Loading...
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 *
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 */
8
9/*
10 * Handle hardware traps and faults.
11 */
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/context_tracking.h>
16#include <linux/interrupt.h>
17#include <linux/kallsyms.h>
18#include <linux/spinlock.h>
19#include <linux/kprobes.h>
20#include <linux/uaccess.h>
21#include <linux/kdebug.h>
22#include <linux/kgdb.h>
23#include <linux/kernel.h>
24#include <linux/export.h>
25#include <linux/ptrace.h>
26#include <linux/uprobes.h>
27#include <linux/string.h>
28#include <linux/delay.h>
29#include <linux/errno.h>
30#include <linux/kexec.h>
31#include <linux/sched.h>
32#include <linux/sched/task_stack.h>
33#include <linux/timer.h>
34#include <linux/init.h>
35#include <linux/bug.h>
36#include <linux/nmi.h>
37#include <linux/mm.h>
38#include <linux/smp.h>
39#include <linux/io.h>
40
41#if defined(CONFIG_EDAC)
42#include <linux/edac.h>
43#endif
44
45#include <asm/stacktrace.h>
46#include <asm/processor.h>
47#include <asm/debugreg.h>
48#include <linux/atomic.h>
49#include <asm/text-patching.h>
50#include <asm/ftrace.h>
51#include <asm/traps.h>
52#include <asm/desc.h>
53#include <asm/fpu/internal.h>
54#include <asm/cpu_entry_area.h>
55#include <asm/mce.h>
56#include <asm/fixmap.h>
57#include <asm/mach_traps.h>
58#include <asm/alternative.h>
59#include <asm/fpu/xstate.h>
60#include <asm/trace/mpx.h>
61#include <asm/mpx.h>
62#include <asm/vm86.h>
63#include <asm/umip.h>
64
65#ifdef CONFIG_X86_64
66#include <asm/x86_init.h>
67#include <asm/pgalloc.h>
68#include <asm/proto.h>
69#else
70#include <asm/processor-flags.h>
71#include <asm/setup.h>
72#include <asm/proto.h>
73#endif
74
75DECLARE_BITMAP(system_vectors, NR_VECTORS);
76
77static inline void cond_local_irq_enable(struct pt_regs *regs)
78{
79 if (regs->flags & X86_EFLAGS_IF)
80 local_irq_enable();
81}
82
83static inline void cond_local_irq_disable(struct pt_regs *regs)
84{
85 if (regs->flags & X86_EFLAGS_IF)
86 local_irq_disable();
87}
88
89/*
90 * In IST context, we explicitly disable preemption. This serves two
91 * purposes: it makes it much less likely that we would accidentally
92 * schedule in IST context and it will force a warning if we somehow
93 * manage to schedule by accident.
94 */
95void ist_enter(struct pt_regs *regs)
96{
97 if (user_mode(regs)) {
98 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
99 } else {
100 /*
101 * We might have interrupted pretty much anything. In
102 * fact, if we're a machine check, we can even interrupt
103 * NMI processing. We don't want in_nmi() to return true,
104 * but we need to notify RCU.
105 */
106 rcu_nmi_enter();
107 }
108
109 preempt_disable();
110
111 /* This code is a bit fragile. Test it. */
112 RCU_LOCKDEP_WARN(!rcu_is_watching(), "ist_enter didn't work");
113}
114NOKPROBE_SYMBOL(ist_enter);
115
116void ist_exit(struct pt_regs *regs)
117{
118 preempt_enable_no_resched();
119
120 if (!user_mode(regs))
121 rcu_nmi_exit();
122}
123
124/**
125 * ist_begin_non_atomic() - begin a non-atomic section in an IST exception
126 * @regs: regs passed to the IST exception handler
127 *
128 * IST exception handlers normally cannot schedule. As a special
129 * exception, if the exception interrupted userspace code (i.e.
130 * user_mode(regs) would return true) and the exception was not
131 * a double fault, it can be safe to schedule. ist_begin_non_atomic()
132 * begins a non-atomic section within an ist_enter()/ist_exit() region.
133 * Callers are responsible for enabling interrupts themselves inside
134 * the non-atomic section, and callers must call ist_end_non_atomic()
135 * before ist_exit().
136 */
137void ist_begin_non_atomic(struct pt_regs *regs)
138{
139 BUG_ON(!user_mode(regs));
140
141 /*
142 * Sanity check: we need to be on the normal thread stack. This
143 * will catch asm bugs and any attempt to use ist_preempt_enable
144 * from double_fault.
145 */
146 BUG_ON(!on_thread_stack());
147
148 preempt_enable_no_resched();
149}
150
151/**
152 * ist_end_non_atomic() - begin a non-atomic section in an IST exception
153 *
154 * Ends a non-atomic section started with ist_begin_non_atomic().
155 */
156void ist_end_non_atomic(void)
157{
158 preempt_disable();
159}
160
161int is_valid_bugaddr(unsigned long addr)
162{
163 unsigned short ud;
164
165 if (addr < TASK_SIZE_MAX)
166 return 0;
167
168 if (probe_kernel_address((unsigned short *)addr, ud))
169 return 0;
170
171 return ud == INSN_UD0 || ud == INSN_UD2;
172}
173
174int fixup_bug(struct pt_regs *regs, int trapnr)
175{
176 if (trapnr != X86_TRAP_UD)
177 return 0;
178
179 switch (report_bug(regs->ip, regs)) {
180 case BUG_TRAP_TYPE_NONE:
181 case BUG_TRAP_TYPE_BUG:
182 break;
183
184 case BUG_TRAP_TYPE_WARN:
185 regs->ip += LEN_UD2;
186 return 1;
187 }
188
189 return 0;
190}
191
192static nokprobe_inline int
193do_trap_no_signal(struct task_struct *tsk, int trapnr, const char *str,
194 struct pt_regs *regs, long error_code)
195{
196 if (v8086_mode(regs)) {
197 /*
198 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
199 * On nmi (interrupt 2), do_trap should not be called.
200 */
201 if (trapnr < X86_TRAP_UD) {
202 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
203 error_code, trapnr))
204 return 0;
205 }
206 } else if (!user_mode(regs)) {
207 if (fixup_exception(regs, trapnr, error_code, 0))
208 return 0;
209
210 tsk->thread.error_code = error_code;
211 tsk->thread.trap_nr = trapnr;
212 die(str, regs, error_code);
213 }
214
215 /*
216 * We want error_code and trap_nr set for userspace faults and
217 * kernelspace faults which result in die(), but not
218 * kernelspace faults which are fixed up. die() gives the
219 * process no chance to handle the signal and notice the
220 * kernel fault information, so that won't result in polluting
221 * the information about previously queued, but not yet
222 * delivered, faults. See also do_general_protection below.
223 */
224 tsk->thread.error_code = error_code;
225 tsk->thread.trap_nr = trapnr;
226
227 return -1;
228}
229
230static void show_signal(struct task_struct *tsk, int signr,
231 const char *type, const char *desc,
232 struct pt_regs *regs, long error_code)
233{
234 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
235 printk_ratelimit()) {
236 pr_info("%s[%d] %s%s ip:%lx sp:%lx error:%lx",
237 tsk->comm, task_pid_nr(tsk), type, desc,
238 regs->ip, regs->sp, error_code);
239 print_vma_addr(KERN_CONT " in ", regs->ip);
240 pr_cont("\n");
241 }
242}
243
244static void
245do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
246 long error_code, int sicode, void __user *addr)
247{
248 struct task_struct *tsk = current;
249
250
251 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
252 return;
253
254 show_signal(tsk, signr, "trap ", str, regs, error_code);
255
256 if (!sicode)
257 force_sig(signr);
258 else
259 force_sig_fault(signr, sicode, addr);
260}
261NOKPROBE_SYMBOL(do_trap);
262
263static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
264 unsigned long trapnr, int signr, int sicode, void __user *addr)
265{
266 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
267
268 /*
269 * WARN*()s end up here; fix them up before we call the
270 * notifier chain.
271 */
272 if (!user_mode(regs) && fixup_bug(regs, trapnr))
273 return;
274
275 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
276 NOTIFY_STOP) {
277 cond_local_irq_enable(regs);
278 do_trap(trapnr, signr, str, regs, error_code, sicode, addr);
279 }
280}
281
282#define IP ((void __user *)uprobe_get_trap_addr(regs))
283#define DO_ERROR(trapnr, signr, sicode, addr, str, name) \
284dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
285{ \
286 do_error_trap(regs, error_code, str, trapnr, signr, sicode, addr); \
287}
288
289DO_ERROR(X86_TRAP_DE, SIGFPE, FPE_INTDIV, IP, "divide error", divide_error)
290DO_ERROR(X86_TRAP_OF, SIGSEGV, 0, NULL, "overflow", overflow)
291DO_ERROR(X86_TRAP_UD, SIGILL, ILL_ILLOPN, IP, "invalid opcode", invalid_op)
292DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, 0, NULL, "coprocessor segment overrun", coprocessor_segment_overrun)
293DO_ERROR(X86_TRAP_TS, SIGSEGV, 0, NULL, "invalid TSS", invalid_TSS)
294DO_ERROR(X86_TRAP_NP, SIGBUS, 0, NULL, "segment not present", segment_not_present)
295DO_ERROR(X86_TRAP_SS, SIGBUS, 0, NULL, "stack segment", stack_segment)
296DO_ERROR(X86_TRAP_AC, SIGBUS, BUS_ADRALN, NULL, "alignment check", alignment_check)
297#undef IP
298
299#ifdef CONFIG_VMAP_STACK
300__visible void __noreturn handle_stack_overflow(const char *message,
301 struct pt_regs *regs,
302 unsigned long fault_address)
303{
304 printk(KERN_EMERG "BUG: stack guard page was hit at %p (stack is %p..%p)\n",
305 (void *)fault_address, current->stack,
306 (char *)current->stack + THREAD_SIZE - 1);
307 die(message, regs, 0);
308
309 /* Be absolutely certain we don't return. */
310 panic("%s", message);
311}
312#endif
313
314#ifdef CONFIG_X86_64
315/* Runs on IST stack */
316dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code, unsigned long cr2)
317{
318 static const char str[] = "double fault";
319 struct task_struct *tsk = current;
320
321#ifdef CONFIG_X86_ESPFIX64
322 extern unsigned char native_irq_return_iret[];
323
324 /*
325 * If IRET takes a non-IST fault on the espfix64 stack, then we
326 * end up promoting it to a doublefault. In that case, take
327 * advantage of the fact that we're not using the normal (TSS.sp0)
328 * stack right now. We can write a fake #GP(0) frame at TSS.sp0
329 * and then modify our own IRET frame so that, when we return,
330 * we land directly at the #GP(0) vector with the stack already
331 * set up according to its expectations.
332 *
333 * The net result is that our #GP handler will think that we
334 * entered from usermode with the bad user context.
335 *
336 * No need for ist_enter here because we don't use RCU.
337 */
338 if (((long)regs->sp >> P4D_SHIFT) == ESPFIX_PGD_ENTRY &&
339 regs->cs == __KERNEL_CS &&
340 regs->ip == (unsigned long)native_irq_return_iret)
341 {
342 struct pt_regs *gpregs = (struct pt_regs *)this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
343
344 /*
345 * regs->sp points to the failing IRET frame on the
346 * ESPFIX64 stack. Copy it to the entry stack. This fills
347 * in gpregs->ss through gpregs->ip.
348 *
349 */
350 memmove(&gpregs->ip, (void *)regs->sp, 5*8);
351 gpregs->orig_ax = 0; /* Missing (lost) #GP error code */
352
353 /*
354 * Adjust our frame so that we return straight to the #GP
355 * vector with the expected RSP value. This is safe because
356 * we won't enable interupts or schedule before we invoke
357 * general_protection, so nothing will clobber the stack
358 * frame we just set up.
359 *
360 * We will enter general_protection with kernel GSBASE,
361 * which is what the stub expects, given that the faulting
362 * RIP will be the IRET instruction.
363 */
364 regs->ip = (unsigned long)general_protection;
365 regs->sp = (unsigned long)&gpregs->orig_ax;
366
367 return;
368 }
369#endif
370
371 ist_enter(regs);
372 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
373
374 tsk->thread.error_code = error_code;
375 tsk->thread.trap_nr = X86_TRAP_DF;
376
377#ifdef CONFIG_VMAP_STACK
378 /*
379 * If we overflow the stack into a guard page, the CPU will fail
380 * to deliver #PF and will send #DF instead. Similarly, if we
381 * take any non-IST exception while too close to the bottom of
382 * the stack, the processor will get a page fault while
383 * delivering the exception and will generate a double fault.
384 *
385 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
386 * Page-Fault Exception (#PF):
387 *
388 * Processors update CR2 whenever a page fault is detected. If a
389 * second page fault occurs while an earlier page fault is being
390 * delivered, the faulting linear address of the second fault will
391 * overwrite the contents of CR2 (replacing the previous
392 * address). These updates to CR2 occur even if the page fault
393 * results in a double fault or occurs during the delivery of a
394 * double fault.
395 *
396 * The logic below has a small possibility of incorrectly diagnosing
397 * some errors as stack overflows. For example, if the IDT or GDT
398 * gets corrupted such that #GP delivery fails due to a bad descriptor
399 * causing #GP and we hit this condition while CR2 coincidentally
400 * points to the stack guard page, we'll think we overflowed the
401 * stack. Given that we're going to panic one way or another
402 * if this happens, this isn't necessarily worth fixing.
403 *
404 * If necessary, we could improve the test by only diagnosing
405 * a stack overflow if the saved RSP points within 47 bytes of
406 * the bottom of the stack: if RSP == tsk_stack + 48 and we
407 * take an exception, the stack is already aligned and there
408 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
409 * possible error code, so a stack overflow would *not* double
410 * fault. With any less space left, exception delivery could
411 * fail, and, as a practical matter, we've overflowed the
412 * stack even if the actual trigger for the double fault was
413 * something else.
414 */
415 if ((unsigned long)task_stack_page(tsk) - 1 - cr2 < PAGE_SIZE)
416 handle_stack_overflow("kernel stack overflow (double-fault)", regs, cr2);
417#endif
418
419#ifdef CONFIG_DOUBLEFAULT
420 df_debug(regs, error_code);
421#endif
422 /*
423 * This is always a kernel trap and never fixable (and thus must
424 * never return).
425 */
426 for (;;)
427 die(str, regs, error_code);
428}
429#endif
430
431dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
432{
433 const struct mpx_bndcsr *bndcsr;
434
435 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
436 if (notify_die(DIE_TRAP, "bounds", regs, error_code,
437 X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
438 return;
439 cond_local_irq_enable(regs);
440
441 if (!user_mode(regs))
442 die("bounds", regs, error_code);
443
444 if (!cpu_feature_enabled(X86_FEATURE_MPX)) {
445 /* The exception is not from Intel MPX */
446 goto exit_trap;
447 }
448
449 /*
450 * We need to look at BNDSTATUS to resolve this exception.
451 * A NULL here might mean that it is in its 'init state',
452 * which is all zeros which indicates MPX was not
453 * responsible for the exception.
454 */
455 bndcsr = get_xsave_field_ptr(XFEATURE_BNDCSR);
456 if (!bndcsr)
457 goto exit_trap;
458
459 trace_bounds_exception_mpx(bndcsr);
460 /*
461 * The error code field of the BNDSTATUS register communicates status
462 * information of a bound range exception #BR or operation involving
463 * bound directory.
464 */
465 switch (bndcsr->bndstatus & MPX_BNDSTA_ERROR_CODE) {
466 case 2: /* Bound directory has invalid entry. */
467 if (mpx_handle_bd_fault())
468 goto exit_trap;
469 break; /* Success, it was handled */
470 case 1: /* Bound violation. */
471 {
472 struct task_struct *tsk = current;
473 struct mpx_fault_info mpx;
474
475 if (mpx_fault_info(&mpx, regs)) {
476 /*
477 * We failed to decode the MPX instruction. Act as if
478 * the exception was not caused by MPX.
479 */
480 goto exit_trap;
481 }
482 /*
483 * Success, we decoded the instruction and retrieved
484 * an 'mpx' containing the address being accessed
485 * which caused the exception. This information
486 * allows and application to possibly handle the
487 * #BR exception itself.
488 */
489 if (!do_trap_no_signal(tsk, X86_TRAP_BR, "bounds", regs,
490 error_code))
491 break;
492
493 show_signal(tsk, SIGSEGV, "trap ", "bounds", regs, error_code);
494
495 force_sig_bnderr(mpx.addr, mpx.lower, mpx.upper);
496 break;
497 }
498 case 0: /* No exception caused by Intel MPX operations. */
499 goto exit_trap;
500 default:
501 die("bounds", regs, error_code);
502 }
503
504 return;
505
506exit_trap:
507 /*
508 * This path out is for all the cases where we could not
509 * handle the exception in some way (like allocating a
510 * table or telling userspace about it. We will also end
511 * up here if the kernel has MPX turned off at compile
512 * time..
513 */
514 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, 0, NULL);
515}
516
517dotraplinkage void
518do_general_protection(struct pt_regs *regs, long error_code)
519{
520 const char *desc = "general protection fault";
521 struct task_struct *tsk;
522
523 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
524 cond_local_irq_enable(regs);
525
526 if (static_cpu_has(X86_FEATURE_UMIP)) {
527 if (user_mode(regs) && fixup_umip_exception(regs))
528 return;
529 }
530
531 if (v8086_mode(regs)) {
532 local_irq_enable();
533 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
534 return;
535 }
536
537 tsk = current;
538 if (!user_mode(regs)) {
539 if (fixup_exception(regs, X86_TRAP_GP, error_code, 0))
540 return;
541
542 tsk->thread.error_code = error_code;
543 tsk->thread.trap_nr = X86_TRAP_GP;
544
545 /*
546 * To be potentially processing a kprobe fault and to
547 * trust the result from kprobe_running(), we have to
548 * be non-preemptible.
549 */
550 if (!preemptible() && kprobe_running() &&
551 kprobe_fault_handler(regs, X86_TRAP_GP))
552 return;
553
554 if (notify_die(DIE_GPF, desc, regs, error_code,
555 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
556 die(desc, regs, error_code);
557 return;
558 }
559
560 tsk->thread.error_code = error_code;
561 tsk->thread.trap_nr = X86_TRAP_GP;
562
563 show_signal(tsk, SIGSEGV, "", desc, regs, error_code);
564
565 force_sig(SIGSEGV);
566}
567NOKPROBE_SYMBOL(do_general_protection);
568
569dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
570{
571#ifdef CONFIG_DYNAMIC_FTRACE
572 /*
573 * ftrace must be first, everything else may cause a recursive crash.
574 * See note by declaration of modifying_ftrace_code in ftrace.c
575 */
576 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
577 ftrace_int3_handler(regs))
578 return;
579#endif
580 if (poke_int3_handler(regs))
581 return;
582
583 /*
584 * Use ist_enter despite the fact that we don't use an IST stack.
585 * We can be called from a kprobe in non-CONTEXT_KERNEL kernel
586 * mode or even during context tracking state changes.
587 *
588 * This means that we can't schedule. That's okay.
589 */
590 ist_enter(regs);
591 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
592#ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
593 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
594 SIGTRAP) == NOTIFY_STOP)
595 goto exit;
596#endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
597
598#ifdef CONFIG_KPROBES
599 if (kprobe_int3_handler(regs))
600 goto exit;
601#endif
602
603 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
604 SIGTRAP) == NOTIFY_STOP)
605 goto exit;
606
607 cond_local_irq_enable(regs);
608 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, 0, NULL);
609 cond_local_irq_disable(regs);
610
611exit:
612 ist_exit(regs);
613}
614NOKPROBE_SYMBOL(do_int3);
615
616#ifdef CONFIG_X86_64
617/*
618 * Help handler running on a per-cpu (IST or entry trampoline) stack
619 * to switch to the normal thread stack if the interrupted code was in
620 * user mode. The actual stack switch is done in entry_64.S
621 */
622asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
623{
624 struct pt_regs *regs = (struct pt_regs *)this_cpu_read(cpu_current_top_of_stack) - 1;
625 if (regs != eregs)
626 *regs = *eregs;
627 return regs;
628}
629NOKPROBE_SYMBOL(sync_regs);
630
631struct bad_iret_stack {
632 void *error_entry_ret;
633 struct pt_regs regs;
634};
635
636asmlinkage __visible notrace
637struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
638{
639 /*
640 * This is called from entry_64.S early in handling a fault
641 * caused by a bad iret to user mode. To handle the fault
642 * correctly, we want to move our stack frame to where it would
643 * be had we entered directly on the entry stack (rather than
644 * just below the IRET frame) and we want to pretend that the
645 * exception came from the IRET target.
646 */
647 struct bad_iret_stack *new_stack =
648 (struct bad_iret_stack *)this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
649
650 /* Copy the IRET target to the new stack. */
651 memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
652
653 /* Copy the remainder of the stack from the current stack. */
654 memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
655
656 BUG_ON(!user_mode(&new_stack->regs));
657 return new_stack;
658}
659NOKPROBE_SYMBOL(fixup_bad_iret);
660#endif
661
662static bool is_sysenter_singlestep(struct pt_regs *regs)
663{
664 /*
665 * We don't try for precision here. If we're anywhere in the region of
666 * code that can be single-stepped in the SYSENTER entry path, then
667 * assume that this is a useless single-step trap due to SYSENTER
668 * being invoked with TF set. (We don't know in advance exactly
669 * which instructions will be hit because BTF could plausibly
670 * be set.)
671 */
672#ifdef CONFIG_X86_32
673 return (regs->ip - (unsigned long)__begin_SYSENTER_singlestep_region) <
674 (unsigned long)__end_SYSENTER_singlestep_region -
675 (unsigned long)__begin_SYSENTER_singlestep_region;
676#elif defined(CONFIG_IA32_EMULATION)
677 return (regs->ip - (unsigned long)entry_SYSENTER_compat) <
678 (unsigned long)__end_entry_SYSENTER_compat -
679 (unsigned long)entry_SYSENTER_compat;
680#else
681 return false;
682#endif
683}
684
685/*
686 * Our handling of the processor debug registers is non-trivial.
687 * We do not clear them on entry and exit from the kernel. Therefore
688 * it is possible to get a watchpoint trap here from inside the kernel.
689 * However, the code in ./ptrace.c has ensured that the user can
690 * only set watchpoints on userspace addresses. Therefore the in-kernel
691 * watchpoint trap can only occur in code which is reading/writing
692 * from user space. Such code must not hold kernel locks (since it
693 * can equally take a page fault), therefore it is safe to call
694 * force_sig_info even though that claims and releases locks.
695 *
696 * Code in ./signal.c ensures that the debug control register
697 * is restored before we deliver any signal, and therefore that
698 * user code runs with the correct debug control register even though
699 * we clear it here.
700 *
701 * Being careful here means that we don't have to be as careful in a
702 * lot of more complicated places (task switching can be a bit lazy
703 * about restoring all the debug state, and ptrace doesn't have to
704 * find every occurrence of the TF bit that could be saved away even
705 * by user code)
706 *
707 * May run on IST stack.
708 */
709dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
710{
711 struct task_struct *tsk = current;
712 int user_icebp = 0;
713 unsigned long dr6;
714 int si_code;
715
716 ist_enter(regs);
717
718 get_debugreg(dr6, 6);
719 /*
720 * The Intel SDM says:
721 *
722 * Certain debug exceptions may clear bits 0-3. The remaining
723 * contents of the DR6 register are never cleared by the
724 * processor. To avoid confusion in identifying debug
725 * exceptions, debug handlers should clear the register before
726 * returning to the interrupted task.
727 *
728 * Keep it simple: clear DR6 immediately.
729 */
730 set_debugreg(0, 6);
731
732 /* Filter out all the reserved bits which are preset to 1 */
733 dr6 &= ~DR6_RESERVED;
734
735 /*
736 * The SDM says "The processor clears the BTF flag when it
737 * generates a debug exception." Clear TIF_BLOCKSTEP to keep
738 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
739 */
740 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
741
742 if (unlikely(!user_mode(regs) && (dr6 & DR_STEP) &&
743 is_sysenter_singlestep(regs))) {
744 dr6 &= ~DR_STEP;
745 if (!dr6)
746 goto exit;
747 /*
748 * else we might have gotten a single-step trap and hit a
749 * watchpoint at the same time, in which case we should fall
750 * through and handle the watchpoint.
751 */
752 }
753
754 /*
755 * If dr6 has no reason to give us about the origin of this trap,
756 * then it's very likely the result of an icebp/int01 trap.
757 * User wants a sigtrap for that.
758 */
759 if (!dr6 && user_mode(regs))
760 user_icebp = 1;
761
762 /* Store the virtualized DR6 value */
763 tsk->thread.debugreg6 = dr6;
764
765#ifdef CONFIG_KPROBES
766 if (kprobe_debug_handler(regs))
767 goto exit;
768#endif
769
770 if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
771 SIGTRAP) == NOTIFY_STOP)
772 goto exit;
773
774 /*
775 * Let others (NMI) know that the debug stack is in use
776 * as we may switch to the interrupt stack.
777 */
778 debug_stack_usage_inc();
779
780 /* It's safe to allow irq's after DR6 has been saved */
781 cond_local_irq_enable(regs);
782
783 if (v8086_mode(regs)) {
784 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
785 X86_TRAP_DB);
786 cond_local_irq_disable(regs);
787 debug_stack_usage_dec();
788 goto exit;
789 }
790
791 if (WARN_ON_ONCE((dr6 & DR_STEP) && !user_mode(regs))) {
792 /*
793 * Historical junk that used to handle SYSENTER single-stepping.
794 * This should be unreachable now. If we survive for a while
795 * without anyone hitting this warning, we'll turn this into
796 * an oops.
797 */
798 tsk->thread.debugreg6 &= ~DR_STEP;
799 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
800 regs->flags &= ~X86_EFLAGS_TF;
801 }
802 si_code = get_si_code(tsk->thread.debugreg6);
803 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
804 send_sigtrap(regs, error_code, si_code);
805 cond_local_irq_disable(regs);
806 debug_stack_usage_dec();
807
808exit:
809 ist_exit(regs);
810}
811NOKPROBE_SYMBOL(do_debug);
812
813/*
814 * Note that we play around with the 'TS' bit in an attempt to get
815 * the correct behaviour even in the presence of the asynchronous
816 * IRQ13 behaviour
817 */
818static void math_error(struct pt_regs *regs, int error_code, int trapnr)
819{
820 struct task_struct *task = current;
821 struct fpu *fpu = &task->thread.fpu;
822 int si_code;
823 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
824 "simd exception";
825
826 cond_local_irq_enable(regs);
827
828 if (!user_mode(regs)) {
829 if (fixup_exception(regs, trapnr, error_code, 0))
830 return;
831
832 task->thread.error_code = error_code;
833 task->thread.trap_nr = trapnr;
834
835 if (notify_die(DIE_TRAP, str, regs, error_code,
836 trapnr, SIGFPE) != NOTIFY_STOP)
837 die(str, regs, error_code);
838 return;
839 }
840
841 /*
842 * Save the info for the exception handler and clear the error.
843 */
844 fpu__save(fpu);
845
846 task->thread.trap_nr = trapnr;
847 task->thread.error_code = error_code;
848
849 si_code = fpu__exception_code(fpu, trapnr);
850 /* Retry when we get spurious exceptions: */
851 if (!si_code)
852 return;
853
854 force_sig_fault(SIGFPE, si_code,
855 (void __user *)uprobe_get_trap_addr(regs));
856}
857
858dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
859{
860 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
861 math_error(regs, error_code, X86_TRAP_MF);
862}
863
864dotraplinkage void
865do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
866{
867 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
868 math_error(regs, error_code, X86_TRAP_XF);
869}
870
871dotraplinkage void
872do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
873{
874 cond_local_irq_enable(regs);
875}
876
877dotraplinkage void
878do_device_not_available(struct pt_regs *regs, long error_code)
879{
880 unsigned long cr0 = read_cr0();
881
882 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
883
884#ifdef CONFIG_MATH_EMULATION
885 if (!boot_cpu_has(X86_FEATURE_FPU) && (cr0 & X86_CR0_EM)) {
886 struct math_emu_info info = { };
887
888 cond_local_irq_enable(regs);
889
890 info.regs = regs;
891 math_emulate(&info);
892 return;
893 }
894#endif
895
896 /* This should not happen. */
897 if (WARN(cr0 & X86_CR0_TS, "CR0.TS was set")) {
898 /* Try to fix it up and carry on. */
899 write_cr0(cr0 & ~X86_CR0_TS);
900 } else {
901 /*
902 * Something terrible happened, and we're better off trying
903 * to kill the task than getting stuck in a never-ending
904 * loop of #NM faults.
905 */
906 die("unexpected #NM exception", regs, error_code);
907 }
908}
909NOKPROBE_SYMBOL(do_device_not_available);
910
911#ifdef CONFIG_X86_32
912dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
913{
914 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
915 local_irq_enable();
916
917 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
918 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
919 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
920 ILL_BADSTK, (void __user *)NULL);
921 }
922}
923#endif
924
925void __init trap_init(void)
926{
927 /* Init cpu_entry_area before IST entries are set up */
928 setup_cpu_entry_areas();
929
930 idt_setup_traps();
931
932 /*
933 * Set the IDT descriptor to a fixed read-only location, so that the
934 * "sidt" instruction will not leak the location of the kernel, and
935 * to defend the IDT against arbitrary memory write vulnerabilities.
936 * It will be reloaded in cpu_init() */
937 cea_set_pte(CPU_ENTRY_AREA_RO_IDT_VADDR, __pa_symbol(idt_table),
938 PAGE_KERNEL_RO);
939 idt_descr.address = CPU_ENTRY_AREA_RO_IDT;
940
941 /*
942 * Should be a barrier for any external CPU state:
943 */
944 cpu_init();
945
946 idt_setup_ist_traps();
947
948 x86_init.irqs.trap_init();
949
950 idt_setup_debugidt_traps();
951}
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 *
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 */
8
9/*
10 * Handle hardware traps and faults.
11 */
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/context_tracking.h>
16#include <linux/interrupt.h>
17#include <linux/kallsyms.h>
18#include <linux/kmsan.h>
19#include <linux/spinlock.h>
20#include <linux/kprobes.h>
21#include <linux/uaccess.h>
22#include <linux/kdebug.h>
23#include <linux/kgdb.h>
24#include <linux/kernel.h>
25#include <linux/export.h>
26#include <linux/ptrace.h>
27#include <linux/uprobes.h>
28#include <linux/string.h>
29#include <linux/delay.h>
30#include <linux/errno.h>
31#include <linux/kexec.h>
32#include <linux/sched.h>
33#include <linux/sched/task_stack.h>
34#include <linux/timer.h>
35#include <linux/init.h>
36#include <linux/bug.h>
37#include <linux/nmi.h>
38#include <linux/mm.h>
39#include <linux/smp.h>
40#include <linux/cpu.h>
41#include <linux/io.h>
42#include <linux/hardirq.h>
43#include <linux/atomic.h>
44#include <linux/iommu.h>
45
46#include <asm/stacktrace.h>
47#include <asm/processor.h>
48#include <asm/debugreg.h>
49#include <asm/realmode.h>
50#include <asm/text-patching.h>
51#include <asm/ftrace.h>
52#include <asm/traps.h>
53#include <asm/desc.h>
54#include <asm/fpu/api.h>
55#include <asm/cpu.h>
56#include <asm/cpu_entry_area.h>
57#include <asm/mce.h>
58#include <asm/fixmap.h>
59#include <asm/mach_traps.h>
60#include <asm/alternative.h>
61#include <asm/fpu/xstate.h>
62#include <asm/vm86.h>
63#include <asm/umip.h>
64#include <asm/insn.h>
65#include <asm/insn-eval.h>
66#include <asm/vdso.h>
67#include <asm/tdx.h>
68#include <asm/cfi.h>
69
70#ifdef CONFIG_X86_64
71#include <asm/x86_init.h>
72#else
73#include <asm/processor-flags.h>
74#include <asm/setup.h>
75#endif
76
77#include <asm/proto.h>
78
79DECLARE_BITMAP(system_vectors, NR_VECTORS);
80
81__always_inline int is_valid_bugaddr(unsigned long addr)
82{
83 if (addr < TASK_SIZE_MAX)
84 return 0;
85
86 /*
87 * We got #UD, if the text isn't readable we'd have gotten
88 * a different exception.
89 */
90 return *(unsigned short *)addr == INSN_UD2;
91}
92
93static nokprobe_inline int
94do_trap_no_signal(struct task_struct *tsk, int trapnr, const char *str,
95 struct pt_regs *regs, long error_code)
96{
97 if (v8086_mode(regs)) {
98 /*
99 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
100 * On nmi (interrupt 2), do_trap should not be called.
101 */
102 if (trapnr < X86_TRAP_UD) {
103 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
104 error_code, trapnr))
105 return 0;
106 }
107 } else if (!user_mode(regs)) {
108 if (fixup_exception(regs, trapnr, error_code, 0))
109 return 0;
110
111 tsk->thread.error_code = error_code;
112 tsk->thread.trap_nr = trapnr;
113 die(str, regs, error_code);
114 } else {
115 if (fixup_vdso_exception(regs, trapnr, error_code, 0))
116 return 0;
117 }
118
119 /*
120 * We want error_code and trap_nr set for userspace faults and
121 * kernelspace faults which result in die(), but not
122 * kernelspace faults which are fixed up. die() gives the
123 * process no chance to handle the signal and notice the
124 * kernel fault information, so that won't result in polluting
125 * the information about previously queued, but not yet
126 * delivered, faults. See also exc_general_protection below.
127 */
128 tsk->thread.error_code = error_code;
129 tsk->thread.trap_nr = trapnr;
130
131 return -1;
132}
133
134static void show_signal(struct task_struct *tsk, int signr,
135 const char *type, const char *desc,
136 struct pt_regs *regs, long error_code)
137{
138 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
139 printk_ratelimit()) {
140 pr_info("%s[%d] %s%s ip:%lx sp:%lx error:%lx",
141 tsk->comm, task_pid_nr(tsk), type, desc,
142 regs->ip, regs->sp, error_code);
143 print_vma_addr(KERN_CONT " in ", regs->ip);
144 pr_cont("\n");
145 }
146}
147
148static void
149do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
150 long error_code, int sicode, void __user *addr)
151{
152 struct task_struct *tsk = current;
153
154 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
155 return;
156
157 show_signal(tsk, signr, "trap ", str, regs, error_code);
158
159 if (!sicode)
160 force_sig(signr);
161 else
162 force_sig_fault(signr, sicode, addr);
163}
164NOKPROBE_SYMBOL(do_trap);
165
166static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
167 unsigned long trapnr, int signr, int sicode, void __user *addr)
168{
169 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
170
171 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
172 NOTIFY_STOP) {
173 cond_local_irq_enable(regs);
174 do_trap(trapnr, signr, str, regs, error_code, sicode, addr);
175 cond_local_irq_disable(regs);
176 }
177}
178
179/*
180 * Posix requires to provide the address of the faulting instruction for
181 * SIGILL (#UD) and SIGFPE (#DE) in the si_addr member of siginfo_t.
182 *
183 * This address is usually regs->ip, but when an uprobe moved the code out
184 * of line then regs->ip points to the XOL code which would confuse
185 * anything which analyzes the fault address vs. the unmodified binary. If
186 * a trap happened in XOL code then uprobe maps regs->ip back to the
187 * original instruction address.
188 */
189static __always_inline void __user *error_get_trap_addr(struct pt_regs *regs)
190{
191 return (void __user *)uprobe_get_trap_addr(regs);
192}
193
194DEFINE_IDTENTRY(exc_divide_error)
195{
196 do_error_trap(regs, 0, "divide error", X86_TRAP_DE, SIGFPE,
197 FPE_INTDIV, error_get_trap_addr(regs));
198}
199
200DEFINE_IDTENTRY(exc_overflow)
201{
202 do_error_trap(regs, 0, "overflow", X86_TRAP_OF, SIGSEGV, 0, NULL);
203}
204
205#ifdef CONFIG_X86_F00F_BUG
206void handle_invalid_op(struct pt_regs *regs)
207#else
208static inline void handle_invalid_op(struct pt_regs *regs)
209#endif
210{
211 do_error_trap(regs, 0, "invalid opcode", X86_TRAP_UD, SIGILL,
212 ILL_ILLOPN, error_get_trap_addr(regs));
213}
214
215static noinstr bool handle_bug(struct pt_regs *regs)
216{
217 bool handled = false;
218
219 /*
220 * Normally @regs are unpoisoned by irqentry_enter(), but handle_bug()
221 * is a rare case that uses @regs without passing them to
222 * irqentry_enter().
223 */
224 kmsan_unpoison_entry_regs(regs);
225 if (!is_valid_bugaddr(regs->ip))
226 return handled;
227
228 /*
229 * All lies, just get the WARN/BUG out.
230 */
231 instrumentation_begin();
232 /*
233 * Since we're emulating a CALL with exceptions, restore the interrupt
234 * state to what it was at the exception site.
235 */
236 if (regs->flags & X86_EFLAGS_IF)
237 raw_local_irq_enable();
238 if (report_bug(regs->ip, regs) == BUG_TRAP_TYPE_WARN ||
239 handle_cfi_failure(regs) == BUG_TRAP_TYPE_WARN) {
240 regs->ip += LEN_UD2;
241 handled = true;
242 }
243 if (regs->flags & X86_EFLAGS_IF)
244 raw_local_irq_disable();
245 instrumentation_end();
246
247 return handled;
248}
249
250DEFINE_IDTENTRY_RAW(exc_invalid_op)
251{
252 irqentry_state_t state;
253
254 /*
255 * We use UD2 as a short encoding for 'CALL __WARN', as such
256 * handle it before exception entry to avoid recursive WARN
257 * in case exception entry is the one triggering WARNs.
258 */
259 if (!user_mode(regs) && handle_bug(regs))
260 return;
261
262 state = irqentry_enter(regs);
263 instrumentation_begin();
264 handle_invalid_op(regs);
265 instrumentation_end();
266 irqentry_exit(regs, state);
267}
268
269DEFINE_IDTENTRY(exc_coproc_segment_overrun)
270{
271 do_error_trap(regs, 0, "coprocessor segment overrun",
272 X86_TRAP_OLD_MF, SIGFPE, 0, NULL);
273}
274
275DEFINE_IDTENTRY_ERRORCODE(exc_invalid_tss)
276{
277 do_error_trap(regs, error_code, "invalid TSS", X86_TRAP_TS, SIGSEGV,
278 0, NULL);
279}
280
281DEFINE_IDTENTRY_ERRORCODE(exc_segment_not_present)
282{
283 do_error_trap(regs, error_code, "segment not present", X86_TRAP_NP,
284 SIGBUS, 0, NULL);
285}
286
287DEFINE_IDTENTRY_ERRORCODE(exc_stack_segment)
288{
289 do_error_trap(regs, error_code, "stack segment", X86_TRAP_SS, SIGBUS,
290 0, NULL);
291}
292
293DEFINE_IDTENTRY_ERRORCODE(exc_alignment_check)
294{
295 char *str = "alignment check";
296
297 if (notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_AC, SIGBUS) == NOTIFY_STOP)
298 return;
299
300 if (!user_mode(regs))
301 die("Split lock detected\n", regs, error_code);
302
303 local_irq_enable();
304
305 if (handle_user_split_lock(regs, error_code))
306 goto out;
307
308 do_trap(X86_TRAP_AC, SIGBUS, "alignment check", regs,
309 error_code, BUS_ADRALN, NULL);
310
311out:
312 local_irq_disable();
313}
314
315#ifdef CONFIG_VMAP_STACK
316__visible void __noreturn handle_stack_overflow(struct pt_regs *regs,
317 unsigned long fault_address,
318 struct stack_info *info)
319{
320 const char *name = stack_type_name(info->type);
321
322 printk(KERN_EMERG "BUG: %s stack guard page was hit at %p (stack is %p..%p)\n",
323 name, (void *)fault_address, info->begin, info->end);
324
325 die("stack guard page", regs, 0);
326
327 /* Be absolutely certain we don't return. */
328 panic("%s stack guard hit", name);
329}
330#endif
331
332/*
333 * Runs on an IST stack for x86_64 and on a special task stack for x86_32.
334 *
335 * On x86_64, this is more or less a normal kernel entry. Notwithstanding the
336 * SDM's warnings about double faults being unrecoverable, returning works as
337 * expected. Presumably what the SDM actually means is that the CPU may get
338 * the register state wrong on entry, so returning could be a bad idea.
339 *
340 * Various CPU engineers have promised that double faults due to an IRET fault
341 * while the stack is read-only are, in fact, recoverable.
342 *
343 * On x86_32, this is entered through a task gate, and regs are synthesized
344 * from the TSS. Returning is, in principle, okay, but changes to regs will
345 * be lost. If, for some reason, we need to return to a context with modified
346 * regs, the shim code could be adjusted to synchronize the registers.
347 *
348 * The 32bit #DF shim provides CR2 already as an argument. On 64bit it needs
349 * to be read before doing anything else.
350 */
351DEFINE_IDTENTRY_DF(exc_double_fault)
352{
353 static const char str[] = "double fault";
354 struct task_struct *tsk = current;
355
356#ifdef CONFIG_VMAP_STACK
357 unsigned long address = read_cr2();
358 struct stack_info info;
359#endif
360
361#ifdef CONFIG_X86_ESPFIX64
362 extern unsigned char native_irq_return_iret[];
363
364 /*
365 * If IRET takes a non-IST fault on the espfix64 stack, then we
366 * end up promoting it to a doublefault. In that case, take
367 * advantage of the fact that we're not using the normal (TSS.sp0)
368 * stack right now. We can write a fake #GP(0) frame at TSS.sp0
369 * and then modify our own IRET frame so that, when we return,
370 * we land directly at the #GP(0) vector with the stack already
371 * set up according to its expectations.
372 *
373 * The net result is that our #GP handler will think that we
374 * entered from usermode with the bad user context.
375 *
376 * No need for nmi_enter() here because we don't use RCU.
377 */
378 if (((long)regs->sp >> P4D_SHIFT) == ESPFIX_PGD_ENTRY &&
379 regs->cs == __KERNEL_CS &&
380 regs->ip == (unsigned long)native_irq_return_iret)
381 {
382 struct pt_regs *gpregs = (struct pt_regs *)this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
383 unsigned long *p = (unsigned long *)regs->sp;
384
385 /*
386 * regs->sp points to the failing IRET frame on the
387 * ESPFIX64 stack. Copy it to the entry stack. This fills
388 * in gpregs->ss through gpregs->ip.
389 *
390 */
391 gpregs->ip = p[0];
392 gpregs->cs = p[1];
393 gpregs->flags = p[2];
394 gpregs->sp = p[3];
395 gpregs->ss = p[4];
396 gpregs->orig_ax = 0; /* Missing (lost) #GP error code */
397
398 /*
399 * Adjust our frame so that we return straight to the #GP
400 * vector with the expected RSP value. This is safe because
401 * we won't enable interrupts or schedule before we invoke
402 * general_protection, so nothing will clobber the stack
403 * frame we just set up.
404 *
405 * We will enter general_protection with kernel GSBASE,
406 * which is what the stub expects, given that the faulting
407 * RIP will be the IRET instruction.
408 */
409 regs->ip = (unsigned long)asm_exc_general_protection;
410 regs->sp = (unsigned long)&gpregs->orig_ax;
411
412 return;
413 }
414#endif
415
416 irqentry_nmi_enter(regs);
417 instrumentation_begin();
418 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
419
420 tsk->thread.error_code = error_code;
421 tsk->thread.trap_nr = X86_TRAP_DF;
422
423#ifdef CONFIG_VMAP_STACK
424 /*
425 * If we overflow the stack into a guard page, the CPU will fail
426 * to deliver #PF and will send #DF instead. Similarly, if we
427 * take any non-IST exception while too close to the bottom of
428 * the stack, the processor will get a page fault while
429 * delivering the exception and will generate a double fault.
430 *
431 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
432 * Page-Fault Exception (#PF):
433 *
434 * Processors update CR2 whenever a page fault is detected. If a
435 * second page fault occurs while an earlier page fault is being
436 * delivered, the faulting linear address of the second fault will
437 * overwrite the contents of CR2 (replacing the previous
438 * address). These updates to CR2 occur even if the page fault
439 * results in a double fault or occurs during the delivery of a
440 * double fault.
441 *
442 * The logic below has a small possibility of incorrectly diagnosing
443 * some errors as stack overflows. For example, if the IDT or GDT
444 * gets corrupted such that #GP delivery fails due to a bad descriptor
445 * causing #GP and we hit this condition while CR2 coincidentally
446 * points to the stack guard page, we'll think we overflowed the
447 * stack. Given that we're going to panic one way or another
448 * if this happens, this isn't necessarily worth fixing.
449 *
450 * If necessary, we could improve the test by only diagnosing
451 * a stack overflow if the saved RSP points within 47 bytes of
452 * the bottom of the stack: if RSP == tsk_stack + 48 and we
453 * take an exception, the stack is already aligned and there
454 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
455 * possible error code, so a stack overflow would *not* double
456 * fault. With any less space left, exception delivery could
457 * fail, and, as a practical matter, we've overflowed the
458 * stack even if the actual trigger for the double fault was
459 * something else.
460 */
461 if (get_stack_guard_info((void *)address, &info))
462 handle_stack_overflow(regs, address, &info);
463#endif
464
465 pr_emerg("PANIC: double fault, error_code: 0x%lx\n", error_code);
466 die("double fault", regs, error_code);
467 panic("Machine halted.");
468 instrumentation_end();
469}
470
471DEFINE_IDTENTRY(exc_bounds)
472{
473 if (notify_die(DIE_TRAP, "bounds", regs, 0,
474 X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
475 return;
476 cond_local_irq_enable(regs);
477
478 if (!user_mode(regs))
479 die("bounds", regs, 0);
480
481 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, 0, 0, NULL);
482
483 cond_local_irq_disable(regs);
484}
485
486enum kernel_gp_hint {
487 GP_NO_HINT,
488 GP_NON_CANONICAL,
489 GP_CANONICAL
490};
491
492/*
493 * When an uncaught #GP occurs, try to determine the memory address accessed by
494 * the instruction and return that address to the caller. Also, try to figure
495 * out whether any part of the access to that address was non-canonical.
496 */
497static enum kernel_gp_hint get_kernel_gp_address(struct pt_regs *regs,
498 unsigned long *addr)
499{
500 u8 insn_buf[MAX_INSN_SIZE];
501 struct insn insn;
502 int ret;
503
504 if (copy_from_kernel_nofault(insn_buf, (void *)regs->ip,
505 MAX_INSN_SIZE))
506 return GP_NO_HINT;
507
508 ret = insn_decode_kernel(&insn, insn_buf);
509 if (ret < 0)
510 return GP_NO_HINT;
511
512 *addr = (unsigned long)insn_get_addr_ref(&insn, regs);
513 if (*addr == -1UL)
514 return GP_NO_HINT;
515
516#ifdef CONFIG_X86_64
517 /*
518 * Check that:
519 * - the operand is not in the kernel half
520 * - the last byte of the operand is not in the user canonical half
521 */
522 if (*addr < ~__VIRTUAL_MASK &&
523 *addr + insn.opnd_bytes - 1 > __VIRTUAL_MASK)
524 return GP_NON_CANONICAL;
525#endif
526
527 return GP_CANONICAL;
528}
529
530#define GPFSTR "general protection fault"
531
532static bool fixup_iopl_exception(struct pt_regs *regs)
533{
534 struct thread_struct *t = ¤t->thread;
535 unsigned char byte;
536 unsigned long ip;
537
538 if (!IS_ENABLED(CONFIG_X86_IOPL_IOPERM) || t->iopl_emul != 3)
539 return false;
540
541 if (insn_get_effective_ip(regs, &ip))
542 return false;
543
544 if (get_user(byte, (const char __user *)ip))
545 return false;
546
547 if (byte != 0xfa && byte != 0xfb)
548 return false;
549
550 if (!t->iopl_warn && printk_ratelimit()) {
551 pr_err("%s[%d] attempts to use CLI/STI, pretending it's a NOP, ip:%lx",
552 current->comm, task_pid_nr(current), ip);
553 print_vma_addr(KERN_CONT " in ", ip);
554 pr_cont("\n");
555 t->iopl_warn = 1;
556 }
557
558 regs->ip += 1;
559 return true;
560}
561
562/*
563 * The unprivileged ENQCMD instruction generates #GPs if the
564 * IA32_PASID MSR has not been populated. If possible, populate
565 * the MSR from a PASID previously allocated to the mm.
566 */
567static bool try_fixup_enqcmd_gp(void)
568{
569#ifdef CONFIG_ARCH_HAS_CPU_PASID
570 u32 pasid;
571
572 /*
573 * MSR_IA32_PASID is managed using XSAVE. Directly
574 * writing to the MSR is only possible when fpregs
575 * are valid and the fpstate is not. This is
576 * guaranteed when handling a userspace exception
577 * in *before* interrupts are re-enabled.
578 */
579 lockdep_assert_irqs_disabled();
580
581 /*
582 * Hardware without ENQCMD will not generate
583 * #GPs that can be fixed up here.
584 */
585 if (!cpu_feature_enabled(X86_FEATURE_ENQCMD))
586 return false;
587
588 /*
589 * If the mm has not been allocated a
590 * PASID, the #GP can not be fixed up.
591 */
592 if (!mm_valid_pasid(current->mm))
593 return false;
594
595 pasid = mm_get_enqcmd_pasid(current->mm);
596
597 /*
598 * Did this thread already have its PASID activated?
599 * If so, the #GP must be from something else.
600 */
601 if (current->pasid_activated)
602 return false;
603
604 wrmsrl(MSR_IA32_PASID, pasid | MSR_IA32_PASID_VALID);
605 current->pasid_activated = 1;
606
607 return true;
608#else
609 return false;
610#endif
611}
612
613static bool gp_try_fixup_and_notify(struct pt_regs *regs, int trapnr,
614 unsigned long error_code, const char *str,
615 unsigned long address)
616{
617 if (fixup_exception(regs, trapnr, error_code, address))
618 return true;
619
620 current->thread.error_code = error_code;
621 current->thread.trap_nr = trapnr;
622
623 /*
624 * To be potentially processing a kprobe fault and to trust the result
625 * from kprobe_running(), we have to be non-preemptible.
626 */
627 if (!preemptible() && kprobe_running() &&
628 kprobe_fault_handler(regs, trapnr))
629 return true;
630
631 return notify_die(DIE_GPF, str, regs, error_code, trapnr, SIGSEGV) == NOTIFY_STOP;
632}
633
634static void gp_user_force_sig_segv(struct pt_regs *regs, int trapnr,
635 unsigned long error_code, const char *str)
636{
637 current->thread.error_code = error_code;
638 current->thread.trap_nr = trapnr;
639 show_signal(current, SIGSEGV, "", str, regs, error_code);
640 force_sig(SIGSEGV);
641}
642
643DEFINE_IDTENTRY_ERRORCODE(exc_general_protection)
644{
645 char desc[sizeof(GPFSTR) + 50 + 2*sizeof(unsigned long) + 1] = GPFSTR;
646 enum kernel_gp_hint hint = GP_NO_HINT;
647 unsigned long gp_addr;
648
649 if (user_mode(regs) && try_fixup_enqcmd_gp())
650 return;
651
652 cond_local_irq_enable(regs);
653
654 if (static_cpu_has(X86_FEATURE_UMIP)) {
655 if (user_mode(regs) && fixup_umip_exception(regs))
656 goto exit;
657 }
658
659 if (v8086_mode(regs)) {
660 local_irq_enable();
661 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
662 local_irq_disable();
663 return;
664 }
665
666 if (user_mode(regs)) {
667 if (fixup_iopl_exception(regs))
668 goto exit;
669
670 if (fixup_vdso_exception(regs, X86_TRAP_GP, error_code, 0))
671 goto exit;
672
673 gp_user_force_sig_segv(regs, X86_TRAP_GP, error_code, desc);
674 goto exit;
675 }
676
677 if (gp_try_fixup_and_notify(regs, X86_TRAP_GP, error_code, desc, 0))
678 goto exit;
679
680 if (error_code)
681 snprintf(desc, sizeof(desc), "segment-related " GPFSTR);
682 else
683 hint = get_kernel_gp_address(regs, &gp_addr);
684
685 if (hint != GP_NO_HINT)
686 snprintf(desc, sizeof(desc), GPFSTR ", %s 0x%lx",
687 (hint == GP_NON_CANONICAL) ? "probably for non-canonical address"
688 : "maybe for address",
689 gp_addr);
690
691 /*
692 * KASAN is interested only in the non-canonical case, clear it
693 * otherwise.
694 */
695 if (hint != GP_NON_CANONICAL)
696 gp_addr = 0;
697
698 die_addr(desc, regs, error_code, gp_addr);
699
700exit:
701 cond_local_irq_disable(regs);
702}
703
704static bool do_int3(struct pt_regs *regs)
705{
706 int res;
707
708#ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
709 if (kgdb_ll_trap(DIE_INT3, "int3", regs, 0, X86_TRAP_BP,
710 SIGTRAP) == NOTIFY_STOP)
711 return true;
712#endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
713
714#ifdef CONFIG_KPROBES
715 if (kprobe_int3_handler(regs))
716 return true;
717#endif
718 res = notify_die(DIE_INT3, "int3", regs, 0, X86_TRAP_BP, SIGTRAP);
719
720 return res == NOTIFY_STOP;
721}
722NOKPROBE_SYMBOL(do_int3);
723
724static void do_int3_user(struct pt_regs *regs)
725{
726 if (do_int3(regs))
727 return;
728
729 cond_local_irq_enable(regs);
730 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, 0, 0, NULL);
731 cond_local_irq_disable(regs);
732}
733
734DEFINE_IDTENTRY_RAW(exc_int3)
735{
736 /*
737 * poke_int3_handler() is completely self contained code; it does (and
738 * must) *NOT* call out to anything, lest it hits upon yet another
739 * INT3.
740 */
741 if (poke_int3_handler(regs))
742 return;
743
744 /*
745 * irqentry_enter_from_user_mode() uses static_branch_{,un}likely()
746 * and therefore can trigger INT3, hence poke_int3_handler() must
747 * be done before. If the entry came from kernel mode, then use
748 * nmi_enter() because the INT3 could have been hit in any context
749 * including NMI.
750 */
751 if (user_mode(regs)) {
752 irqentry_enter_from_user_mode(regs);
753 instrumentation_begin();
754 do_int3_user(regs);
755 instrumentation_end();
756 irqentry_exit_to_user_mode(regs);
757 } else {
758 irqentry_state_t irq_state = irqentry_nmi_enter(regs);
759
760 instrumentation_begin();
761 if (!do_int3(regs))
762 die("int3", regs, 0);
763 instrumentation_end();
764 irqentry_nmi_exit(regs, irq_state);
765 }
766}
767
768#ifdef CONFIG_X86_64
769/*
770 * Help handler running on a per-cpu (IST or entry trampoline) stack
771 * to switch to the normal thread stack if the interrupted code was in
772 * user mode. The actual stack switch is done in entry_64.S
773 */
774asmlinkage __visible noinstr struct pt_regs *sync_regs(struct pt_regs *eregs)
775{
776 struct pt_regs *regs = (struct pt_regs *)this_cpu_read(pcpu_hot.top_of_stack) - 1;
777 if (regs != eregs)
778 *regs = *eregs;
779 return regs;
780}
781
782#ifdef CONFIG_AMD_MEM_ENCRYPT
783asmlinkage __visible noinstr struct pt_regs *vc_switch_off_ist(struct pt_regs *regs)
784{
785 unsigned long sp, *stack;
786 struct stack_info info;
787 struct pt_regs *regs_ret;
788
789 /*
790 * In the SYSCALL entry path the RSP value comes from user-space - don't
791 * trust it and switch to the current kernel stack
792 */
793 if (ip_within_syscall_gap(regs)) {
794 sp = this_cpu_read(pcpu_hot.top_of_stack);
795 goto sync;
796 }
797
798 /*
799 * From here on the RSP value is trusted. Now check whether entry
800 * happened from a safe stack. Not safe are the entry or unknown stacks,
801 * use the fall-back stack instead in this case.
802 */
803 sp = regs->sp;
804 stack = (unsigned long *)sp;
805
806 if (!get_stack_info_noinstr(stack, current, &info) || info.type == STACK_TYPE_ENTRY ||
807 info.type > STACK_TYPE_EXCEPTION_LAST)
808 sp = __this_cpu_ist_top_va(VC2);
809
810sync:
811 /*
812 * Found a safe stack - switch to it as if the entry didn't happen via
813 * IST stack. The code below only copies pt_regs, the real switch happens
814 * in assembly code.
815 */
816 sp = ALIGN_DOWN(sp, 8) - sizeof(*regs_ret);
817
818 regs_ret = (struct pt_regs *)sp;
819 *regs_ret = *regs;
820
821 return regs_ret;
822}
823#endif
824
825asmlinkage __visible noinstr struct pt_regs *fixup_bad_iret(struct pt_regs *bad_regs)
826{
827 struct pt_regs tmp, *new_stack;
828
829 /*
830 * This is called from entry_64.S early in handling a fault
831 * caused by a bad iret to user mode. To handle the fault
832 * correctly, we want to move our stack frame to where it would
833 * be had we entered directly on the entry stack (rather than
834 * just below the IRET frame) and we want to pretend that the
835 * exception came from the IRET target.
836 */
837 new_stack = (struct pt_regs *)__this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
838
839 /* Copy the IRET target to the temporary storage. */
840 __memcpy(&tmp.ip, (void *)bad_regs->sp, 5*8);
841
842 /* Copy the remainder of the stack from the current stack. */
843 __memcpy(&tmp, bad_regs, offsetof(struct pt_regs, ip));
844
845 /* Update the entry stack */
846 __memcpy(new_stack, &tmp, sizeof(tmp));
847
848 BUG_ON(!user_mode(new_stack));
849 return new_stack;
850}
851#endif
852
853static bool is_sysenter_singlestep(struct pt_regs *regs)
854{
855 /*
856 * We don't try for precision here. If we're anywhere in the region of
857 * code that can be single-stepped in the SYSENTER entry path, then
858 * assume that this is a useless single-step trap due to SYSENTER
859 * being invoked with TF set. (We don't know in advance exactly
860 * which instructions will be hit because BTF could plausibly
861 * be set.)
862 */
863#ifdef CONFIG_X86_32
864 return (regs->ip - (unsigned long)__begin_SYSENTER_singlestep_region) <
865 (unsigned long)__end_SYSENTER_singlestep_region -
866 (unsigned long)__begin_SYSENTER_singlestep_region;
867#elif defined(CONFIG_IA32_EMULATION)
868 return (regs->ip - (unsigned long)entry_SYSENTER_compat) <
869 (unsigned long)__end_entry_SYSENTER_compat -
870 (unsigned long)entry_SYSENTER_compat;
871#else
872 return false;
873#endif
874}
875
876static __always_inline unsigned long debug_read_clear_dr6(void)
877{
878 unsigned long dr6;
879
880 /*
881 * The Intel SDM says:
882 *
883 * Certain debug exceptions may clear bits 0-3. The remaining
884 * contents of the DR6 register are never cleared by the
885 * processor. To avoid confusion in identifying debug
886 * exceptions, debug handlers should clear the register before
887 * returning to the interrupted task.
888 *
889 * Keep it simple: clear DR6 immediately.
890 */
891 get_debugreg(dr6, 6);
892 set_debugreg(DR6_RESERVED, 6);
893 dr6 ^= DR6_RESERVED; /* Flip to positive polarity */
894
895 return dr6;
896}
897
898/*
899 * Our handling of the processor debug registers is non-trivial.
900 * We do not clear them on entry and exit from the kernel. Therefore
901 * it is possible to get a watchpoint trap here from inside the kernel.
902 * However, the code in ./ptrace.c has ensured that the user can
903 * only set watchpoints on userspace addresses. Therefore the in-kernel
904 * watchpoint trap can only occur in code which is reading/writing
905 * from user space. Such code must not hold kernel locks (since it
906 * can equally take a page fault), therefore it is safe to call
907 * force_sig_info even though that claims and releases locks.
908 *
909 * Code in ./signal.c ensures that the debug control register
910 * is restored before we deliver any signal, and therefore that
911 * user code runs with the correct debug control register even though
912 * we clear it here.
913 *
914 * Being careful here means that we don't have to be as careful in a
915 * lot of more complicated places (task switching can be a bit lazy
916 * about restoring all the debug state, and ptrace doesn't have to
917 * find every occurrence of the TF bit that could be saved away even
918 * by user code)
919 *
920 * May run on IST stack.
921 */
922
923static bool notify_debug(struct pt_regs *regs, unsigned long *dr6)
924{
925 /*
926 * Notifiers will clear bits in @dr6 to indicate the event has been
927 * consumed - hw_breakpoint_handler(), single_stop_cont().
928 *
929 * Notifiers will set bits in @virtual_dr6 to indicate the desire
930 * for signals - ptrace_triggered(), kgdb_hw_overflow_handler().
931 */
932 if (notify_die(DIE_DEBUG, "debug", regs, (long)dr6, 0, SIGTRAP) == NOTIFY_STOP)
933 return true;
934
935 return false;
936}
937
938static __always_inline void exc_debug_kernel(struct pt_regs *regs,
939 unsigned long dr6)
940{
941 /*
942 * Disable breakpoints during exception handling; recursive exceptions
943 * are exceedingly 'fun'.
944 *
945 * Since this function is NOKPROBE, and that also applies to
946 * HW_BREAKPOINT_X, we can't hit a breakpoint before this (XXX except a
947 * HW_BREAKPOINT_W on our stack)
948 *
949 * Entry text is excluded for HW_BP_X and cpu_entry_area, which
950 * includes the entry stack is excluded for everything.
951 */
952 unsigned long dr7 = local_db_save();
953 irqentry_state_t irq_state = irqentry_nmi_enter(regs);
954 instrumentation_begin();
955
956 /*
957 * If something gets miswired and we end up here for a user mode
958 * #DB, we will malfunction.
959 */
960 WARN_ON_ONCE(user_mode(regs));
961
962 if (test_thread_flag(TIF_BLOCKSTEP)) {
963 /*
964 * The SDM says "The processor clears the BTF flag when it
965 * generates a debug exception." but PTRACE_BLOCKSTEP requested
966 * it for userspace, but we just took a kernel #DB, so re-set
967 * BTF.
968 */
969 unsigned long debugctl;
970
971 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
972 debugctl |= DEBUGCTLMSR_BTF;
973 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
974 }
975
976 /*
977 * Catch SYSENTER with TF set and clear DR_STEP. If this hit a
978 * watchpoint at the same time then that will still be handled.
979 */
980 if ((dr6 & DR_STEP) && is_sysenter_singlestep(regs))
981 dr6 &= ~DR_STEP;
982
983 /*
984 * The kernel doesn't use INT1
985 */
986 if (!dr6)
987 goto out;
988
989 if (notify_debug(regs, &dr6))
990 goto out;
991
992 /*
993 * The kernel doesn't use TF single-step outside of:
994 *
995 * - Kprobes, consumed through kprobe_debug_handler()
996 * - KGDB, consumed through notify_debug()
997 *
998 * So if we get here with DR_STEP set, something is wonky.
999 *
1000 * A known way to trigger this is through QEMU's GDB stub,
1001 * which leaks #DB into the guest and causes IST recursion.
1002 */
1003 if (WARN_ON_ONCE(dr6 & DR_STEP))
1004 regs->flags &= ~X86_EFLAGS_TF;
1005out:
1006 instrumentation_end();
1007 irqentry_nmi_exit(regs, irq_state);
1008
1009 local_db_restore(dr7);
1010}
1011
1012static __always_inline void exc_debug_user(struct pt_regs *regs,
1013 unsigned long dr6)
1014{
1015 bool icebp;
1016
1017 /*
1018 * If something gets miswired and we end up here for a kernel mode
1019 * #DB, we will malfunction.
1020 */
1021 WARN_ON_ONCE(!user_mode(regs));
1022
1023 /*
1024 * NB: We can't easily clear DR7 here because
1025 * irqentry_exit_to_usermode() can invoke ptrace, schedule, access
1026 * user memory, etc. This means that a recursive #DB is possible. If
1027 * this happens, that #DB will hit exc_debug_kernel() and clear DR7.
1028 * Since we're not on the IST stack right now, everything will be
1029 * fine.
1030 */
1031
1032 irqentry_enter_from_user_mode(regs);
1033 instrumentation_begin();
1034
1035 /*
1036 * Start the virtual/ptrace DR6 value with just the DR_STEP mask
1037 * of the real DR6. ptrace_triggered() will set the DR_TRAPn bits.
1038 *
1039 * Userspace expects DR_STEP to be visible in ptrace_get_debugreg(6)
1040 * even if it is not the result of PTRACE_SINGLESTEP.
1041 */
1042 current->thread.virtual_dr6 = (dr6 & DR_STEP);
1043
1044 /*
1045 * The SDM says "The processor clears the BTF flag when it
1046 * generates a debug exception." Clear TIF_BLOCKSTEP to keep
1047 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
1048 */
1049 clear_thread_flag(TIF_BLOCKSTEP);
1050
1051 /*
1052 * If dr6 has no reason to give us about the origin of this trap,
1053 * then it's very likely the result of an icebp/int01 trap.
1054 * User wants a sigtrap for that.
1055 */
1056 icebp = !dr6;
1057
1058 if (notify_debug(regs, &dr6))
1059 goto out;
1060
1061 /* It's safe to allow irq's after DR6 has been saved */
1062 local_irq_enable();
1063
1064 if (v8086_mode(regs)) {
1065 handle_vm86_trap((struct kernel_vm86_regs *)regs, 0, X86_TRAP_DB);
1066 goto out_irq;
1067 }
1068
1069 /* #DB for bus lock can only be triggered from userspace. */
1070 if (dr6 & DR_BUS_LOCK)
1071 handle_bus_lock(regs);
1072
1073 /* Add the virtual_dr6 bits for signals. */
1074 dr6 |= current->thread.virtual_dr6;
1075 if (dr6 & (DR_STEP | DR_TRAP_BITS) || icebp)
1076 send_sigtrap(regs, 0, get_si_code(dr6));
1077
1078out_irq:
1079 local_irq_disable();
1080out:
1081 instrumentation_end();
1082 irqentry_exit_to_user_mode(regs);
1083}
1084
1085#ifdef CONFIG_X86_64
1086/* IST stack entry */
1087DEFINE_IDTENTRY_DEBUG(exc_debug)
1088{
1089 exc_debug_kernel(regs, debug_read_clear_dr6());
1090}
1091
1092/* User entry, runs on regular task stack */
1093DEFINE_IDTENTRY_DEBUG_USER(exc_debug)
1094{
1095 exc_debug_user(regs, debug_read_clear_dr6());
1096}
1097#else
1098/* 32 bit does not have separate entry points. */
1099DEFINE_IDTENTRY_RAW(exc_debug)
1100{
1101 unsigned long dr6 = debug_read_clear_dr6();
1102
1103 if (user_mode(regs))
1104 exc_debug_user(regs, dr6);
1105 else
1106 exc_debug_kernel(regs, dr6);
1107}
1108#endif
1109
1110/*
1111 * Note that we play around with the 'TS' bit in an attempt to get
1112 * the correct behaviour even in the presence of the asynchronous
1113 * IRQ13 behaviour
1114 */
1115static void math_error(struct pt_regs *regs, int trapnr)
1116{
1117 struct task_struct *task = current;
1118 struct fpu *fpu = &task->thread.fpu;
1119 int si_code;
1120 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
1121 "simd exception";
1122
1123 cond_local_irq_enable(regs);
1124
1125 if (!user_mode(regs)) {
1126 if (fixup_exception(regs, trapnr, 0, 0))
1127 goto exit;
1128
1129 task->thread.error_code = 0;
1130 task->thread.trap_nr = trapnr;
1131
1132 if (notify_die(DIE_TRAP, str, regs, 0, trapnr,
1133 SIGFPE) != NOTIFY_STOP)
1134 die(str, regs, 0);
1135 goto exit;
1136 }
1137
1138 /*
1139 * Synchronize the FPU register state to the memory register state
1140 * if necessary. This allows the exception handler to inspect it.
1141 */
1142 fpu_sync_fpstate(fpu);
1143
1144 task->thread.trap_nr = trapnr;
1145 task->thread.error_code = 0;
1146
1147 si_code = fpu__exception_code(fpu, trapnr);
1148 /* Retry when we get spurious exceptions: */
1149 if (!si_code)
1150 goto exit;
1151
1152 if (fixup_vdso_exception(regs, trapnr, 0, 0))
1153 goto exit;
1154
1155 force_sig_fault(SIGFPE, si_code,
1156 (void __user *)uprobe_get_trap_addr(regs));
1157exit:
1158 cond_local_irq_disable(regs);
1159}
1160
1161DEFINE_IDTENTRY(exc_coprocessor_error)
1162{
1163 math_error(regs, X86_TRAP_MF);
1164}
1165
1166DEFINE_IDTENTRY(exc_simd_coprocessor_error)
1167{
1168 if (IS_ENABLED(CONFIG_X86_INVD_BUG)) {
1169 /* AMD 486 bug: INVD in CPL 0 raises #XF instead of #GP */
1170 if (!static_cpu_has(X86_FEATURE_XMM)) {
1171 __exc_general_protection(regs, 0);
1172 return;
1173 }
1174 }
1175 math_error(regs, X86_TRAP_XF);
1176}
1177
1178DEFINE_IDTENTRY(exc_spurious_interrupt_bug)
1179{
1180 /*
1181 * This addresses a Pentium Pro Erratum:
1182 *
1183 * PROBLEM: If the APIC subsystem is configured in mixed mode with
1184 * Virtual Wire mode implemented through the local APIC, an
1185 * interrupt vector of 0Fh (Intel reserved encoding) may be
1186 * generated by the local APIC (Int 15). This vector may be
1187 * generated upon receipt of a spurious interrupt (an interrupt
1188 * which is removed before the system receives the INTA sequence)
1189 * instead of the programmed 8259 spurious interrupt vector.
1190 *
1191 * IMPLICATION: The spurious interrupt vector programmed in the
1192 * 8259 is normally handled by an operating system's spurious
1193 * interrupt handler. However, a vector of 0Fh is unknown to some
1194 * operating systems, which would crash if this erratum occurred.
1195 *
1196 * In theory this could be limited to 32bit, but the handler is not
1197 * hurting and who knows which other CPUs suffer from this.
1198 */
1199}
1200
1201static bool handle_xfd_event(struct pt_regs *regs)
1202{
1203 u64 xfd_err;
1204 int err;
1205
1206 if (!IS_ENABLED(CONFIG_X86_64) || !cpu_feature_enabled(X86_FEATURE_XFD))
1207 return false;
1208
1209 rdmsrl(MSR_IA32_XFD_ERR, xfd_err);
1210 if (!xfd_err)
1211 return false;
1212
1213 wrmsrl(MSR_IA32_XFD_ERR, 0);
1214
1215 /* Die if that happens in kernel space */
1216 if (WARN_ON(!user_mode(regs)))
1217 return false;
1218
1219 local_irq_enable();
1220
1221 err = xfd_enable_feature(xfd_err);
1222
1223 switch (err) {
1224 case -EPERM:
1225 force_sig_fault(SIGILL, ILL_ILLOPC, error_get_trap_addr(regs));
1226 break;
1227 case -EFAULT:
1228 force_sig(SIGSEGV);
1229 break;
1230 }
1231
1232 local_irq_disable();
1233 return true;
1234}
1235
1236DEFINE_IDTENTRY(exc_device_not_available)
1237{
1238 unsigned long cr0 = read_cr0();
1239
1240 if (handle_xfd_event(regs))
1241 return;
1242
1243#ifdef CONFIG_MATH_EMULATION
1244 if (!boot_cpu_has(X86_FEATURE_FPU) && (cr0 & X86_CR0_EM)) {
1245 struct math_emu_info info = { };
1246
1247 cond_local_irq_enable(regs);
1248
1249 info.regs = regs;
1250 math_emulate(&info);
1251
1252 cond_local_irq_disable(regs);
1253 return;
1254 }
1255#endif
1256
1257 /* This should not happen. */
1258 if (WARN(cr0 & X86_CR0_TS, "CR0.TS was set")) {
1259 /* Try to fix it up and carry on. */
1260 write_cr0(cr0 & ~X86_CR0_TS);
1261 } else {
1262 /*
1263 * Something terrible happened, and we're better off trying
1264 * to kill the task than getting stuck in a never-ending
1265 * loop of #NM faults.
1266 */
1267 die("unexpected #NM exception", regs, 0);
1268 }
1269}
1270
1271#ifdef CONFIG_INTEL_TDX_GUEST
1272
1273#define VE_FAULT_STR "VE fault"
1274
1275static void ve_raise_fault(struct pt_regs *regs, long error_code,
1276 unsigned long address)
1277{
1278 if (user_mode(regs)) {
1279 gp_user_force_sig_segv(regs, X86_TRAP_VE, error_code, VE_FAULT_STR);
1280 return;
1281 }
1282
1283 if (gp_try_fixup_and_notify(regs, X86_TRAP_VE, error_code,
1284 VE_FAULT_STR, address)) {
1285 return;
1286 }
1287
1288 die_addr(VE_FAULT_STR, regs, error_code, address);
1289}
1290
1291/*
1292 * Virtualization Exceptions (#VE) are delivered to TDX guests due to
1293 * specific guest actions which may happen in either user space or the
1294 * kernel:
1295 *
1296 * * Specific instructions (WBINVD, for example)
1297 * * Specific MSR accesses
1298 * * Specific CPUID leaf accesses
1299 * * Access to specific guest physical addresses
1300 *
1301 * In the settings that Linux will run in, virtualization exceptions are
1302 * never generated on accesses to normal, TD-private memory that has been
1303 * accepted (by BIOS or with tdx_enc_status_changed()).
1304 *
1305 * Syscall entry code has a critical window where the kernel stack is not
1306 * yet set up. Any exception in this window leads to hard to debug issues
1307 * and can be exploited for privilege escalation. Exceptions in the NMI
1308 * entry code also cause issues. Returning from the exception handler with
1309 * IRET will re-enable NMIs and nested NMI will corrupt the NMI stack.
1310 *
1311 * For these reasons, the kernel avoids #VEs during the syscall gap and
1312 * the NMI entry code. Entry code paths do not access TD-shared memory,
1313 * MMIO regions, use #VE triggering MSRs, instructions, or CPUID leaves
1314 * that might generate #VE. VMM can remove memory from TD at any point,
1315 * but access to unaccepted (or missing) private memory leads to VM
1316 * termination, not to #VE.
1317 *
1318 * Similarly to page faults and breakpoints, #VEs are allowed in NMI
1319 * handlers once the kernel is ready to deal with nested NMIs.
1320 *
1321 * During #VE delivery, all interrupts, including NMIs, are blocked until
1322 * TDGETVEINFO is called. It prevents #VE nesting until the kernel reads
1323 * the VE info.
1324 *
1325 * If a guest kernel action which would normally cause a #VE occurs in
1326 * the interrupt-disabled region before TDGETVEINFO, a #DF (fault
1327 * exception) is delivered to the guest which will result in an oops.
1328 *
1329 * The entry code has been audited carefully for following these expectations.
1330 * Changes in the entry code have to be audited for correctness vs. this
1331 * aspect. Similarly to #PF, #VE in these places will expose kernel to
1332 * privilege escalation or may lead to random crashes.
1333 */
1334DEFINE_IDTENTRY(exc_virtualization_exception)
1335{
1336 struct ve_info ve;
1337
1338 /*
1339 * NMIs/Machine-checks/Interrupts will be in a disabled state
1340 * till TDGETVEINFO TDCALL is executed. This ensures that VE
1341 * info cannot be overwritten by a nested #VE.
1342 */
1343 tdx_get_ve_info(&ve);
1344
1345 cond_local_irq_enable(regs);
1346
1347 /*
1348 * If tdx_handle_virt_exception() could not process
1349 * it successfully, treat it as #GP(0) and handle it.
1350 */
1351 if (!tdx_handle_virt_exception(regs, &ve))
1352 ve_raise_fault(regs, 0, ve.gla);
1353
1354 cond_local_irq_disable(regs);
1355}
1356
1357#endif
1358
1359#ifdef CONFIG_X86_32
1360DEFINE_IDTENTRY_SW(iret_error)
1361{
1362 local_irq_enable();
1363 if (notify_die(DIE_TRAP, "iret exception", regs, 0,
1364 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
1365 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, 0,
1366 ILL_BADSTK, (void __user *)NULL);
1367 }
1368 local_irq_disable();
1369}
1370#endif
1371
1372void __init trap_init(void)
1373{
1374 /* Init cpu_entry_area before IST entries are set up */
1375 setup_cpu_entry_areas();
1376
1377 /* Init GHCB memory pages when running as an SEV-ES guest */
1378 sev_es_init_vc_handling();
1379
1380 /* Initialize TSS before setting up traps so ISTs work */
1381 cpu_init_exception_handling();
1382 /* Setup traps as cpu_init() might #GP */
1383 idt_setup_traps();
1384 cpu_init();
1385}