |
1da177e4 LT |
323 | |
324 | /* Clear and disable the memory parity error line. */ | |
7970479c AH |
325 | reason = (reason & 0xf) | 4; |
326 | outb(reason, 0x61); | |
1da177e4 LT |
327 | } |
328 | ||
5deb45e3 | 329 | static notrace __kprobes void |
b5964405 | 330 | io_check_error(unsigned char reason, struct pt_regs *regs) |
1da177e4 LT |
331 | { |
332 | unsigned long i; | |
333 | ||
9c107805 | 334 | printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n"); |
1da177e4 LT |
335 | show_registers(regs); |
336 | ||
5211a242 KG |
337 | if (panic_on_io_nmi) |
338 | panic("NMI IOCK error: Not continuing"); | |
339 | ||
1da177e4 LT |
340 | /* Re-enable the IOCK line, wait for a few seconds */ |
341 | reason = (reason & 0xf) | 8; | |
342 | outb(reason, 0x61); | |
b5964405 | 343 | |
1da177e4 | 344 | i = 2000; |
b5964405 IM |
345 | while (--i) |
346 | udelay(1000); | |
347 | ||
1da177e4 LT |
348 | reason &= ~8; |
349 | outb(reason, 0x61); | |
350 | } | |
351 | ||
5deb45e3 | 352 | static notrace __kprobes void |
b5964405 | 353 | unknown_nmi_error(unsigned char reason, struct pt_regs *regs) |
1da177e4 | 354 | { |
c1d518c8 AH |
355 | if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == |
356 | NOTIFY_STOP) | |
d3597524 | 357 | return; |
1da177e4 | 358 | #ifdef CONFIG_MCA |
b5964405 IM |
359 | /* |
360 | * Might actually be able to figure out what the guilty party | |
361 | * is: | |
362 | */ | |
363 | if (MCA_bus) { | |
1da177e4 LT |
364 | mca_handle_nmi(); |
365 | return; | |
366 | } | |
367 | #endif | |
b5964405 IM |
368 | printk(KERN_EMERG |
369 | "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", | |
370 | reason, smp_processor_id()); | |
371 | ||
c41c5cd3 | 372 | printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n"); |
8da5adda | 373 | if (panic_on_unrecovered_nmi) |
b5964405 | 374 | panic("NMI: Not continuing"); |
8da5adda | 375 | |
c41c5cd3 | 376 | printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); |
1da177e4 LT |
377 | } |
378 | ||
5deb45e3 | 379 | static notrace __kprobes void default_do_nmi(struct pt_regs *regs) |
1da177e4 LT |
380 | { |
381 | unsigned char reason = 0; | |
abd34807 AH |
382 | int cpu; |
383 | ||
384 | cpu = smp_processor_id(); | |
1da177e4 | 385 | |
abd34807 AH |
386 | /* Only the BSP gets external NMIs from the system. */ |
387 | if (!cpu) | |
1da177e4 | 388 | reason = get_nmi_reason(); |
b5964405 | 389 | |
1da177e4 | 390 | if (!(reason & 0xc0)) { |
20c0d2d4 | 391 | if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT) |
a8c1be9d | 392 | == NOTIFY_STOP) |
1da177e4 LT |
393 | return; |
394 | #ifdef CONFIG_X86_LOCAL_APIC | |
395 | /* | |
396 | * Ok, so this is none of the documented NMI sources, | |
397 | * so it must be the NMI watchdog. | |
398 | */ | |
3adbbcce | 399 | if (nmi_watchdog_tick(regs, reason)) |
1da177e4 | 400 | return; |
abd34807 | 401 | if (!do_nmi_callback(regs, cpu)) |
3adbbcce | 402 | unknown_nmi_error(reason, regs); |
b5964405 IM |
403 | #else |
404 | unknown_nmi_error(reason, regs); | |
405 | #endif | |
2fbe7b25 | 406 | |
1da177e4 LT |
407 | return; |
408 | } | |
20c0d2d4 | 409 | if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) |
1da177e4 | 410 | return; |
a8c1be9d AH |
411 | |
412 | /* AK: following checks seem to be broken on modern chipsets. FIXME */ | |
1da177e4 LT |
413 | if (reason & 0x80) |
414 | mem_parity_error(reason, regs); | |
415 | if (reason & 0x40) | |
416 | io_check_error(reason, regs); | |
081f75bb | 417 | #ifdef CONFIG_X86_32 |
1da177e4 LT |
418 | /* |
419 | * Reassert NMI in case it became active meanwhile | |
b5964405 | 420 | * as it's edge-triggered: |
1da177e4 LT |
421 | */ |
422 | reassert_nmi(); | |
081f75bb | 423 | #endif |
1da177e4 LT |
424 | } |
425 | ||
e407d620 AH |
426 | dotraplinkage notrace __kprobes void |
427 | do_nmi(struct pt_regs *regs, long error_code) | |
1da177e4 | 428 | { |
1da177e4 LT |
429 | nmi_enter(); |
430 | ||
915b0d01 | 431 | inc_irq_stat(__nmi_count); |
1da177e4 | 432 | |
8f4e956b AK |
433 | if (!ignore_nmis) |
434 | default_do_nmi(regs); | |
1da177e4 LT |
435 | |
436 | nmi_exit(); | |
437 | } | |
438 | ||
8f4e956b AK |
439 | void stop_nmi(void) |
440 | { | |
441 | acpi_nmi_disable(); | |
442 | ignore_nmis++; | |
443 | } | |
444 | ||
445 | void restart_nmi(void) | |
446 | { | |
447 | ignore_nmis--; | |
448 | acpi_nmi_enable(); | |
449 | } | |
450 | ||
c1d518c8 | 451 | /* May run on IST stack. */ |
e407d620 | 452 | dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code) |
1da177e4 | 453 | { |
b94da1e4 | 454 | #ifdef CONFIG_KPROBES |
1da177e4 LT |
455 | if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) |
456 | == NOTIFY_STOP) | |
48c88211 | 457 | return; |
b94da1e4 AH |
458 | #else |
459 | if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP) | |
460 | == NOTIFY_STOP) | |
461 | return; | |
462 | #endif | |
b5964405 | 463 | |
4915a35e | 464 | preempt_conditional_sti(regs); |
3c1326f8 | 465 | do_trap(3, SIGTRAP, "int3", regs, error_code, NULL); |
4915a35e | 466 | preempt_conditional_cli(regs); |
1da177e4 | 467 | } |
1da177e4 | 468 | |
081f75bb | 469 | #ifdef CONFIG_X86_64 |
bd8b96df IM |
470 | /* |
471 | * Help handler running on IST stack to switch back to user stack | |
472 | * for scheduling or signal handling. The actual stack switch is done in | |
473 | * entry.S | |
474 | */ | |
081f75bb AH |
475 | asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs) |
476 | { | |
477 | struct pt_regs *regs = eregs; | |
478 | /* Did already sync */ | |
479 | if (eregs == (struct pt_regs *)eregs->sp) | |
480 | ; | |
481 | /* Exception from user space */ | |
482 | else if (user_mode(eregs)) | |
483 | regs = task_pt_regs(current); | |
bd8b96df IM |
484 | /* |
485 | * Exception from kernel and interrupts are enabled. Move to | |
486 | * kernel process stack. | |
487 | */ | |
081f75bb AH |
488 | else if (eregs->flags & X86_EFLAGS_IF) |
489 | regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs)); | |
490 | if (eregs != regs) | |
491 | *regs = *eregs; | |
492 | return regs; | |
493 | } | |
494 | #endif | |
495 | ||
1da177e4 LT |
496 | /* |
497 | * Our handling of the processor debug registers is non-trivial. | |
498 | * We do not clear them on entry and exit from the kernel. Therefore | |
499 | * it is possible to get a watchpoint trap here from inside the kernel. | |
500 | * However, the code in ./ptrace.c has ensured that the user can | |
501 | * only set watchpoints on userspace addresses. Therefore the in-kernel | |
502 | * watchpoint trap can only occur in code which is reading/writing | |
503 | * from user space. Such code must not hold kernel locks (since it | |
504 | * can equally take a page fault), therefore it is safe to call | |
505 | * force_sig_info even though that claims and releases locks. | |
b5964405 | 506 | * |
1da177e4 LT |
507 | * Code in ./signal.c ensures that the debug control register |
508 | * is restored before we deliver any signal, and therefore that | |
509 | * user code runs with the correct debug control register even though | |
510 | * we clear it here. | |
511 | * | |
512 | * Being careful here means that we don't have to be as careful in a | |
513 | * lot of more complicated places (task switching can be a bit lazy | |
514 | * about restoring all the debug state, and ptrace doesn't have to | |
515 | * find every occurrence of the TF bit that could be saved away even | |
516 | * by user code) | |
c1d518c8 AH |
517 | * |
518 | * May run on IST stack. | |
1da177e4 | 519 | */ |
e407d620 | 520 | dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code) |
1da177e4 | 521 | { |
1da177e4 | 522 | struct task_struct *tsk = current; |
08d68323 | 523 | unsigned long dr6; |
da654b74 | 524 | int si_code; |
1da177e4 | 525 | |
08d68323 | 526 | get_debugreg(dr6, 6); |
1da177e4 | 527 | |
40f9249a P |
528 | /* Filter out all the reserved bits which are preset to 1 */ |
529 | dr6 &= ~DR6_RESERVED; | |
530 | ||
f8561296 | 531 | /* Catch kmemcheck conditions first of all! */ |
eadb8a09 | 532 | if ((dr6 & DR_STEP) && kmemcheck_trap(regs)) |
f8561296 VN |
533 | return; |
534 | ||
08d68323 P |
535 | /* DR6 may or may not be cleared by the CPU */ |
536 | set_debugreg(0, 6); | |
10faa81e RM |
537 | /* |
538 | * The processor cleared BTF, so don't mark that we need it set. | |
539 | */ | |
540 | clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR); | |
541 | tsk->thread.debugctlmsr = 0; | |
542 | ||
08d68323 P |
543 | /* Store the virtualized DR6 value */ |
544 | tsk->thread.debugreg6 = dr6; | |
545 | ||
62edab90 P |
546 | if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code, |
547 | SIGTRAP) == NOTIFY_STOP) | |
1da177e4 | 548 | return; |
3d2a71a5 | 549 | |
1da177e4 | 550 | /* It's safe to allow irq's after DR6 has been saved */ |
3d2a71a5 | 551 | preempt_conditional_sti(regs); |
1da177e4 | 552 | |
08d68323 P |
553 | if (regs->flags & X86_VM_MASK) { |
554 | handle_vm86_trap((struct kernel_vm86_regs *) regs, | |
555 | error_code, 1); | |
556 | return; | |
1da177e4 LT |
557 | } |
558 | ||
1da177e4 | 559 | /* |
08d68323 P |
560 | * Single-stepping through system calls: ignore any exceptions in |
561 | * kernel space, but re-enable TF when returning to user mode. | |
562 | * | |
563 | * We already checked v86 mode above, so we can check for kernel mode | |
564 | * by just checking the CPL of CS. | |
1da177e4 | 565 | */ |
08d68323 P |
566 | if ((dr6 & DR_STEP) && !user_mode(regs)) { |
567 | tsk->thread.debugreg6 &= ~DR_STEP; | |
568 | set_tsk_thread_flag(tsk, TIF_SINGLESTEP); | |
569 | regs->flags &= ~X86_EFLAGS_TF; | |
1da177e4 | 570 | } |
08d68323 P |
571 | si_code = get_si_code(tsk->thread.debugreg6); |
572 | if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS)) | |
573 | send_sigtrap(tsk, regs, error_code, si_code); | |
3d2a71a5 | 574 | preempt_conditional_cli(regs); |
1da177e4 | 575 | |
1da177e4 LT |
576 | return; |
577 | } | |
578 | ||
081f75bb AH |
579 | #ifdef CONFIG_X86_64 |
580 | static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr) | |
581 | { | |
582 | if (fixup_exception(regs)) | |
583 | return 1; | |
584 | ||
585 | notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE); | |
586 | /* Illegal floating point operation in the kernel */ | |
587 | current->thread.trap_no = trapnr; | |
588 | die(str, regs, 0); | |
589 | return 0; | |
590 | } | |
591 | #endif | |
592 | ||
1da177e4 LT |
593 | /* |
594 | * Note that we play around with the 'TS' bit in an attempt to get | |
595 | * the correct behaviour even in the presence of the asynchronous | |
596 | * IRQ13 behaviour | |
597 | */ | |
65ea5b03 | 598 | void math_error(void __user *ip) |
1da177e4 | 599 | { |
b5964405 | 600 | struct task_struct *task; |
1da177e4 | 601 | siginfo_t info; |
adf77bac | 602 | unsigned short cwd, swd, err; |
1da177e4 LT |
603 | |
604 | /* | |
605 | * Save the info for the exception handler and clear the error. | |
606 | */ | |
607 | task = current; | |
608 | save_init_fpu(task); | |
609 | task->thread.trap_no = 16; | |
610 | task->thread.error_code = 0; | |
611 | info.si_signo = SIGFPE; | |
612 | info.si_errno = 0; | |
65ea5b03 | 613 | info.si_addr = ip; |
1da177e4 LT |
614 | /* |
615 | * (~cwd & swd) will mask out exceptions that are not set to unmasked | |
616 | * status. 0x3f is the exception bits in these regs, 0x200 is the | |
617 | * C1 reg you need in case of a stack fault, 0x040 is the stack | |
618 | * fault bit. We should only be taking one exception at a time, | |
619 | * so if this combination doesn't produce any single exception, | |
a8c1be9d | 620 | * then we have a bad program that isn't synchronizing its FPU usage |
1da177e4 LT |
621 | * and it will suffer the consequences since we won't be able to |
622 | * fully reproduce the context of the exception | |
623 | */ | |
624 | cwd = get_fpu_cwd(task); | |
625 | swd = get_fpu_swd(task); | |
adf77bac | 626 | |
a73ad333 | 627 | err = swd & ~cwd; |
adf77bac PA |
628 | |
629 | if (err & 0x001) { /* Invalid op */ | |
b5964405 IM |
630 | /* |
631 | * swd & 0x240 == 0x040: Stack Underflow | |
632 | * swd & 0x240 == 0x240: Stack Overflow | |
633 | * User must clear the SF bit (0x40) if set | |
634 | */ | |
635 | info.si_code = FPE_FLTINV; | |
adf77bac | 636 | } else if (err & 0x004) { /* Divide by Zero */ |
b5964405 | 637 | info.si_code = FPE_FLTDIV; |
adf77bac | 638 | } else if (err & 0x008) { /* Overflow */ |
b5964405 | 639 | info.si_code = FPE_FLTOVF; |
adf77bac PA |
640 | } else if (err & 0x012) { /* Denormal, Underflow */ |
641 | info.si_code = FPE_FLTUND; | |
642 | } else if (err & 0x020) { /* Precision */ | |
b5964405 | 643 | info.si_code = FPE_FLTRES; |
adf77bac | 644 | } else { |
bd8b96df IM |
645 | /* |
646 | * If we're using IRQ 13, or supposedly even some trap 16 | |
647 | * implementations, it's possible we get a spurious trap... | |
648 | */ | |
a73ad333 | 649 | return; /* Spurious trap, no error */ |
1da177e4 LT |
650 | } |
651 | force_sig_info(SIGFPE, &info, task); | |
652 | } | |
653 | ||
e407d620 | 654 | dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code) |
1da177e4 | 655 | { |
252d28fe | 656 | conditional_sti(regs); |
081f75bb AH |
657 | |
658 | #ifdef CONFIG_X86_32 | |
1da177e4 | 659 | ignore_fpu_irq = 1; |
081f75bb AH |
660 | #else |
661 | if (!user_mode(regs) && | |
662 | kernel_math_error(regs, "kernel x87 math error", 16)) | |
663 | return; | |
664 | #endif | |
665 | ||
65ea5b03 | 666 | math_error((void __user *)regs->ip); |
1da177e4 LT |
667 | } |
668 | ||
65ea5b03 | 669 | static void simd_math_error(void __user *ip) |
1da177e4 | 670 | { |
b5964405 | 671 | struct task_struct *task; |
b5964405 | 672 | siginfo_t info; |
7b4fd4bb | 673 | unsigned short mxcsr; |
1da177e4 LT |
674 | |
675 | /* | |
676 | * Save the info for the exception handler and clear the error. | |
677 | */ | |
678 | task = current; | |
679 | save_init_fpu(task); | |
680 | task->thread.trap_no = 19; | |
681 | task->thread.error_code = 0; | |
682 | info.si_signo = SIGFPE; | |
683 | info.si_errno = 0; | |
684 | info.si_code = __SI_FAULT; | |
65ea5b03 | 685 | info.si_addr = ip; |
1da177e4 LT |
686 | /* |
687 | * The SIMD FPU exceptions are handled a little differently, as there | |
688 | * is only a single status/control register. Thus, to determine which | |
689 | * unmasked exception was caught we must mask the exception mask bits | |
690 | * at 0x1f80, and then use these to mask the exception bits at 0x3f. | |
691 | */ | |
692 | mxcsr = get_fpu_mxcsr(task); | |
693 | switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) { | |
b5964405 IM |
694 | case 0x000: |
695 | default: | |
696 | break; | |
697 | case 0x001: /* Invalid Op */ | |
698 | info.si_code = FPE_FLTINV; | |
699 | break; | |
700 | case 0x002: /* Denormalize */ | |
701 | case 0x010: /* Underflow */ | |
702 | info.si_code = FPE_FLTUND; | |
703 | break; | |
704 | case 0x004: /* Zero Divide */ | |
705 | info.si_code = FPE_FLTDIV; | |
706 | break; | |
707 | case 0x008: /* Overflow */ | |
708 | info.si_code = FPE_FLTOVF; | |
709 | break; | |
710 | case 0x020: /* Precision */ | |
711 | info.si_code = FPE_FLTRES; | |
712 | break; | |
1da177e4 LT |
713 | } |
714 | force_sig_info(SIGFPE, &info, task); | |
715 | } | |
716 | ||
e407d620 AH |
717 | dotraplinkage void |
718 | do_simd_coprocessor_error(struct pt_regs *regs, long error_code) | |
1da177e4 | 719 | { |
b939bde2 AH |
720 | conditional_sti(regs); |
721 | ||
081f75bb | 722 | #ifdef CONFIG_X86_32 |
40d2e763 | 723 | ignore_fpu_irq = 1; |
081f75bb AH |
724 | #else |
725 | if (!user_mode(regs) && | |
726 | kernel_math_error(regs, "kernel simd math error", 19)) | |
727 | return; | |
081f75bb | 728 | #endif |
40d2e763 BG |
729 | |
730 | simd_math_error((void __user *)regs->ip); | |
1da177e4 LT |
731 | } |
732 | ||
e407d620 AH |
733 | dotraplinkage void |
734 | do_spurious_interrupt_bug(struct pt_regs *regs, long error_code) | |
1da177e4 | 735 | { |
cf81978d | 736 | conditional_sti(regs); |
1da177e4 LT |
737 | #if 0 |
738 | /* No need to warn about this any longer. */ | |
b5964405 | 739 | printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n"); |
1da177e4 LT |
740 | #endif |
741 | } | |
742 | ||
081f75bb | 743 | asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void) |
1da177e4 | 744 | { |
1da177e4 | 745 | } |
4efc0670 | 746 | |
7856f6cc | 747 | asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void) |
081f75bb AH |
748 | { |
749 | } | |
750 | ||
e6e9cac8 JF |
751 | /* |
752 | * __math_state_restore assumes that cr0.TS is already clear and the | |
753 | * fpu state is all ready for use. Used during context switch. | |
754 | */ | |
755 | void __math_state_restore(void) | |
081f75bb | 756 | { |
e6e9cac8 JF |
757 | struct thread_info *thread = current_thread_info(); |
758 | struct task_struct *tsk = thread->task; | |
759 | ||
760 | /* | |
761 | * Paranoid restore. send a SIGSEGV if we fail to restore the state. | |
762 | */ | |
763 | if (unlikely(restore_fpu_checking(tsk))) { | |
764 | stts(); | |
765 | force_sig(SIGSEGV, tsk); | |
766 | return; | |
767 | } | |
768 | ||
769 | thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */ | |
770 | tsk->fpu_counter++; | |
081f75bb | 771 | } |
1da177e4 LT |
772 | |
773 | /* | |
b5964405 | 774 | * 'math_state_restore()' saves the current math information in the |
1da177e4 LT |
775 | * old math state array, and gets the new ones from the current task |
776 | * | |
777 | * Careful.. There are problems with IBM-designed IRQ13 behaviour. | |
778 | * Don't touch unless you *really* know how it works. | |
779 | * | |
780 | * Must be called with kernel preemption disabled (in this case, | |
781 | * local interrupts are disabled at the call-site in entry.S). | |
782 | */ | |
acc20761 | 783 | asmlinkage void math_state_restore(void) |
1da177e4 LT |
784 | { |
785 | struct thread_info *thread = current_thread_info(); | |
786 | struct task_struct *tsk = thread->task; | |
787 | ||
aa283f49 SS |
788 | if (!tsk_used_math(tsk)) { |
789 | local_irq_enable(); | |
790 | /* | |
791 | * does a slab alloc which can sleep | |
792 | */ | |
793 | if (init_fpu(tsk)) { | |
794 | /* | |
795 | * ran out of memory! | |
796 | */ | |
797 | do_group_exit(SIGKILL); | |
798 | return; | |
799 | } | |
800 | local_irq_disable(); | |
801 | } | |
802 | ||
b5964405 | 803 | clts(); /* Allow maths ops (or we recurse) */ |
fcb2ac5b | 804 | |
e6e9cac8 | 805 | __math_state_restore(); |
1da177e4 | 806 | } |
5992b6da | 807 | EXPORT_SYMBOL_GPL(math_state_restore); |
1da177e4 LT |
808 | |
809 | #ifndef CONFIG_MATH_EMULATION | |
d315760f | 810 | void math_emulate(struct math_emu_info *info) |
1da177e4 | 811 | { |
b5964405 IM |
812 | printk(KERN_EMERG |
813 | "math-emulation not enabled and no coprocessor found.\n"); | |
814 | printk(KERN_EMERG "killing %s.\n", current->comm); | |
815 | force_sig(SIGFPE, current); | |
1da177e4 LT |
816 | schedule(); |
817 | } | |
1da177e4 LT |
818 | #endif /* CONFIG_MATH_EMULATION */ |
819 | ||
e407d620 | 820 | dotraplinkage void __kprobes |
aa78bcfa | 821 | do_device_not_available(struct pt_regs *regs, long error_code) |
7643e9b9 | 822 | { |
081f75bb | 823 | #ifdef CONFIG_X86_32 |
7643e9b9 | 824 | if (read_cr0() & X86_CR0_EM) { |
d315760f TH |
825 | struct math_emu_info info = { }; |
826 | ||
7643e9b9 | 827 | conditional_sti(regs); |
d315760f | 828 | |
aa78bcfa | 829 | info.regs = regs; |
d315760f | 830 | math_emulate(&info); |
7643e9b9 AH |
831 | } else { |
832 | math_state_restore(); /* interrupts still off */ | |
833 | conditional_sti(regs); | |
834 | } | |
081f75bb AH |
835 | #else |
836 | math_state_restore(); | |
837 | #endif | |
7643e9b9 AH |
838 | } |
839 | ||
081f75bb | 840 | #ifdef CONFIG_X86_32 |
e407d620 | 841 | dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code) |
f8e0870f AH |
842 | { |
843 | siginfo_t info; | |
844 | local_irq_enable(); | |
845 | ||
846 | info.si_signo = SIGILL; | |
847 | info.si_errno = 0; | |
848 | info.si_code = ILL_BADSTK; | |
fc6fcdfb | 849 | info.si_addr = NULL; |
f8e0870f AH |
850 | if (notify_die(DIE_TRAP, "iret exception", |
851 | regs, error_code, 32, SIGILL) == NOTIFY_STOP) | |
852 | return; | |
3c1326f8 | 853 | do_trap(32, SIGILL, "iret exception", regs, error_code, &info); |
f8e0870f | 854 | } |
081f75bb | 855 | #endif |
f8e0870f | 856 | |
1da177e4 LT |
857 | void __init trap_init(void) |
858 | { | |
dbeb2be2 RR |
859 | int i; |
860 | ||
1da177e4 | 861 | #ifdef CONFIG_EISA |
927222b1 | 862 | void __iomem *p = early_ioremap(0x0FFFD9, 4); |
b5964405 IM |
863 | |
864 | if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24)) | |
1da177e4 | 865 | EISA_bus = 1; |
927222b1 | 866 | early_iounmap(p, 4); |
1da177e4 LT |
867 | #endif |
868 | ||
976382dc | 869 | set_intr_gate(0, ÷_error); |
699d2937 AH |
870 | set_intr_gate_ist(1, &debug, DEBUG_STACK); |
871 | set_intr_gate_ist(2, &nmi, NMI_STACK); | |
872 | /* int3 can be called from all */ | |
873 | set_system_intr_gate_ist(3, &int3, DEBUG_STACK); | |
874 | /* int4 can be called from all */ | |
875 | set_system_intr_gate(4, &overflow); | |
64f644c0 | 876 | set_intr_gate(5, &bounds); |
12394cf5 | 877 | set_intr_gate(6, &invalid_op); |
7643e9b9 | 878 | set_intr_gate(7, &device_not_available); |
081f75bb | 879 | #ifdef CONFIG_X86_32 |
a8c1be9d | 880 | set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS); |
081f75bb AH |
881 | #else |
882 | set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK); | |
883 | #endif | |
51bc1ed6 | 884 | set_intr_gate(9, &coprocessor_segment_overrun); |
6bf77bf9 | 885 | set_intr_gate(10, &invalid_TSS); |
36d936c7 | 886 | set_intr_gate(11, &segment_not_present); |
699d2937 | 887 | set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK); |
c6df0d71 | 888 | set_intr_gate(13, &general_protection); |
b5964405 | 889 | set_intr_gate(14, &page_fault); |
cf81978d | 890 | set_intr_gate(15, &spurious_interrupt_bug); |
252d28fe | 891 | set_intr_gate(16, &coprocessor_error); |
5feedfd4 | 892 | set_intr_gate(17, &alignment_check); |
1da177e4 | 893 | #ifdef CONFIG_X86_MCE |
699d2937 | 894 | set_intr_gate_ist(18, &machine_check, MCE_STACK); |
1da177e4 | 895 | #endif |
b939bde2 | 896 | set_intr_gate(19, &simd_coprocessor_error); |
1da177e4 | 897 | |
bb3f0b59 YL |
898 | /* Reserve all the builtin and the syscall vector: */ |
899 | for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++) | |
900 | set_bit(i, used_vectors); | |
901 | ||
081f75bb AH |
902 | #ifdef CONFIG_IA32_EMULATION |
903 | set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall); | |
bb3f0b59 | 904 | set_bit(IA32_SYSCALL_VECTOR, used_vectors); |
081f75bb AH |
905 | #endif |
906 | ||
907 | #ifdef CONFIG_X86_32 | |
d43c6e80 | 908 | if (cpu_has_fxsr) { |
d43c6e80 JB |
909 | printk(KERN_INFO "Enabling fast FPU save and restore... "); |
910 | set_in_cr4(X86_CR4_OSFXSR); | |
911 | printk("done.\n"); | |
912 | } | |
913 | if (cpu_has_xmm) { | |
b5964405 IM |
914 | printk(KERN_INFO |
915 | "Enabling unmasked SIMD FPU exception support... "); | |
d43c6e80 JB |
916 | set_in_cr4(X86_CR4_OSXMMEXCPT); |
917 | printk("done.\n"); | |
918 | } | |
919 | ||
699d2937 | 920 | set_system_trap_gate(SYSCALL_VECTOR, &system_call); |
dbeb2be2 | 921 | set_bit(SYSCALL_VECTOR, used_vectors); |
081f75bb | 922 | #endif |
bb3f0b59 | 923 | |
1da177e4 | 924 | /* |
b5964405 | 925 | * Should be a barrier for any external CPU state: |
1da177e4 LT |
926 | */ |
927 | cpu_init(); | |
928 | ||
428cf902 | 929 | x86_init.irqs.trap_init(); |
1da177e4 | 930 | } |