2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * Handle hardware traps and faults.
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
38 #include <linux/smp.h>
40 #include <linux/hardirq.h>
41 #include <linux/atomic.h>
43 #include <asm/stacktrace.h>
44 #include <asm/processor.h>
45 #include <asm/debugreg.h>
46 #include <asm/realmode.h>
47 #include <asm/text-patching.h>
48 #include <asm/ftrace.h>
49 #include <asm/traps.h>
51 #include <asm/fpu/internal.h>
53 #include <asm/cpu_entry_area.h>
55 #include <asm/fixmap.h>
56 #include <asm/mach_traps.h>
57 #include <asm/alternative.h>
58 #include <asm/fpu/xstate.h>
62 #include <asm/insn-eval.h>
65 #include <asm/x86_init.h>
66 #include <asm/proto.h>
68 #include <asm/processor-flags.h>
69 #include <asm/setup.h>
70 #include <asm/proto.h>
73 DECLARE_BITMAP(system_vectors, NR_VECTORS);
75 static inline void cond_local_irq_enable(struct pt_regs *regs)
77 if (regs->flags & X86_EFLAGS_IF)
81 static inline void cond_local_irq_disable(struct pt_regs *regs)
83 if (regs->flags & X86_EFLAGS_IF)
87 __always_inline int is_valid_bugaddr(unsigned long addr)
89 if (addr < TASK_SIZE_MAX)
93 * We got #UD, if the text isn't readable we'd have gotten
94 * a different exception.
96 return *(unsigned short *)addr == INSN_UD2;
99 static nokprobe_inline int
100 do_trap_no_signal(struct task_struct *tsk, int trapnr, const char *str,
101 struct pt_regs *regs, long error_code)
103 if (v8086_mode(regs)) {
105 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
106 * On nmi (interrupt 2), do_trap should not be called.
108 if (trapnr < X86_TRAP_UD) {
109 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
113 } else if (!user_mode(regs)) {
114 if (fixup_exception(regs, trapnr, error_code, 0))
117 tsk->thread.error_code = error_code;
118 tsk->thread.trap_nr = trapnr;
119 die(str, regs, error_code);
123 * We want error_code and trap_nr set for userspace faults and
124 * kernelspace faults which result in die(), but not
125 * kernelspace faults which are fixed up. die() gives the
126 * process no chance to handle the signal and notice the
127 * kernel fault information, so that won't result in polluting
128 * the information about previously queued, but not yet
129 * delivered, faults. See also exc_general_protection below.
131 tsk->thread.error_code = error_code;
132 tsk->thread.trap_nr = trapnr;
137 static void show_signal(struct task_struct *tsk, int signr,
138 const char *type, const char *desc,
139 struct pt_regs *regs, long error_code)
141 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
142 printk_ratelimit()) {
143 pr_info("%s[%d] %s%s ip:%lx sp:%lx error:%lx",
144 tsk->comm, task_pid_nr(tsk), type, desc,
145 regs->ip, regs->sp, error_code);
146 print_vma_addr(KERN_CONT " in ", regs->ip);
152 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
153 long error_code, int sicode, void __user *addr)
155 struct task_struct *tsk = current;
157 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
160 show_signal(tsk, signr, "trap ", str, regs, error_code);
165 force_sig_fault(signr, sicode, addr);
167 NOKPROBE_SYMBOL(do_trap);
169 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
170 unsigned long trapnr, int signr, int sicode, void __user *addr)
172 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
174 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
176 cond_local_irq_enable(regs);
177 do_trap(trapnr, signr, str, regs, error_code, sicode, addr);
178 cond_local_irq_disable(regs);
183 * Posix requires to provide the address of the faulting instruction for
184 * SIGILL (#UD) and SIGFPE (#DE) in the si_addr member of siginfo_t.
186 * This address is usually regs->ip, but when an uprobe moved the code out
187 * of line then regs->ip points to the XOL code which would confuse
188 * anything which analyzes the fault address vs. the unmodified binary. If
189 * a trap happened in XOL code then uprobe maps regs->ip back to the
190 * original instruction address.
192 static __always_inline void __user *error_get_trap_addr(struct pt_regs *regs)
194 return (void __user *)uprobe_get_trap_addr(regs);
197 DEFINE_IDTENTRY(exc_divide_error)
199 do_error_trap(regs, 0, "divide error", X86_TRAP_DE, SIGFPE,
200 FPE_INTDIV, error_get_trap_addr(regs));
203 DEFINE_IDTENTRY(exc_overflow)
205 do_error_trap(regs, 0, "overflow", X86_TRAP_OF, SIGSEGV, 0, NULL);
208 #ifdef CONFIG_X86_F00F_BUG
209 void handle_invalid_op(struct pt_regs *regs)
211 static inline void handle_invalid_op(struct pt_regs *regs)
214 do_error_trap(regs, 0, "invalid opcode", X86_TRAP_UD, SIGILL,
215 ILL_ILLOPN, error_get_trap_addr(regs));
218 static noinstr bool handle_bug(struct pt_regs *regs)
220 bool handled = false;
222 if (!is_valid_bugaddr(regs->ip))
226 * All lies, just get the WARN/BUG out.
228 instrumentation_begin();
230 * Since we're emulating a CALL with exceptions, restore the interrupt
231 * state to what it was at the exception site.
233 if (regs->flags & X86_EFLAGS_IF)
234 raw_local_irq_enable();
235 if (report_bug(regs->ip, regs) == BUG_TRAP_TYPE_WARN) {
239 if (regs->flags & X86_EFLAGS_IF)
240 raw_local_irq_disable();
241 instrumentation_end();
246 DEFINE_IDTENTRY_RAW(exc_invalid_op)
248 irqentry_state_t state;
251 * We use UD2 as a short encoding for 'CALL __WARN', as such
252 * handle it before exception entry to avoid recursive WARN
253 * in case exception entry is the one triggering WARNs.
255 if (!user_mode(regs) && handle_bug(regs))
258 state = irqentry_enter(regs);
259 instrumentation_begin();
260 handle_invalid_op(regs);
261 instrumentation_end();
262 irqentry_exit(regs, state);
265 DEFINE_IDTENTRY(exc_coproc_segment_overrun)
267 do_error_trap(regs, 0, "coprocessor segment overrun",
268 X86_TRAP_OLD_MF, SIGFPE, 0, NULL);
271 DEFINE_IDTENTRY_ERRORCODE(exc_invalid_tss)
273 do_error_trap(regs, error_code, "invalid TSS", X86_TRAP_TS, SIGSEGV,
277 DEFINE_IDTENTRY_ERRORCODE(exc_segment_not_present)
279 do_error_trap(regs, error_code, "segment not present", X86_TRAP_NP,
283 DEFINE_IDTENTRY_ERRORCODE(exc_stack_segment)
285 do_error_trap(regs, error_code, "stack segment", X86_TRAP_SS, SIGBUS,
289 DEFINE_IDTENTRY_ERRORCODE(exc_alignment_check)
291 char *str = "alignment check";
293 if (notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_AC, SIGBUS) == NOTIFY_STOP)
296 if (!user_mode(regs))
297 die("Split lock detected\n", regs, error_code);
301 if (handle_user_split_lock(regs, error_code))
304 do_trap(X86_TRAP_AC, SIGBUS, "alignment check", regs,
305 error_code, BUS_ADRALN, NULL);
311 #ifdef CONFIG_VMAP_STACK
312 __visible void __noreturn handle_stack_overflow(const char *message,
313 struct pt_regs *regs,
314 unsigned long fault_address)
316 printk(KERN_EMERG "BUG: stack guard page was hit at %p (stack is %p..%p)\n",
317 (void *)fault_address, current->stack,
318 (char *)current->stack + THREAD_SIZE - 1);
319 die(message, regs, 0);
321 /* Be absolutely certain we don't return. */
322 panic("%s", message);
327 * Runs on an IST stack for x86_64 and on a special task stack for x86_32.
329 * On x86_64, this is more or less a normal kernel entry. Notwithstanding the
330 * SDM's warnings about double faults being unrecoverable, returning works as
331 * expected. Presumably what the SDM actually means is that the CPU may get
332 * the register state wrong on entry, so returning could be a bad idea.
334 * Various CPU engineers have promised that double faults due to an IRET fault
335 * while the stack is read-only are, in fact, recoverable.
337 * On x86_32, this is entered through a task gate, and regs are synthesized
338 * from the TSS. Returning is, in principle, okay, but changes to regs will
339 * be lost. If, for some reason, we need to return to a context with modified
340 * regs, the shim code could be adjusted to synchronize the registers.
342 * The 32bit #DF shim provides CR2 already as an argument. On 64bit it needs
343 * to be read before doing anything else.
345 DEFINE_IDTENTRY_DF(exc_double_fault)
347 static const char str[] = "double fault";
348 struct task_struct *tsk = current;
350 #ifdef CONFIG_VMAP_STACK
351 unsigned long address = read_cr2();
354 #ifdef CONFIG_X86_ESPFIX64
355 extern unsigned char native_irq_return_iret[];
358 * If IRET takes a non-IST fault on the espfix64 stack, then we
359 * end up promoting it to a doublefault. In that case, take
360 * advantage of the fact that we're not using the normal (TSS.sp0)
361 * stack right now. We can write a fake #GP(0) frame at TSS.sp0
362 * and then modify our own IRET frame so that, when we return,
363 * we land directly at the #GP(0) vector with the stack already
364 * set up according to its expectations.
366 * The net result is that our #GP handler will think that we
367 * entered from usermode with the bad user context.
369 * No need for nmi_enter() here because we don't use RCU.
371 if (((long)regs->sp >> P4D_SHIFT) == ESPFIX_PGD_ENTRY &&
372 regs->cs == __KERNEL_CS &&
373 regs->ip == (unsigned long)native_irq_return_iret)
375 struct pt_regs *gpregs = (struct pt_regs *)this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
376 unsigned long *p = (unsigned long *)regs->sp;
379 * regs->sp points to the failing IRET frame on the
380 * ESPFIX64 stack. Copy it to the entry stack. This fills
381 * in gpregs->ss through gpregs->ip.
386 gpregs->flags = p[2];
389 gpregs->orig_ax = 0; /* Missing (lost) #GP error code */
392 * Adjust our frame so that we return straight to the #GP
393 * vector with the expected RSP value. This is safe because
394 * we won't enable interupts or schedule before we invoke
395 * general_protection, so nothing will clobber the stack
396 * frame we just set up.
398 * We will enter general_protection with kernel GSBASE,
399 * which is what the stub expects, given that the faulting
400 * RIP will be the IRET instruction.
402 regs->ip = (unsigned long)asm_exc_general_protection;
403 regs->sp = (unsigned long)&gpregs->orig_ax;
409 irqentry_nmi_enter(regs);
410 instrumentation_begin();
411 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
413 tsk->thread.error_code = error_code;
414 tsk->thread.trap_nr = X86_TRAP_DF;
416 #ifdef CONFIG_VMAP_STACK
418 * If we overflow the stack into a guard page, the CPU will fail
419 * to deliver #PF and will send #DF instead. Similarly, if we
420 * take any non-IST exception while too close to the bottom of
421 * the stack, the processor will get a page fault while
422 * delivering the exception and will generate a double fault.
424 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
425 * Page-Fault Exception (#PF):
427 * Processors update CR2 whenever a page fault is detected. If a
428 * second page fault occurs while an earlier page fault is being
429 * delivered, the faulting linear address of the second fault will
430 * overwrite the contents of CR2 (replacing the previous
431 * address). These updates to CR2 occur even if the page fault
432 * results in a double fault or occurs during the delivery of a
435 * The logic below has a small possibility of incorrectly diagnosing
436 * some errors as stack overflows. For example, if the IDT or GDT
437 * gets corrupted such that #GP delivery fails due to a bad descriptor
438 * causing #GP and we hit this condition while CR2 coincidentally
439 * points to the stack guard page, we'll think we overflowed the
440 * stack. Given that we're going to panic one way or another
441 * if this happens, this isn't necessarily worth fixing.
443 * If necessary, we could improve the test by only diagnosing
444 * a stack overflow if the saved RSP points within 47 bytes of
445 * the bottom of the stack: if RSP == tsk_stack + 48 and we
446 * take an exception, the stack is already aligned and there
447 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
448 * possible error code, so a stack overflow would *not* double
449 * fault. With any less space left, exception delivery could
450 * fail, and, as a practical matter, we've overflowed the
451 * stack even if the actual trigger for the double fault was
454 if ((unsigned long)task_stack_page(tsk) - 1 - address < PAGE_SIZE) {
455 handle_stack_overflow("kernel stack overflow (double-fault)",
460 pr_emerg("PANIC: double fault, error_code: 0x%lx\n", error_code);
461 die("double fault", regs, error_code);
462 panic("Machine halted.");
463 instrumentation_end();
466 DEFINE_IDTENTRY(exc_bounds)
468 if (notify_die(DIE_TRAP, "bounds", regs, 0,
469 X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
471 cond_local_irq_enable(regs);
473 if (!user_mode(regs))
474 die("bounds", regs, 0);
476 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, 0, 0, NULL);
478 cond_local_irq_disable(regs);
481 enum kernel_gp_hint {
488 * When an uncaught #GP occurs, try to determine the memory address accessed by
489 * the instruction and return that address to the caller. Also, try to figure
490 * out whether any part of the access to that address was non-canonical.
492 static enum kernel_gp_hint get_kernel_gp_address(struct pt_regs *regs,
495 u8 insn_buf[MAX_INSN_SIZE];
498 if (copy_from_kernel_nofault(insn_buf, (void *)regs->ip,
502 kernel_insn_init(&insn, insn_buf, MAX_INSN_SIZE);
503 insn_get_modrm(&insn);
506 *addr = (unsigned long)insn_get_addr_ref(&insn, regs);
513 * - the operand is not in the kernel half
514 * - the last byte of the operand is not in the user canonical half
516 if (*addr < ~__VIRTUAL_MASK &&
517 *addr + insn.opnd_bytes - 1 > __VIRTUAL_MASK)
518 return GP_NON_CANONICAL;
524 #define GPFSTR "general protection fault"
526 static bool fixup_iopl_exception(struct pt_regs *regs)
528 struct thread_struct *t = ¤t->thread;
532 if (!IS_ENABLED(CONFIG_X86_IOPL_IOPERM) || t->iopl_emul != 3)
535 ip = insn_get_effective_ip(regs);
539 if (get_user(byte, (const char __user *)ip))
542 if (byte != 0xfa && byte != 0xfb)
545 if (!t->iopl_warn && printk_ratelimit()) {
546 pr_err("%s[%d] attempts to use CLI/STI, pretending it's a NOP, ip:%lx",
547 current->comm, task_pid_nr(current), ip);
548 print_vma_addr(KERN_CONT " in ", ip);
557 DEFINE_IDTENTRY_ERRORCODE(exc_general_protection)
559 char desc[sizeof(GPFSTR) + 50 + 2*sizeof(unsigned long) + 1] = GPFSTR;
560 enum kernel_gp_hint hint = GP_NO_HINT;
561 struct task_struct *tsk;
562 unsigned long gp_addr;
565 cond_local_irq_enable(regs);
567 if (static_cpu_has(X86_FEATURE_UMIP)) {
568 if (user_mode(regs) && fixup_umip_exception(regs))
572 if (v8086_mode(regs)) {
574 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
581 if (user_mode(regs)) {
582 if (fixup_iopl_exception(regs))
585 tsk->thread.error_code = error_code;
586 tsk->thread.trap_nr = X86_TRAP_GP;
588 show_signal(tsk, SIGSEGV, "", desc, regs, error_code);
593 if (fixup_exception(regs, X86_TRAP_GP, error_code, 0))
596 tsk->thread.error_code = error_code;
597 tsk->thread.trap_nr = X86_TRAP_GP;
600 * To be potentially processing a kprobe fault and to trust the result
601 * from kprobe_running(), we have to be non-preemptible.
603 if (!preemptible() &&
605 kprobe_fault_handler(regs, X86_TRAP_GP))
608 ret = notify_die(DIE_GPF, desc, regs, error_code, X86_TRAP_GP, SIGSEGV);
609 if (ret == NOTIFY_STOP)
613 snprintf(desc, sizeof(desc), "segment-related " GPFSTR);
615 hint = get_kernel_gp_address(regs, &gp_addr);
617 if (hint != GP_NO_HINT)
618 snprintf(desc, sizeof(desc), GPFSTR ", %s 0x%lx",
619 (hint == GP_NON_CANONICAL) ? "probably for non-canonical address"
620 : "maybe for address",
624 * KASAN is interested only in the non-canonical case, clear it
627 if (hint != GP_NON_CANONICAL)
630 die_addr(desc, regs, error_code, gp_addr);
633 cond_local_irq_disable(regs);
636 static bool do_int3(struct pt_regs *regs)
640 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
641 if (kgdb_ll_trap(DIE_INT3, "int3", regs, 0, X86_TRAP_BP,
642 SIGTRAP) == NOTIFY_STOP)
644 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
646 #ifdef CONFIG_KPROBES
647 if (kprobe_int3_handler(regs))
650 res = notify_die(DIE_INT3, "int3", regs, 0, X86_TRAP_BP, SIGTRAP);
652 return res == NOTIFY_STOP;
654 NOKPROBE_SYMBOL(do_int3);
656 static void do_int3_user(struct pt_regs *regs)
661 cond_local_irq_enable(regs);
662 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, 0, 0, NULL);
663 cond_local_irq_disable(regs);
666 DEFINE_IDTENTRY_RAW(exc_int3)
669 * poke_int3_handler() is completely self contained code; it does (and
670 * must) *NOT* call out to anything, lest it hits upon yet another
673 if (poke_int3_handler(regs))
677 * irqentry_enter_from_user_mode() uses static_branch_{,un}likely()
678 * and therefore can trigger INT3, hence poke_int3_handler() must
679 * be done before. If the entry came from kernel mode, then use
680 * nmi_enter() because the INT3 could have been hit in any context
683 if (user_mode(regs)) {
684 irqentry_enter_from_user_mode(regs);
685 instrumentation_begin();
687 instrumentation_end();
688 irqentry_exit_to_user_mode(regs);
690 irqentry_state_t irq_state = irqentry_nmi_enter(regs);
692 instrumentation_begin();
694 die("int3", regs, 0);
695 instrumentation_end();
696 irqentry_nmi_exit(regs, irq_state);
702 * Help handler running on a per-cpu (IST or entry trampoline) stack
703 * to switch to the normal thread stack if the interrupted code was in
704 * user mode. The actual stack switch is done in entry_64.S
706 asmlinkage __visible noinstr struct pt_regs *sync_regs(struct pt_regs *eregs)
708 struct pt_regs *regs = (struct pt_regs *)this_cpu_read(cpu_current_top_of_stack) - 1;
714 #ifdef CONFIG_AMD_MEM_ENCRYPT
715 asmlinkage __visible noinstr struct pt_regs *vc_switch_off_ist(struct pt_regs *regs)
717 unsigned long sp, *stack;
718 struct stack_info info;
719 struct pt_regs *regs_ret;
722 * In the SYSCALL entry path the RSP value comes from user-space - don't
723 * trust it and switch to the current kernel stack
725 if (ip_within_syscall_gap(regs)) {
726 sp = this_cpu_read(cpu_current_top_of_stack);
731 * From here on the RSP value is trusted. Now check whether entry
732 * happened from a safe stack. Not safe are the entry or unknown stacks,
733 * use the fall-back stack instead in this case.
736 stack = (unsigned long *)sp;
738 if (!get_stack_info_noinstr(stack, current, &info) || info.type == STACK_TYPE_ENTRY ||
739 info.type > STACK_TYPE_EXCEPTION_LAST)
740 sp = __this_cpu_ist_top_va(VC2);
744 * Found a safe stack - switch to it as if the entry didn't happen via
745 * IST stack. The code below only copies pt_regs, the real switch happens
748 sp = ALIGN_DOWN(sp, 8) - sizeof(*regs_ret);
750 regs_ret = (struct pt_regs *)sp;
757 struct bad_iret_stack {
758 void *error_entry_ret;
762 asmlinkage __visible noinstr
763 struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
766 * This is called from entry_64.S early in handling a fault
767 * caused by a bad iret to user mode. To handle the fault
768 * correctly, we want to move our stack frame to where it would
769 * be had we entered directly on the entry stack (rather than
770 * just below the IRET frame) and we want to pretend that the
771 * exception came from the IRET target.
773 struct bad_iret_stack tmp, *new_stack =
774 (struct bad_iret_stack *)__this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
776 /* Copy the IRET target to the temporary storage. */
777 __memcpy(&tmp.regs.ip, (void *)s->regs.sp, 5*8);
779 /* Copy the remainder of the stack from the current stack. */
780 __memcpy(&tmp, s, offsetof(struct bad_iret_stack, regs.ip));
782 /* Update the entry stack */
783 __memcpy(new_stack, &tmp, sizeof(tmp));
785 BUG_ON(!user_mode(&new_stack->regs));
790 static bool is_sysenter_singlestep(struct pt_regs *regs)
793 * We don't try for precision here. If we're anywhere in the region of
794 * code that can be single-stepped in the SYSENTER entry path, then
795 * assume that this is a useless single-step trap due to SYSENTER
796 * being invoked with TF set. (We don't know in advance exactly
797 * which instructions will be hit because BTF could plausibly
801 return (regs->ip - (unsigned long)__begin_SYSENTER_singlestep_region) <
802 (unsigned long)__end_SYSENTER_singlestep_region -
803 (unsigned long)__begin_SYSENTER_singlestep_region;
804 #elif defined(CONFIG_IA32_EMULATION)
805 return (regs->ip - (unsigned long)entry_SYSENTER_compat) <
806 (unsigned long)__end_entry_SYSENTER_compat -
807 (unsigned long)entry_SYSENTER_compat;
813 static __always_inline unsigned long debug_read_clear_dr6(void)
818 * The Intel SDM says:
820 * Certain debug exceptions may clear bits 0-3. The remaining
821 * contents of the DR6 register are never cleared by the
822 * processor. To avoid confusion in identifying debug
823 * exceptions, debug handlers should clear the register before
824 * returning to the interrupted task.
826 * Keep it simple: clear DR6 immediately.
828 get_debugreg(dr6, 6);
829 set_debugreg(DR6_RESERVED, 6);
830 dr6 ^= DR6_RESERVED; /* Flip to positive polarity */
836 * Our handling of the processor debug registers is non-trivial.
837 * We do not clear them on entry and exit from the kernel. Therefore
838 * it is possible to get a watchpoint trap here from inside the kernel.
839 * However, the code in ./ptrace.c has ensured that the user can
840 * only set watchpoints on userspace addresses. Therefore the in-kernel
841 * watchpoint trap can only occur in code which is reading/writing
842 * from user space. Such code must not hold kernel locks (since it
843 * can equally take a page fault), therefore it is safe to call
844 * force_sig_info even though that claims and releases locks.
846 * Code in ./signal.c ensures that the debug control register
847 * is restored before we deliver any signal, and therefore that
848 * user code runs with the correct debug control register even though
851 * Being careful here means that we don't have to be as careful in a
852 * lot of more complicated places (task switching can be a bit lazy
853 * about restoring all the debug state, and ptrace doesn't have to
854 * find every occurrence of the TF bit that could be saved away even
857 * May run on IST stack.
860 static bool notify_debug(struct pt_regs *regs, unsigned long *dr6)
863 * Notifiers will clear bits in @dr6 to indicate the event has been
864 * consumed - hw_breakpoint_handler(), single_stop_cont().
866 * Notifiers will set bits in @virtual_dr6 to indicate the desire
867 * for signals - ptrace_triggered(), kgdb_hw_overflow_handler().
869 if (notify_die(DIE_DEBUG, "debug", regs, (long)dr6, 0, SIGTRAP) == NOTIFY_STOP)
875 static __always_inline void exc_debug_kernel(struct pt_regs *regs,
879 * Disable breakpoints during exception handling; recursive exceptions
880 * are exceedingly 'fun'.
882 * Since this function is NOKPROBE, and that also applies to
883 * HW_BREAKPOINT_X, we can't hit a breakpoint before this (XXX except a
884 * HW_BREAKPOINT_W on our stack)
886 * Entry text is excluded for HW_BP_X and cpu_entry_area, which
887 * includes the entry stack is excluded for everything.
889 unsigned long dr7 = local_db_save();
890 irqentry_state_t irq_state = irqentry_nmi_enter(regs);
891 instrumentation_begin();
894 * If something gets miswired and we end up here for a user mode
895 * #DB, we will malfunction.
897 WARN_ON_ONCE(user_mode(regs));
899 if (test_thread_flag(TIF_BLOCKSTEP)) {
901 * The SDM says "The processor clears the BTF flag when it
902 * generates a debug exception." but PTRACE_BLOCKSTEP requested
903 * it for userspace, but we just took a kernel #DB, so re-set
906 unsigned long debugctl;
908 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
909 debugctl |= DEBUGCTLMSR_BTF;
910 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
914 * Catch SYSENTER with TF set and clear DR_STEP. If this hit a
915 * watchpoint at the same time then that will still be handled.
917 if ((dr6 & DR_STEP) && is_sysenter_singlestep(regs))
921 * The kernel doesn't use INT1
926 if (notify_debug(regs, &dr6))
930 * The kernel doesn't use TF single-step outside of:
932 * - Kprobes, consumed through kprobe_debug_handler()
933 * - KGDB, consumed through notify_debug()
935 * So if we get here with DR_STEP set, something is wonky.
937 * A known way to trigger this is through QEMU's GDB stub,
938 * which leaks #DB into the guest and causes IST recursion.
940 if (WARN_ON_ONCE(dr6 & DR_STEP))
941 regs->flags &= ~X86_EFLAGS_TF;
943 instrumentation_end();
944 irqentry_nmi_exit(regs, irq_state);
946 local_db_restore(dr7);
949 static __always_inline void exc_debug_user(struct pt_regs *regs,
955 * If something gets miswired and we end up here for a kernel mode
956 * #DB, we will malfunction.
958 WARN_ON_ONCE(!user_mode(regs));
961 * NB: We can't easily clear DR7 here because
962 * irqentry_exit_to_usermode() can invoke ptrace, schedule, access
963 * user memory, etc. This means that a recursive #DB is possible. If
964 * this happens, that #DB will hit exc_debug_kernel() and clear DR7.
965 * Since we're not on the IST stack right now, everything will be
969 irqentry_enter_from_user_mode(regs);
970 instrumentation_begin();
973 * Start the virtual/ptrace DR6 value with just the DR_STEP mask
974 * of the real DR6. ptrace_triggered() will set the DR_TRAPn bits.
976 * Userspace expects DR_STEP to be visible in ptrace_get_debugreg(6)
977 * even if it is not the result of PTRACE_SINGLESTEP.
979 current->thread.virtual_dr6 = (dr6 & DR_STEP);
982 * The SDM says "The processor clears the BTF flag when it
983 * generates a debug exception." Clear TIF_BLOCKSTEP to keep
984 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
986 clear_thread_flag(TIF_BLOCKSTEP);
989 * If dr6 has no reason to give us about the origin of this trap,
990 * then it's very likely the result of an icebp/int01 trap.
991 * User wants a sigtrap for that.
995 if (notify_debug(regs, &dr6))
998 /* It's safe to allow irq's after DR6 has been saved */
1001 if (v8086_mode(regs)) {
1002 handle_vm86_trap((struct kernel_vm86_regs *)regs, 0, X86_TRAP_DB);
1006 /* Add the virtual_dr6 bits for signals. */
1007 dr6 |= current->thread.virtual_dr6;
1008 if (dr6 & (DR_STEP | DR_TRAP_BITS) || icebp)
1009 send_sigtrap(regs, 0, get_si_code(dr6));
1012 local_irq_disable();
1014 instrumentation_end();
1015 irqentry_exit_to_user_mode(regs);
1018 #ifdef CONFIG_X86_64
1019 /* IST stack entry */
1020 DEFINE_IDTENTRY_DEBUG(exc_debug)
1022 exc_debug_kernel(regs, debug_read_clear_dr6());
1025 /* User entry, runs on regular task stack */
1026 DEFINE_IDTENTRY_DEBUG_USER(exc_debug)
1028 exc_debug_user(regs, debug_read_clear_dr6());
1031 /* 32 bit does not have separate entry points. */
1032 DEFINE_IDTENTRY_RAW(exc_debug)
1034 unsigned long dr6 = debug_read_clear_dr6();
1036 if (user_mode(regs))
1037 exc_debug_user(regs, dr6);
1039 exc_debug_kernel(regs, dr6);
1044 * Note that we play around with the 'TS' bit in an attempt to get
1045 * the correct behaviour even in the presence of the asynchronous
1048 static void math_error(struct pt_regs *regs, int trapnr)
1050 struct task_struct *task = current;
1051 struct fpu *fpu = &task->thread.fpu;
1053 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
1056 cond_local_irq_enable(regs);
1058 if (!user_mode(regs)) {
1059 if (fixup_exception(regs, trapnr, 0, 0))
1062 task->thread.error_code = 0;
1063 task->thread.trap_nr = trapnr;
1065 if (notify_die(DIE_TRAP, str, regs, 0, trapnr,
1066 SIGFPE) != NOTIFY_STOP)
1072 * Save the info for the exception handler and clear the error.
1076 task->thread.trap_nr = trapnr;
1077 task->thread.error_code = 0;
1079 si_code = fpu__exception_code(fpu, trapnr);
1080 /* Retry when we get spurious exceptions: */
1084 force_sig_fault(SIGFPE, si_code,
1085 (void __user *)uprobe_get_trap_addr(regs));
1087 cond_local_irq_disable(regs);
1090 DEFINE_IDTENTRY(exc_coprocessor_error)
1092 math_error(regs, X86_TRAP_MF);
1095 DEFINE_IDTENTRY(exc_simd_coprocessor_error)
1097 if (IS_ENABLED(CONFIG_X86_INVD_BUG)) {
1098 /* AMD 486 bug: INVD in CPL 0 raises #XF instead of #GP */
1099 if (!static_cpu_has(X86_FEATURE_XMM)) {
1100 __exc_general_protection(regs, 0);
1104 math_error(regs, X86_TRAP_XF);
1107 DEFINE_IDTENTRY(exc_spurious_interrupt_bug)
1110 * This addresses a Pentium Pro Erratum:
1112 * PROBLEM: If the APIC subsystem is configured in mixed mode with
1113 * Virtual Wire mode implemented through the local APIC, an
1114 * interrupt vector of 0Fh (Intel reserved encoding) may be
1115 * generated by the local APIC (Int 15). This vector may be
1116 * generated upon receipt of a spurious interrupt (an interrupt
1117 * which is removed before the system receives the INTA sequence)
1118 * instead of the programmed 8259 spurious interrupt vector.
1120 * IMPLICATION: The spurious interrupt vector programmed in the
1121 * 8259 is normally handled by an operating system's spurious
1122 * interrupt handler. However, a vector of 0Fh is unknown to some
1123 * operating systems, which would crash if this erratum occurred.
1125 * In theory this could be limited to 32bit, but the handler is not
1126 * hurting and who knows which other CPUs suffer from this.
1130 DEFINE_IDTENTRY(exc_device_not_available)
1132 unsigned long cr0 = read_cr0();
1134 #ifdef CONFIG_MATH_EMULATION
1135 if (!boot_cpu_has(X86_FEATURE_FPU) && (cr0 & X86_CR0_EM)) {
1136 struct math_emu_info info = { };
1138 cond_local_irq_enable(regs);
1141 math_emulate(&info);
1143 cond_local_irq_disable(regs);
1148 /* This should not happen. */
1149 if (WARN(cr0 & X86_CR0_TS, "CR0.TS was set")) {
1150 /* Try to fix it up and carry on. */
1151 write_cr0(cr0 & ~X86_CR0_TS);
1154 * Something terrible happened, and we're better off trying
1155 * to kill the task than getting stuck in a never-ending
1156 * loop of #NM faults.
1158 die("unexpected #NM exception", regs, 0);
1162 #ifdef CONFIG_X86_32
1163 DEFINE_IDTENTRY_SW(iret_error)
1166 if (notify_die(DIE_TRAP, "iret exception", regs, 0,
1167 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
1168 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, 0,
1169 ILL_BADSTK, (void __user *)NULL);
1171 local_irq_disable();
1175 void __init trap_init(void)
1177 /* Init cpu_entry_area before IST entries are set up */
1178 setup_cpu_entry_areas();
1180 /* Init GHCB memory pages when running as an SEV-ES guest */
1181 sev_es_init_vc_handling();
1185 cpu_init_exception_handling();
1188 idt_setup_ist_traps();