projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] FPU context corrupted after resume
[linux-2.6-block.git] / arch / i386 / kernel / traps.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/arch/i386/traps.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * Pentium III FXSR, SSE support
7 * Gareth Hughes <gareth@valinux.com>, May 2000
8 */
9
10/*
11 * 'Traps.c' handles hardware traps and faults after we have saved some
12 * state in 'asm.s'.
13 */
14#include <linux/config.h>
15#include <linux/sched.h>
16#include <linux/kernel.h>
17#include <linux/string.h>
18#include <linux/errno.h>
19#include <linux/timer.h>
20#include <linux/mm.h>
21#include <linux/init.h>
22#include <linux/delay.h>
23#include <linux/spinlock.h>
24#include <linux/interrupt.h>
25#include <linux/highmem.h>
26#include <linux/kallsyms.h>
27#include <linux/ptrace.h>
28#include <linux/utsname.h>
29#include <linux/kprobes.h>
6e274d14 30#include <linux/kexec.h>
1da177e4
LT		 31
32#ifdef CONFIG_EISA
33#include <linux/ioport.h>
34#include <linux/eisa.h>
35#endif
36
37#ifdef CONFIG_MCA
38#include <linux/mca.h>
39#endif
40
41#include <asm/processor.h>
42#include <asm/system.h>
43#include <asm/uaccess.h>
44#include <asm/io.h>
45#include <asm/atomic.h>
46#include <asm/debugreg.h>
47#include <asm/desc.h>
48#include <asm/i387.h>
49#include <asm/nmi.h>
50
51#include <asm/smp.h>
52#include <asm/arch_hooks.h>
53#include <asm/kdebug.h>
54
1da177e4
LT		 55#include <linux/module.h>
56
57#include "mach_traps.h"
58
59asmlinkage int system_call(void);
60
61struct desc_struct default_ldt[] = { { 0, 0 }, { 0, 0 }, { 0, 0 },
62 { 0, 0 }, { 0, 0 } };
63
64/* Do we ignore FPU interrupts ? */
65char ignore_fpu_irq = 0;
66
67/*
68 * The IDT has to be page-aligned to simplify the Pentium
69 * F0 0F bug workaround.. We have a special link segment
70 * for this.
71 */
72struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
73
74asmlinkage void divide_error(void);
75asmlinkage void debug(void);
76asmlinkage void nmi(void);
77asmlinkage void int3(void);
78asmlinkage void overflow(void);
79asmlinkage void bounds(void);
80asmlinkage void invalid_op(void);
81asmlinkage void device_not_available(void);
82asmlinkage void coprocessor_segment_overrun(void);
83asmlinkage void invalid_TSS(void);
84asmlinkage void segment_not_present(void);
85asmlinkage void stack_segment(void);
86asmlinkage void general_protection(void);
87asmlinkage void page_fault(void);
88asmlinkage void coprocessor_error(void);
89asmlinkage void simd_coprocessor_error(void);
90asmlinkage void alignment_check(void);
91asmlinkage void spurious_interrupt_bug(void);
92asmlinkage void machine_check(void);
93
94static int kstack_depth_to_print = 24;
95struct notifier_block *i386die_chain;
96static DEFINE_SPINLOCK(die_notifier_lock);
97
98int register_die_notifier(struct notifier_block *nb)
99{
100 int err = 0;
101 unsigned long flags;
102 spin_lock_irqsave(&die_notifier_lock, flags);
103 err = notifier_chain_register(&i386die_chain, nb);
104 spin_unlock_irqrestore(&die_notifier_lock, flags);
105 return err;
106}
129f6946 107EXPORT_SYMBOL(register_die_notifier);
1da177e4
LT		 108
109static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
110{
111 return p > (void *)tinfo &&
112 p < (void *)tinfo + THREAD_SIZE - 3;
113}
114
115static inline unsigned long print_context_stack(struct thread_info *tinfo,
116 unsigned long *stack, unsigned long ebp)
117{
118 unsigned long addr;
119
120#ifdef CONFIG_FRAME_POINTER
121 while (valid_stack_ptr(tinfo, (void *)ebp)) {
122 addr = *(unsigned long *)(ebp + 4);
123 printk(" [<%08lx>] ", addr);
124 print_symbol("%s", addr);
125 printk("\n");
126 ebp = *(unsigned long *)ebp;
127 }
128#else
129 while (valid_stack_ptr(tinfo, stack)) {
130 addr = *stack++;
131 if (__kernel_text_address(addr)) {
132 printk(" [<%08lx>]", addr);
133 print_symbol(" %s", addr);
134 printk("\n");
135 }
136 }
137#endif
138 return ebp;
139}
140
141void show_trace(struct task_struct *task, unsigned long * stack)
142{
143 unsigned long ebp;
144
145 if (!task)
146 task = current;
147
148 if (task == current) {
149 /* Grab ebp right from our regs */
150 asm ("movl %%ebp, %0" : "=r" (ebp) : );
151 } else {
152 /* ebp is the last reg pushed by switch_to */
153 ebp = *(unsigned long *) task->thread.esp;
154 }
155
156 while (1) {
157 struct thread_info *context;
158 context = (struct thread_info *)
159 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
160 ebp = print_context_stack(context, stack, ebp);
161 stack = (unsigned long*)context->previous_esp;
162 if (!stack)
163 break;
164 printk(" =======================\n");
165 }
166}
167
168void show_stack(struct task_struct *task, unsigned long *esp)
169{
170 unsigned long *stack;
171 int i;
172
173 if (esp == NULL) {
174 if (task)
175 esp = (unsigned long*)task->thread.esp;
176 else
177 esp = (unsigned long *)&esp;
178 }
179
180 stack = esp;
181 for(i = 0; i < kstack_depth_to_print; i++) {
182 if (kstack_end(stack))
183 break;
184 if (i && ((i % 8) == 0))
185 printk("\n ");
186 printk("%08lx ", *stack++);
187 }
188 printk("\nCall Trace:\n");
189 show_trace(task, esp);
190}
191
192/*
193 * The architecture-independent dump_stack generator
194 */
195void dump_stack(void)
196{
197 unsigned long stack;
198
199 show_trace(current, &stack);
200}
201
202EXPORT_SYMBOL(dump_stack);
203
204void show_registers(struct pt_regs *regs)
205{
206 int i;
207 int in_kernel = 1;
208 unsigned long esp;
209 unsigned short ss;
210
211 esp = (unsigned long) (&regs->esp);
0998e422 212 savesegment(ss, ss);
717b594a 213 if (user_mode(regs)) {
1da177e4
LT		 214 in_kernel = 0;
215 esp = regs->esp;
216 ss = regs->xss & 0xffff;
217 }
218 print_modules();
219 printk("CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\nEFLAGS: %08lx"
220 " (%s) \n",
221 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
222 print_tainted(), regs->eflags, system_utsname.release);
223 print_symbol("EIP is at %s\n", regs->eip);
224 printk("eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
225 regs->eax, regs->ebx, regs->ecx, regs->edx);
226 printk("esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
227 regs->esi, regs->edi, regs->ebp, esp);
228 printk("ds: %04x es: %04x ss: %04x\n",
229 regs->xds & 0xffff, regs->xes & 0xffff, ss);
230 printk("Process %s (pid: %d, threadinfo=%p task=%p)",
231 current->comm, current->pid, current_thread_info(), current);
232 /*
233 * When in-kernel, we also print out the stack and code at the
234 * time of the fault..
235 */
236 if (in_kernel) {
3f3ae347 237 u8 __user *eip;
1da177e4
LT		 238
239 printk("\nStack: ");
240 show_stack(NULL, (unsigned long*)esp);
241
242 printk("Code: ");
243
3f3ae347 244 eip = (u8 __user *)regs->eip - 43;
1da177e4
LT		 245 for (i = 0; i < 64; i++, eip++) {
246 unsigned char c;
247
3f3ae347 248 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
1da177e4
LT		 249 printk(" Bad EIP value.");
250 break;
251 }
3f3ae347 252 if (eip == (u8 __user *)regs->eip)
1da177e4
LT		 253 printk("<%02x> ", c);
254 else
255 printk("%02x ", c);
256 }
257 }
258 printk("\n");
259}
260
261static void handle_BUG(struct pt_regs *regs)
262{
263 unsigned short ud2;
264 unsigned short line;
265 char *file;
266 char c;
267 unsigned long eip;
268
1da177e4
LT		 269 eip = regs->eip;
270
271 if (eip < PAGE_OFFSET)
272 goto no_bug;
3f3ae347 273 if (__get_user(ud2, (unsigned short __user *)eip))
1da177e4
LT		 274 goto no_bug;
275 if (ud2 != 0x0b0f)
276 goto no_bug;
3f3ae347 277 if (__get_user(line, (unsigned short __user *)(eip + 2)))
1da177e4 278 goto bug;
3f3ae347 279 if (__get_user(file, (char * __user *)(eip + 4)) ||
1da177e4
LT		 280 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
281 file = "<bad filename>";
282
283 printk("------------[ cut here ]------------\n");
284 printk(KERN_ALERT "kernel BUG at %s:%d!\n", file, line);
285
286no_bug:
287 return;
288
289 /* Here we know it was a BUG but file-n-line is unavailable */
290bug:
291 printk("Kernel BUG\n");
292}
293
6e274d14
AN		 294/* This is gone through when something in the kernel
295 * has done something bad and is about to be terminated.
296*/
1da177e4
LT		 297void die(const char * str, struct pt_regs * regs, long err)
298{
299 static struct {
300 spinlock_t lock;
301 u32 lock_owner;
302 int lock_owner_depth;
303 } die = {
304 .lock = SPIN_LOCK_UNLOCKED,
305 .lock_owner = -1,
306 .lock_owner_depth = 0
307 };
308 static int die_counter;
309
39c715b7 310 if (die.lock_owner != raw_smp_processor_id()) {
1da177e4
LT		 311 console_verbose();
312 spin_lock_irq(&die.lock);
313 die.lock_owner = smp_processor_id();
314 die.lock_owner_depth = 0;
315 bust_spinlocks(1);
316 }
317
318 if (++die.lock_owner_depth < 3) {
319 int nl = 0;
320 handle_BUG(regs);
321 printk(KERN_ALERT "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
322#ifdef CONFIG_PREEMPT
323 printk("PREEMPT ");
324 nl = 1;
325#endif
326#ifdef CONFIG_SMP
327 printk("SMP ");
328 nl = 1;
329#endif
330#ifdef CONFIG_DEBUG_PAGEALLOC
331 printk("DEBUG_PAGEALLOC");
332 nl = 1;
333#endif
334 if (nl)
335 printk("\n");
336 notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
337 show_registers(regs);
338 } else
339 printk(KERN_ERR "Recursive die() failure, output suppressed\n");
340
341 bust_spinlocks(0);
342 die.lock_owner = -1;
343 spin_unlock_irq(&die.lock);
6e274d14
AN		 344
345 if (kexec_should_crash(current))
346 crash_kexec(regs);
347
1da177e4
LT		 348 if (in_interrupt())
349 panic("Fatal exception in interrupt");
350
351 if (panic_on_oops) {
352 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
353 ssleep(5);
354 panic("Fatal exception");
355 }
356 do_exit(SIGSEGV);
357}
358
359static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
360{
717b594a 361 if (!user_mode_vm(regs))
1da177e4
LT		 362 die(str, regs, err);
363}
364
3d97ae5b
PP		 365static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
366 struct pt_regs * regs, long error_code,
367 siginfo_t *info)
1da177e4 368{
4f339ecb
AN		 369 struct task_struct *tsk = current;
370 tsk->thread.error_code = error_code;
371 tsk->thread.trap_no = trapnr;
372
1da177e4
LT		 373 if (regs->eflags & VM_MASK) {
374 if (vm86)
375 goto vm86_trap;
376 goto trap_signal;
377 }
378
717b594a 379 if (!user_mode(regs))
1da177e4
LT		 380 goto kernel_trap;
381
382 trap_signal: {
1da177e4
LT		 383 if (info)
384 force_sig_info(signr, info, tsk);
385 else
386 force_sig(signr, tsk);
387 return;
388 }
389
390 kernel_trap: {
391 if (!fixup_exception(regs))
392 die(str, regs, error_code);
393 return;
394 }
395
396 vm86_trap: {
397 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
398 if (ret) goto trap_signal;
399 return;
400 }
401}
402
403#define DO_ERROR(trapnr, signr, str, name) \
404fastcall void do_##name(struct pt_regs * regs, long error_code) \
405{ \
406 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
407 == NOTIFY_STOP) \
408 return; \
409 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
410}
411
412#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
413fastcall void do_##name(struct pt_regs * regs, long error_code) \
414{ \
415 siginfo_t info; \
416 info.si_signo = signr; \
417 info.si_errno = 0; \
418 info.si_code = sicode; \
419 info.si_addr = (void __user *)siaddr; \
420 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
421 == NOTIFY_STOP) \
422 return; \
423 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
424}
425
426#define DO_VM86_ERROR(trapnr, signr, str, name) \
427fastcall void do_##name(struct pt_regs * regs, long error_code) \
428{ \
429 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
430 == NOTIFY_STOP) \
431 return; \
432 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
433}
434
435#define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
436fastcall void do_##name(struct pt_regs * regs, long error_code) \
437{ \
438 siginfo_t info; \
439 info.si_signo = signr; \
440 info.si_errno = 0; \
441 info.si_code = sicode; \
442 info.si_addr = (void __user *)siaddr; \
443 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
444 == NOTIFY_STOP) \
445 return; \
446 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
447}
448
449DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
450#ifndef CONFIG_KPROBES
451DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
452#endif
453DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
454DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
455DO_ERROR_INFO( 6, SIGILL, "invalid operand", invalid_op, ILL_ILLOPN, regs->eip)
456DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
457DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
458DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
459DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
460DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
a879cbbb 461DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
1da177e4 462
3d97ae5b
PP		 463fastcall void __kprobes do_general_protection(struct pt_regs * regs,
464 long error_code)
1da177e4
LT		 465{
466 int cpu = get_cpu();
467 struct tss_struct *tss = &per_cpu(init_tss, cpu);
468 struct thread_struct *thread = &current->thread;
469
470 /*
471 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
472 * invalid offset set (the LAZY one) and the faulting thread has
473 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
474 * and we set the offset field correctly. Then we let the CPU to
475 * restart the faulting instruction.
476 */
477 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
478 thread->io_bitmap_ptr) {
479 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
480 thread->io_bitmap_max);
481 /*
482 * If the previously set map was extending to higher ports
483 * than the current one, pad extra space with 0xff (no access).
484 */
485 if (thread->io_bitmap_max < tss->io_bitmap_max)
486 memset((char *) tss->io_bitmap +
487 thread->io_bitmap_max, 0xff,
488 tss->io_bitmap_max - thread->io_bitmap_max);
489 tss->io_bitmap_max = thread->io_bitmap_max;
490 tss->io_bitmap_base = IO_BITMAP_OFFSET;
491 put_cpu();
492 return;
493 }
494 put_cpu();
495
4f339ecb
AN		 496 current->thread.error_code = error_code;
497 current->thread.trap_no = 13;
498
1da177e4
LT		 499 if (regs->eflags & VM_MASK)
500 goto gp_in_vm86;
501
717b594a 502 if (!user_mode(regs))
1da177e4
LT		 503 goto gp_in_kernel;
504
505 current->thread.error_code = error_code;
506 current->thread.trap_no = 13;
507 force_sig(SIGSEGV, current);
508 return;
509
510gp_in_vm86:
511 local_irq_enable();
512 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
513 return;
514
515gp_in_kernel:
516 if (!fixup_exception(regs)) {
517 if (notify_die(DIE_GPF, "general protection fault", regs,
518 error_code, 13, SIGSEGV) == NOTIFY_STOP)
519 return;
520 die("general protection fault", regs, error_code);
521 }
522}
523
524static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
525{
526 printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
527 printk("You probably have a hardware problem with your RAM chips\n");
528
529 /* Clear and disable the memory parity error line. */
530 clear_mem_error(reason);
531}
532
533static void io_check_error(unsigned char reason, struct pt_regs * regs)
534{
535 unsigned long i;
536
537 printk("NMI: IOCK error (debug interrupt?)\n");
538 show_registers(regs);
539
540 /* Re-enable the IOCK line, wait for a few seconds */
541 reason = (reason & 0xf) | 8;
542 outb(reason, 0x61);
543 i = 2000;
544 while (--i) udelay(1000);
545 reason &= ~8;
546 outb(reason, 0x61);
547}
548
549static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
550{
551#ifdef CONFIG_MCA
552 /* Might actually be able to figure out what the guilty party
553 * is. */
554 if( MCA_bus ) {
555 mca_handle_nmi();
556 return;
557 }
558#endif
559 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
560 reason, smp_processor_id());
561 printk("Dazed and confused, but trying to continue\n");
562 printk("Do you have a strange power saving mode enabled?\n");
563}
564
565static DEFINE_SPINLOCK(nmi_print_lock);
566
567void die_nmi (struct pt_regs *regs, const char *msg)
568{
748f2edb
GA		 569 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 0, SIGINT) ==
570 NOTIFY_STOP)
571 return;
572
1da177e4
LT		 573 spin_lock(&nmi_print_lock);
574 /*
575 * We are in trouble anyway, lets at least try
576 * to get a message out.
577 */
578 bust_spinlocks(1);
579 printk(msg);
580 printk(" on CPU%d, eip %08lx, registers:\n",
581 smp_processor_id(), regs->eip);
582 show_registers(regs);
583 printk("console shuts up ...\n");
584 console_silent();
585 spin_unlock(&nmi_print_lock);
586 bust_spinlocks(0);
6e274d14
AN		 587
588 /* If we are in kernel we are probably nested up pretty bad
589 * and might aswell get out now while we still can.
590 */
591 if (!user_mode(regs)) {
592 current->thread.trap_no = 2;
593 crash_kexec(regs);
594 }
595
1da177e4
LT		 596 do_exit(SIGSEGV);
597}
598
599static void default_do_nmi(struct pt_regs * regs)
600{
601 unsigned char reason = 0;
602
603 /* Only the BSP gets external NMIs from the system. */
604 if (!smp_processor_id())
605 reason = get_nmi_reason();
606
607 if (!(reason & 0xc0)) {
608 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
609 == NOTIFY_STOP)
610 return;
611#ifdef CONFIG_X86_LOCAL_APIC
612 /*
613 * Ok, so this is none of the documented NMI sources,
614 * so it must be the NMI watchdog.
615 */
616 if (nmi_watchdog) {
617 nmi_watchdog_tick(regs);
618 return;
619 }
620#endif
621 unknown_nmi_error(reason, regs);
622 return;
623 }
624 if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
625 return;
626 if (reason & 0x80)
627 mem_parity_error(reason, regs);
628 if (reason & 0x40)
629 io_check_error(reason, regs);
630 /*
631 * Reassert NMI in case it became active meanwhile
632 * as it's edge-triggered.
633 */
634 reassert_nmi();
635}
636
637static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
638{
639 return 0;
640}
641
642static nmi_callback_t nmi_callback = dummy_nmi_callback;
643
644fastcall void do_nmi(struct pt_regs * regs, long error_code)
645{
646 int cpu;
647
648 nmi_enter();
649
650 cpu = smp_processor_id();
f3705136
ZM		 651
652#ifdef CONFIG_HOTPLUG_CPU
653 if (!cpu_online(cpu)) {
654 nmi_exit();
655 return;
656 }
657#endif
658
1da177e4
LT		 659 ++nmi_count(cpu);
660
19306059 661 if (!rcu_dereference(nmi_callback)(regs, cpu))
1da177e4
LT		 662 default_do_nmi(regs);
663
664 nmi_exit();
665}
666
667void set_nmi_callback(nmi_callback_t callback)
668{
19306059 669 rcu_assign_pointer(nmi_callback, callback);
1da177e4 670}
129f6946 671EXPORT_SYMBOL_GPL(set_nmi_callback);
1da177e4
LT		 672
673void unset_nmi_callback(void)
674{
675 nmi_callback = dummy_nmi_callback;
676}
129f6946 677EXPORT_SYMBOL_GPL(unset_nmi_callback);
1da177e4
LT		 678
679#ifdef CONFIG_KPROBES
3d97ae5b 680fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
1da177e4
LT		 681{
682 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
683 == NOTIFY_STOP)
48c88211 684 return;
1da177e4
LT		 685 /* This is an interrupt gate, because kprobes wants interrupts
686 disabled. Normal trap handlers don't. */
687 restore_interrupts(regs);
688 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
1da177e4
LT		 689}
690#endif
691
692/*
693 * Our handling of the processor debug registers is non-trivial.
694 * We do not clear them on entry and exit from the kernel. Therefore
695 * it is possible to get a watchpoint trap here from inside the kernel.
696 * However, the code in ./ptrace.c has ensured that the user can
697 * only set watchpoints on userspace addresses. Therefore the in-kernel
698 * watchpoint trap can only occur in code which is reading/writing
699 * from user space. Such code must not hold kernel locks (since it
700 * can equally take a page fault), therefore it is safe to call
701 * force_sig_info even though that claims and releases locks.
702 *
703 * Code in ./signal.c ensures that the debug control register
704 * is restored before we deliver any signal, and therefore that
705 * user code runs with the correct debug control register even though
706 * we clear it here.
707 *
708 * Being careful here means that we don't have to be as careful in a
709 * lot of more complicated places (task switching can be a bit lazy
710 * about restoring all the debug state, and ptrace doesn't have to
711 * find every occurrence of the TF bit that could be saved away even
712 * by user code)
713 */
3d97ae5b 714fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
1da177e4
LT		 715{
716 unsigned int condition;
717 struct task_struct *tsk = current;
718
1cc6f12e 719 get_debugreg(condition, 6);
1da177e4
LT		 720
721 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
722 SIGTRAP) == NOTIFY_STOP)
723 return;
724 /* It's safe to allow irq's after DR6 has been saved */
725 if (regs->eflags & X86_EFLAGS_IF)
726 local_irq_enable();
727
728 /* Mask out spurious debug traps due to lazy DR7 setting */
729 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
730 if (!tsk->thread.debugreg[7])
731 goto clear_dr7;
732 }
733
734 if (regs->eflags & VM_MASK)
735 goto debug_vm86;
736
737 /* Save debug status register where ptrace can see it */
738 tsk->thread.debugreg[6] = condition;
739
740 /*
741 * Single-stepping through TF: make sure we ignore any events in
742 * kernel space (but re-enable TF when returning to user mode).
743 */
744 if (condition & DR_STEP) {
745 /*
746 * We already checked v86 mode above, so we can
747 * check for kernel mode by just checking the CPL
748 * of CS.
749 */
717b594a 750 if (!user_mode(regs))
1da177e4
LT		 751 goto clear_TF_reenable;
752 }
753
754 /* Ok, finally something we can handle */
755 send_sigtrap(tsk, regs, error_code);
756
757 /* Disable additional traps. They'll be re-enabled when
758 * the signal is delivered.
759 */
760clear_dr7:
1cc6f12e 761 set_debugreg(0, 7);
1da177e4
LT		 762 return;
763
764debug_vm86:
765 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
766 return;
767
768clear_TF_reenable:
769 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
770 regs->eflags &= ~TF_MASK;
771 return;
772}
773
774/*
775 * Note that we play around with the 'TS' bit in an attempt to get
776 * the correct behaviour even in the presence of the asynchronous
777 * IRQ13 behaviour
778 */
779void math_error(void __user *eip)
780{
781 struct task_struct * task;
782 siginfo_t info;
783 unsigned short cwd, swd;
784
785 /*
786 * Save the info for the exception handler and clear the error.
787 */
788 task = current;
789 save_init_fpu(task);
790 task->thread.trap_no = 16;
791 task->thread.error_code = 0;
792 info.si_signo = SIGFPE;
793 info.si_errno = 0;
794 info.si_code = __SI_FAULT;
795 info.si_addr = eip;
796 /*
797 * (~cwd & swd) will mask out exceptions that are not set to unmasked
798 * status. 0x3f is the exception bits in these regs, 0x200 is the
799 * C1 reg you need in case of a stack fault, 0x040 is the stack
800 * fault bit. We should only be taking one exception at a time,
801 * so if this combination doesn't produce any single exception,
802 * then we have a bad program that isn't syncronizing its FPU usage
803 * and it will suffer the consequences since we won't be able to
804 * fully reproduce the context of the exception
805 */
806 cwd = get_fpu_cwd(task);
807 swd = get_fpu_swd(task);
b1daec30 808 switch (swd & ~cwd & 0x3f) {
33333373
CE		 809 case 0x000: /* No unmasked exception */
810 return;
811 default: /* Multiple exceptions */
1da177e4
LT		 812 break;
813 case 0x001: /* Invalid Op */
b1daec30
CE		 814 /*
815 * swd & 0x240 == 0x040: Stack Underflow
816 * swd & 0x240 == 0x240: Stack Overflow
817 * User must clear the SF bit (0x40) if set
818 */
1da177e4 819 info.si_code = FPE_FLTINV;
1da177e4
LT		 820 break;
821 case 0x002: /* Denormalize */
822 case 0x010: /* Underflow */
823 info.si_code = FPE_FLTUND;
824 break;
825 case 0x004: /* Zero Divide */
826 info.si_code = FPE_FLTDIV;
827 break;
828 case 0x008: /* Overflow */
829 info.si_code = FPE_FLTOVF;
830 break;
831 case 0x020: /* Precision */
832 info.si_code = FPE_FLTRES;
833 break;
834 }
835 force_sig_info(SIGFPE, &info, task);
836}
837
838fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
839{
840 ignore_fpu_irq = 1;
841 math_error((void __user *)regs->eip);
842}
843
844static void simd_math_error(void __user *eip)
845{
846 struct task_struct * task;
847 siginfo_t info;
848 unsigned short mxcsr;
849
850 /*
851 * Save the info for the exception handler and clear the error.
852 */
853 task = current;
854 save_init_fpu(task);
855 task->thread.trap_no = 19;
856 task->thread.error_code = 0;
857 info.si_signo = SIGFPE;
858 info.si_errno = 0;
859 info.si_code = __SI_FAULT;
860 info.si_addr = eip;
861 /*
862 * The SIMD FPU exceptions are handled a little differently, as there
863 * is only a single status/control register. Thus, to determine which
864 * unmasked exception was caught we must mask the exception mask bits
865 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
866 */
867 mxcsr = get_fpu_mxcsr(task);
868 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
869 case 0x000:
870 default:
871 break;
872 case 0x001: /* Invalid Op */
873 info.si_code = FPE_FLTINV;
874 break;
875 case 0x002: /* Denormalize */
876 case 0x010: /* Underflow */
877 info.si_code = FPE_FLTUND;
878 break;
879 case 0x004: /* Zero Divide */
880 info.si_code = FPE_FLTDIV;
881 break;
882 case 0x008: /* Overflow */
883 info.si_code = FPE_FLTOVF;
884 break;
885 case 0x020: /* Precision */
886 info.si_code = FPE_FLTRES;
887 break;
888 }
889 force_sig_info(SIGFPE, &info, task);
890}
891
892fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
893 long error_code)
894{
895 if (cpu_has_xmm) {
896 /* Handle SIMD FPU exceptions on PIII+ processors. */
897 ignore_fpu_irq = 1;
898 simd_math_error((void __user *)regs->eip);
899 } else {
900 /*
901 * Handle strange cache flush from user space exception
902 * in all other cases. This is undocumented behaviour.
903 */
904 if (regs->eflags & VM_MASK) {
905 handle_vm86_fault((struct kernel_vm86_regs *)regs,
906 error_code);
907 return;
908 }
1da177e4
LT		 909 current->thread.trap_no = 19;
910 current->thread.error_code = error_code;
4f339ecb 911 die_if_kernel("cache flush denied", regs, error_code);
1da177e4
LT		 912 force_sig(SIGSEGV, current);
913 }
914}
915
916fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
917 long error_code)
918{
919#if 0
920 /* No need to warn about this any longer. */
921 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
922#endif
923}
924
925fastcall void setup_x86_bogus_stack(unsigned char * stk)
926{
927 unsigned long *switch16_ptr, *switch32_ptr;
928 struct pt_regs *regs;
929 unsigned long stack_top, stack_bot;
930 unsigned short iret_frame16_off;
931 int cpu = smp_processor_id();
932 /* reserve the space on 32bit stack for the magic switch16 pointer */
933 memmove(stk, stk + 8, sizeof(struct pt_regs));
934 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
935 regs = (struct pt_regs *)stk;
936 /* now the switch32 on 16bit stack */
937 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
938 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
939 switch32_ptr = (unsigned long *)(stack_top - 8);
940 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
941 /* copy iret frame on 16bit stack */
942 memcpy((void *)(stack_bot + iret_frame16_off), &regs->eip, 20);
943 /* fill in the switch pointers */
944 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
945 switch16_ptr[1] = __ESPFIX_SS;
946 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
947 8 - CPU_16BIT_STACK_SIZE;
948 switch32_ptr[1] = __KERNEL_DS;
949}
950
951fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
952{
953 unsigned long *switch32_ptr;
954 unsigned char *stack16, *stack32;
955 unsigned long stack_top, stack_bot;
956 int len;
957 int cpu = smp_processor_id();
958 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
959 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
960 switch32_ptr = (unsigned long *)(stack_top - 8);
961 /* copy the data from 16bit stack to 32bit stack */
962 len = CPU_16BIT_STACK_SIZE - 8 - sp;
963 stack16 = (unsigned char *)(stack_bot + sp);
964 stack32 = (unsigned char *)
965 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
966 memcpy(stack32, stack16, len);
967 return stack32;
968}
969
970/*
971 * 'math_state_restore()' saves the current math information in the
972 * old math state array, and gets the new ones from the current task
973 *
974 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
975 * Don't touch unless you *really* know how it works.
976 *
977 * Must be called with kernel preemption disabled (in this case,
978 * local interrupts are disabled at the call-site in entry.S).
979 */
980asmlinkage void math_state_restore(struct pt_regs regs)
981{
982 struct thread_info *thread = current_thread_info();
983 struct task_struct *tsk = thread->task;
984
985 clts(); /* Allow maths ops (or we recurse) */
986 if (!tsk_used_math(tsk))
987 init_fpu(tsk);
988 restore_fpu(tsk);
989 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
990}
991
992#ifndef CONFIG_MATH_EMULATION
993
994asmlinkage void math_emulate(long arg)
995{
996 printk("math-emulation not enabled and no coprocessor found.\n");
997 printk("killing %s.\n",current->comm);
998 force_sig(SIGFPE,current);
999 schedule();
1000}
1001
1002#endif /* CONFIG_MATH_EMULATION */
1003
1004#ifdef CONFIG_X86_F00F_BUG
1005void __init trap_init_f00f_bug(void)
1006{
1007 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1008
1009 /*
1010 * Update the IDT descriptor and reload the IDT so that
1011 * it uses the read-only mapped virtual address.
1012 */
1013 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
4d37e7e3 1014 load_idt(&idt_descr);
1da177e4
LT		 1015}
1016#endif
1017
1018#define _set_gate(gate_addr,type,dpl,addr,seg) \
1019do { \
1020 int __d0, __d1; \
1021 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1022 "movw %4,%%dx\n\t" \
1023 "movl %%eax,%0\n\t" \
1024 "movl %%edx,%1" \
1025 :"=m" (*((long *) (gate_addr))), \
1026 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1027 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1028 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1029} while (0)
1030
1031
1032/*
1033 * This needs to use 'idt_table' rather than 'idt', and
1034 * thus use the _nonmapped_ version of the IDT, as the
1035 * Pentium F0 0F bugfix can have resulted in the mapped
1036 * IDT being write-protected.
1037 */
1038void set_intr_gate(unsigned int n, void *addr)
1039{
1040 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1041}
1042
1043/*
1044 * This routine sets up an interrupt gate at directory privilege level 3.
1045 */
1046static inline void set_system_intr_gate(unsigned int n, void *addr)
1047{
1048 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1049}
1050
1051static void __init set_trap_gate(unsigned int n, void *addr)
1052{
1053 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1054}
1055
1056static void __init set_system_gate(unsigned int n, void *addr)
1057{
1058 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1059}
1060
1061static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1062{
1063 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1064}
1065
1066
1067void __init trap_init(void)
1068{
1069#ifdef CONFIG_EISA
1070 void __iomem *p = ioremap(0x0FFFD9, 4);
1071 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1072 EISA_bus = 1;
1073 }
1074 iounmap(p);
1075#endif
1076
1077#ifdef CONFIG_X86_LOCAL_APIC
1078 init_apic_mappings();
1079#endif
1080
1081 set_trap_gate(0,&divide_error);
1082 set_intr_gate(1,&debug);
1083 set_intr_gate(2,&nmi);
1084 set_system_intr_gate(3, &int3); /* int3-5 can be called from all */
1085 set_system_gate(4,&overflow);
1086 set_system_gate(5,&bounds);
1087 set_trap_gate(6,&invalid_op);
1088 set_trap_gate(7,&device_not_available);
1089 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1090 set_trap_gate(9,&coprocessor_segment_overrun);
1091 set_trap_gate(10,&invalid_TSS);
1092 set_trap_gate(11,&segment_not_present);
1093 set_trap_gate(12,&stack_segment);
1094 set_trap_gate(13,&general_protection);
1095 set_intr_gate(14,&page_fault);
1096 set_trap_gate(15,&spurious_interrupt_bug);
1097 set_trap_gate(16,&coprocessor_error);
1098 set_trap_gate(17,&alignment_check);
1099#ifdef CONFIG_X86_MCE
1100 set_trap_gate(18,&machine_check);
1101#endif
1102 set_trap_gate(19,&simd_coprocessor_error);
1103
1104 set_system_gate(SYSCALL_VECTOR,&system_call);
1105
1106 /*
1107 * Should be a barrier for any external CPU state.
1108 */
1109 cpu_init();
1110
1111 trap_init_hook();
1112}
1113
1114static int __init kstack_setup(char *s)
1115{
1116 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1117 return 0;
1118}
1119__setup("kstack=", kstack_setup);
Linux 6.x block layer and io_uring tree(s)