xref: /openbmc/linux/arch/x86/kernel/traps.c (revision 31354121)
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 #include <linux/hardirq.h>
41 #include <linux/atomic.h>
42 #include <linux/ioasid.h>
43 
44 #include <asm/stacktrace.h>
45 #include <asm/processor.h>
46 #include <asm/debugreg.h>
47 #include <asm/realmode.h>
48 #include <asm/text-patching.h>
49 #include <asm/ftrace.h>
50 #include <asm/traps.h>
51 #include <asm/desc.h>
52 #include <asm/fpu/api.h>
53 #include <asm/cpu.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/vm86.h>
61 #include <asm/umip.h>
62 #include <asm/insn.h>
63 #include <asm/insn-eval.h>
64 #include <asm/vdso.h>
65 #include <asm/tdx.h>
66 #include <asm/cfi.h>
67 
68 #ifdef CONFIG_X86_64
69 #include <asm/x86_init.h>
70 #include <asm/proto.h>
71 #else
72 #include <asm/processor-flags.h>
73 #include <asm/setup.h>
74 #include <asm/proto.h>
75 #endif
76 
77 DECLARE_BITMAP(system_vectors, NR_VECTORS);
78 
79 static inline void cond_local_irq_enable(struct pt_regs *regs)
80 {
81 	if (regs->flags & X86_EFLAGS_IF)
82 		local_irq_enable();
83 }
84 
85 static inline void cond_local_irq_disable(struct pt_regs *regs)
86 {
87 	if (regs->flags & X86_EFLAGS_IF)
88 		local_irq_disable();
89 }
90 
91 __always_inline int is_valid_bugaddr(unsigned long addr)
92 {
93 	if (addr < TASK_SIZE_MAX)
94 		return 0;
95 
96 	/*
97 	 * We got #UD, if the text isn't readable we'd have gotten
98 	 * a different exception.
99 	 */
100 	return *(unsigned short *)addr == INSN_UD2;
101 }
102 
103 static nokprobe_inline int
104 do_trap_no_signal(struct task_struct *tsk, int trapnr, const char *str,
105 		  struct pt_regs *regs,	long error_code)
106 {
107 	if (v8086_mode(regs)) {
108 		/*
109 		 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
110 		 * On nmi (interrupt 2), do_trap should not be called.
111 		 */
112 		if (trapnr < X86_TRAP_UD) {
113 			if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
114 						error_code, trapnr))
115 				return 0;
116 		}
117 	} else if (!user_mode(regs)) {
118 		if (fixup_exception(regs, trapnr, error_code, 0))
119 			return 0;
120 
121 		tsk->thread.error_code = error_code;
122 		tsk->thread.trap_nr = trapnr;
123 		die(str, regs, error_code);
124 	} else {
125 		if (fixup_vdso_exception(regs, trapnr, error_code, 0))
126 			return 0;
127 	}
128 
129 	/*
130 	 * We want error_code and trap_nr set for userspace faults and
131 	 * kernelspace faults which result in die(), but not
132 	 * kernelspace faults which are fixed up.  die() gives the
133 	 * process no chance to handle the signal and notice the
134 	 * kernel fault information, so that won't result in polluting
135 	 * the information about previously queued, but not yet
136 	 * delivered, faults.  See also exc_general_protection below.
137 	 */
138 	tsk->thread.error_code = error_code;
139 	tsk->thread.trap_nr = trapnr;
140 
141 	return -1;
142 }
143 
144 static void show_signal(struct task_struct *tsk, int signr,
145 			const char *type, const char *desc,
146 			struct pt_regs *regs, long error_code)
147 {
148 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
149 	    printk_ratelimit()) {
150 		pr_info("%s[%d] %s%s ip:%lx sp:%lx error:%lx",
151 			tsk->comm, task_pid_nr(tsk), type, desc,
152 			regs->ip, regs->sp, error_code);
153 		print_vma_addr(KERN_CONT " in ", regs->ip);
154 		pr_cont("\n");
155 	}
156 }
157 
158 static void
159 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
160 	long error_code, int sicode, void __user *addr)
161 {
162 	struct task_struct *tsk = current;
163 
164 	if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
165 		return;
166 
167 	show_signal(tsk, signr, "trap ", str, regs, error_code);
168 
169 	if (!sicode)
170 		force_sig(signr);
171 	else
172 		force_sig_fault(signr, sicode, addr);
173 }
174 NOKPROBE_SYMBOL(do_trap);
175 
176 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
177 	unsigned long trapnr, int signr, int sicode, void __user *addr)
178 {
179 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
180 
181 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
182 			NOTIFY_STOP) {
183 		cond_local_irq_enable(regs);
184 		do_trap(trapnr, signr, str, regs, error_code, sicode, addr);
185 		cond_local_irq_disable(regs);
186 	}
187 }
188 
189 /*
190  * Posix requires to provide the address of the faulting instruction for
191  * SIGILL (#UD) and SIGFPE (#DE) in the si_addr member of siginfo_t.
192  *
193  * This address is usually regs->ip, but when an uprobe moved the code out
194  * of line then regs->ip points to the XOL code which would confuse
195  * anything which analyzes the fault address vs. the unmodified binary. If
196  * a trap happened in XOL code then uprobe maps regs->ip back to the
197  * original instruction address.
198  */
199 static __always_inline void __user *error_get_trap_addr(struct pt_regs *regs)
200 {
201 	return (void __user *)uprobe_get_trap_addr(regs);
202 }
203 
204 DEFINE_IDTENTRY(exc_divide_error)
205 {
206 	do_error_trap(regs, 0, "divide error", X86_TRAP_DE, SIGFPE,
207 		      FPE_INTDIV, error_get_trap_addr(regs));
208 }
209 
210 DEFINE_IDTENTRY(exc_overflow)
211 {
212 	do_error_trap(regs, 0, "overflow", X86_TRAP_OF, SIGSEGV, 0, NULL);
213 }
214 
215 #ifdef CONFIG_X86_KERNEL_IBT
216 
217 static __ro_after_init bool ibt_fatal = true;
218 
219 extern void ibt_selftest_ip(void); /* code label defined in asm below */
220 
221 enum cp_error_code {
222 	CP_EC        = (1 << 15) - 1,
223 
224 	CP_RET       = 1,
225 	CP_IRET      = 2,
226 	CP_ENDBR     = 3,
227 	CP_RSTRORSSP = 4,
228 	CP_SETSSBSY  = 5,
229 
230 	CP_ENCL	     = 1 << 15,
231 };
232 
233 DEFINE_IDTENTRY_ERRORCODE(exc_control_protection)
234 {
235 	if (!cpu_feature_enabled(X86_FEATURE_IBT)) {
236 		pr_err("Unexpected #CP\n");
237 		BUG();
238 	}
239 
240 	if (WARN_ON_ONCE(user_mode(regs) || (error_code & CP_EC) != CP_ENDBR))
241 		return;
242 
243 	if (unlikely(regs->ip == (unsigned long)&ibt_selftest_ip)) {
244 		regs->ax = 0;
245 		return;
246 	}
247 
248 	pr_err("Missing ENDBR: %pS\n", (void *)instruction_pointer(regs));
249 	if (!ibt_fatal) {
250 		printk(KERN_DEFAULT CUT_HERE);
251 		__warn(__FILE__, __LINE__, (void *)regs->ip, TAINT_WARN, regs, NULL);
252 		return;
253 	}
254 	BUG();
255 }
256 
257 /* Must be noinline to ensure uniqueness of ibt_selftest_ip. */
258 noinline bool ibt_selftest(void)
259 {
260 	unsigned long ret;
261 
262 	asm ("	lea ibt_selftest_ip(%%rip), %%rax\n\t"
263 	     ANNOTATE_RETPOLINE_SAFE
264 	     "	jmp *%%rax\n\t"
265 	     "ibt_selftest_ip:\n\t"
266 	     UNWIND_HINT_FUNC
267 	     ANNOTATE_NOENDBR
268 	     "	nop\n\t"
269 
270 	     : "=a" (ret) : : "memory");
271 
272 	return !ret;
273 }
274 
275 static int __init ibt_setup(char *str)
276 {
277 	if (!strcmp(str, "off"))
278 		setup_clear_cpu_cap(X86_FEATURE_IBT);
279 
280 	if (!strcmp(str, "warn"))
281 		ibt_fatal = false;
282 
283 	return 1;
284 }
285 
286 __setup("ibt=", ibt_setup);
287 
288 #endif /* CONFIG_X86_KERNEL_IBT */
289 
290 #ifdef CONFIG_X86_F00F_BUG
291 void handle_invalid_op(struct pt_regs *regs)
292 #else
293 static inline void handle_invalid_op(struct pt_regs *regs)
294 #endif
295 {
296 	do_error_trap(regs, 0, "invalid opcode", X86_TRAP_UD, SIGILL,
297 		      ILL_ILLOPN, error_get_trap_addr(regs));
298 }
299 
300 static noinstr bool handle_bug(struct pt_regs *regs)
301 {
302 	bool handled = false;
303 
304 	if (!is_valid_bugaddr(regs->ip))
305 		return handled;
306 
307 	/*
308 	 * All lies, just get the WARN/BUG out.
309 	 */
310 	instrumentation_begin();
311 	/*
312 	 * Since we're emulating a CALL with exceptions, restore the interrupt
313 	 * state to what it was at the exception site.
314 	 */
315 	if (regs->flags & X86_EFLAGS_IF)
316 		raw_local_irq_enable();
317 	if (report_bug(regs->ip, regs) == BUG_TRAP_TYPE_WARN ||
318 	    handle_cfi_failure(regs) == BUG_TRAP_TYPE_WARN) {
319 		regs->ip += LEN_UD2;
320 		handled = true;
321 	}
322 	if (regs->flags & X86_EFLAGS_IF)
323 		raw_local_irq_disable();
324 	instrumentation_end();
325 
326 	return handled;
327 }
328 
329 DEFINE_IDTENTRY_RAW(exc_invalid_op)
330 {
331 	irqentry_state_t state;
332 
333 	/*
334 	 * We use UD2 as a short encoding for 'CALL __WARN', as such
335 	 * handle it before exception entry to avoid recursive WARN
336 	 * in case exception entry is the one triggering WARNs.
337 	 */
338 	if (!user_mode(regs) && handle_bug(regs))
339 		return;
340 
341 	state = irqentry_enter(regs);
342 	instrumentation_begin();
343 	handle_invalid_op(regs);
344 	instrumentation_end();
345 	irqentry_exit(regs, state);
346 }
347 
348 DEFINE_IDTENTRY(exc_coproc_segment_overrun)
349 {
350 	do_error_trap(regs, 0, "coprocessor segment overrun",
351 		      X86_TRAP_OLD_MF, SIGFPE, 0, NULL);
352 }
353 
354 DEFINE_IDTENTRY_ERRORCODE(exc_invalid_tss)
355 {
356 	do_error_trap(regs, error_code, "invalid TSS", X86_TRAP_TS, SIGSEGV,
357 		      0, NULL);
358 }
359 
360 DEFINE_IDTENTRY_ERRORCODE(exc_segment_not_present)
361 {
362 	do_error_trap(regs, error_code, "segment not present", X86_TRAP_NP,
363 		      SIGBUS, 0, NULL);
364 }
365 
366 DEFINE_IDTENTRY_ERRORCODE(exc_stack_segment)
367 {
368 	do_error_trap(regs, error_code, "stack segment", X86_TRAP_SS, SIGBUS,
369 		      0, NULL);
370 }
371 
372 DEFINE_IDTENTRY_ERRORCODE(exc_alignment_check)
373 {
374 	char *str = "alignment check";
375 
376 	if (notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_AC, SIGBUS) == NOTIFY_STOP)
377 		return;
378 
379 	if (!user_mode(regs))
380 		die("Split lock detected\n", regs, error_code);
381 
382 	local_irq_enable();
383 
384 	if (handle_user_split_lock(regs, error_code))
385 		goto out;
386 
387 	do_trap(X86_TRAP_AC, SIGBUS, "alignment check", regs,
388 		error_code, BUS_ADRALN, NULL);
389 
390 out:
391 	local_irq_disable();
392 }
393 
394 #ifdef CONFIG_VMAP_STACK
395 __visible void __noreturn handle_stack_overflow(struct pt_regs *regs,
396 						unsigned long fault_address,
397 						struct stack_info *info)
398 {
399 	const char *name = stack_type_name(info->type);
400 
401 	printk(KERN_EMERG "BUG: %s stack guard page was hit at %p (stack is %p..%p)\n",
402 	       name, (void *)fault_address, info->begin, info->end);
403 
404 	die("stack guard page", regs, 0);
405 
406 	/* Be absolutely certain we don't return. */
407 	panic("%s stack guard hit", name);
408 }
409 #endif
410 
411 /*
412  * Runs on an IST stack for x86_64 and on a special task stack for x86_32.
413  *
414  * On x86_64, this is more or less a normal kernel entry.  Notwithstanding the
415  * SDM's warnings about double faults being unrecoverable, returning works as
416  * expected.  Presumably what the SDM actually means is that the CPU may get
417  * the register state wrong on entry, so returning could be a bad idea.
418  *
419  * Various CPU engineers have promised that double faults due to an IRET fault
420  * while the stack is read-only are, in fact, recoverable.
421  *
422  * On x86_32, this is entered through a task gate, and regs are synthesized
423  * from the TSS.  Returning is, in principle, okay, but changes to regs will
424  * be lost.  If, for some reason, we need to return to a context with modified
425  * regs, the shim code could be adjusted to synchronize the registers.
426  *
427  * The 32bit #DF shim provides CR2 already as an argument. On 64bit it needs
428  * to be read before doing anything else.
429  */
430 DEFINE_IDTENTRY_DF(exc_double_fault)
431 {
432 	static const char str[] = "double fault";
433 	struct task_struct *tsk = current;
434 
435 #ifdef CONFIG_VMAP_STACK
436 	unsigned long address = read_cr2();
437 	struct stack_info info;
438 #endif
439 
440 #ifdef CONFIG_X86_ESPFIX64
441 	extern unsigned char native_irq_return_iret[];
442 
443 	/*
444 	 * If IRET takes a non-IST fault on the espfix64 stack, then we
445 	 * end up promoting it to a doublefault.  In that case, take
446 	 * advantage of the fact that we're not using the normal (TSS.sp0)
447 	 * stack right now.  We can write a fake #GP(0) frame at TSS.sp0
448 	 * and then modify our own IRET frame so that, when we return,
449 	 * we land directly at the #GP(0) vector with the stack already
450 	 * set up according to its expectations.
451 	 *
452 	 * The net result is that our #GP handler will think that we
453 	 * entered from usermode with the bad user context.
454 	 *
455 	 * No need for nmi_enter() here because we don't use RCU.
456 	 */
457 	if (((long)regs->sp >> P4D_SHIFT) == ESPFIX_PGD_ENTRY &&
458 		regs->cs == __KERNEL_CS &&
459 		regs->ip == (unsigned long)native_irq_return_iret)
460 	{
461 		struct pt_regs *gpregs = (struct pt_regs *)this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
462 		unsigned long *p = (unsigned long *)regs->sp;
463 
464 		/*
465 		 * regs->sp points to the failing IRET frame on the
466 		 * ESPFIX64 stack.  Copy it to the entry stack.  This fills
467 		 * in gpregs->ss through gpregs->ip.
468 		 *
469 		 */
470 		gpregs->ip	= p[0];
471 		gpregs->cs	= p[1];
472 		gpregs->flags	= p[2];
473 		gpregs->sp	= p[3];
474 		gpregs->ss	= p[4];
475 		gpregs->orig_ax = 0;  /* Missing (lost) #GP error code */
476 
477 		/*
478 		 * Adjust our frame so that we return straight to the #GP
479 		 * vector with the expected RSP value.  This is safe because
480 		 * we won't enable interrupts or schedule before we invoke
481 		 * general_protection, so nothing will clobber the stack
482 		 * frame we just set up.
483 		 *
484 		 * We will enter general_protection with kernel GSBASE,
485 		 * which is what the stub expects, given that the faulting
486 		 * RIP will be the IRET instruction.
487 		 */
488 		regs->ip = (unsigned long)asm_exc_general_protection;
489 		regs->sp = (unsigned long)&gpregs->orig_ax;
490 
491 		return;
492 	}
493 #endif
494 
495 	irqentry_nmi_enter(regs);
496 	instrumentation_begin();
497 	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
498 
499 	tsk->thread.error_code = error_code;
500 	tsk->thread.trap_nr = X86_TRAP_DF;
501 
502 #ifdef CONFIG_VMAP_STACK
503 	/*
504 	 * If we overflow the stack into a guard page, the CPU will fail
505 	 * to deliver #PF and will send #DF instead.  Similarly, if we
506 	 * take any non-IST exception while too close to the bottom of
507 	 * the stack, the processor will get a page fault while
508 	 * delivering the exception and will generate a double fault.
509 	 *
510 	 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
511 	 * Page-Fault Exception (#PF):
512 	 *
513 	 *   Processors update CR2 whenever a page fault is detected. If a
514 	 *   second page fault occurs while an earlier page fault is being
515 	 *   delivered, the faulting linear address of the second fault will
516 	 *   overwrite the contents of CR2 (replacing the previous
517 	 *   address). These updates to CR2 occur even if the page fault
518 	 *   results in a double fault or occurs during the delivery of a
519 	 *   double fault.
520 	 *
521 	 * The logic below has a small possibility of incorrectly diagnosing
522 	 * some errors as stack overflows.  For example, if the IDT or GDT
523 	 * gets corrupted such that #GP delivery fails due to a bad descriptor
524 	 * causing #GP and we hit this condition while CR2 coincidentally
525 	 * points to the stack guard page, we'll think we overflowed the
526 	 * stack.  Given that we're going to panic one way or another
527 	 * if this happens, this isn't necessarily worth fixing.
528 	 *
529 	 * If necessary, we could improve the test by only diagnosing
530 	 * a stack overflow if the saved RSP points within 47 bytes of
531 	 * the bottom of the stack: if RSP == tsk_stack + 48 and we
532 	 * take an exception, the stack is already aligned and there
533 	 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
534 	 * possible error code, so a stack overflow would *not* double
535 	 * fault.  With any less space left, exception delivery could
536 	 * fail, and, as a practical matter, we've overflowed the
537 	 * stack even if the actual trigger for the double fault was
538 	 * something else.
539 	 */
540 	if (get_stack_guard_info((void *)address, &info))
541 		handle_stack_overflow(regs, address, &info);
542 #endif
543 
544 	pr_emerg("PANIC: double fault, error_code: 0x%lx\n", error_code);
545 	die("double fault", regs, error_code);
546 	panic("Machine halted.");
547 	instrumentation_end();
548 }
549 
550 DEFINE_IDTENTRY(exc_bounds)
551 {
552 	if (notify_die(DIE_TRAP, "bounds", regs, 0,
553 			X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
554 		return;
555 	cond_local_irq_enable(regs);
556 
557 	if (!user_mode(regs))
558 		die("bounds", regs, 0);
559 
560 	do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, 0, 0, NULL);
561 
562 	cond_local_irq_disable(regs);
563 }
564 
565 enum kernel_gp_hint {
566 	GP_NO_HINT,
567 	GP_NON_CANONICAL,
568 	GP_CANONICAL
569 };
570 
571 /*
572  * When an uncaught #GP occurs, try to determine the memory address accessed by
573  * the instruction and return that address to the caller. Also, try to figure
574  * out whether any part of the access to that address was non-canonical.
575  */
576 static enum kernel_gp_hint get_kernel_gp_address(struct pt_regs *regs,
577 						 unsigned long *addr)
578 {
579 	u8 insn_buf[MAX_INSN_SIZE];
580 	struct insn insn;
581 	int ret;
582 
583 	if (copy_from_kernel_nofault(insn_buf, (void *)regs->ip,
584 			MAX_INSN_SIZE))
585 		return GP_NO_HINT;
586 
587 	ret = insn_decode_kernel(&insn, insn_buf);
588 	if (ret < 0)
589 		return GP_NO_HINT;
590 
591 	*addr = (unsigned long)insn_get_addr_ref(&insn, regs);
592 	if (*addr == -1UL)
593 		return GP_NO_HINT;
594 
595 #ifdef CONFIG_X86_64
596 	/*
597 	 * Check that:
598 	 *  - the operand is not in the kernel half
599 	 *  - the last byte of the operand is not in the user canonical half
600 	 */
601 	if (*addr < ~__VIRTUAL_MASK &&
602 	    *addr + insn.opnd_bytes - 1 > __VIRTUAL_MASK)
603 		return GP_NON_CANONICAL;
604 #endif
605 
606 	return GP_CANONICAL;
607 }
608 
609 #define GPFSTR "general protection fault"
610 
611 static bool fixup_iopl_exception(struct pt_regs *regs)
612 {
613 	struct thread_struct *t = &current->thread;
614 	unsigned char byte;
615 	unsigned long ip;
616 
617 	if (!IS_ENABLED(CONFIG_X86_IOPL_IOPERM) || t->iopl_emul != 3)
618 		return false;
619 
620 	if (insn_get_effective_ip(regs, &ip))
621 		return false;
622 
623 	if (get_user(byte, (const char __user *)ip))
624 		return false;
625 
626 	if (byte != 0xfa && byte != 0xfb)
627 		return false;
628 
629 	if (!t->iopl_warn && printk_ratelimit()) {
630 		pr_err("%s[%d] attempts to use CLI/STI, pretending it's a NOP, ip:%lx",
631 		       current->comm, task_pid_nr(current), ip);
632 		print_vma_addr(KERN_CONT " in ", ip);
633 		pr_cont("\n");
634 		t->iopl_warn = 1;
635 	}
636 
637 	regs->ip += 1;
638 	return true;
639 }
640 
641 /*
642  * The unprivileged ENQCMD instruction generates #GPs if the
643  * IA32_PASID MSR has not been populated.  If possible, populate
644  * the MSR from a PASID previously allocated to the mm.
645  */
646 static bool try_fixup_enqcmd_gp(void)
647 {
648 #ifdef CONFIG_IOMMU_SVA
649 	u32 pasid;
650 
651 	/*
652 	 * MSR_IA32_PASID is managed using XSAVE.  Directly
653 	 * writing to the MSR is only possible when fpregs
654 	 * are valid and the fpstate is not.  This is
655 	 * guaranteed when handling a userspace exception
656 	 * in *before* interrupts are re-enabled.
657 	 */
658 	lockdep_assert_irqs_disabled();
659 
660 	/*
661 	 * Hardware without ENQCMD will not generate
662 	 * #GPs that can be fixed up here.
663 	 */
664 	if (!cpu_feature_enabled(X86_FEATURE_ENQCMD))
665 		return false;
666 
667 	pasid = current->mm->pasid;
668 
669 	/*
670 	 * If the mm has not been allocated a
671 	 * PASID, the #GP can not be fixed up.
672 	 */
673 	if (!pasid_valid(pasid))
674 		return false;
675 
676 	/*
677 	 * Did this thread already have its PASID activated?
678 	 * If so, the #GP must be from something else.
679 	 */
680 	if (current->pasid_activated)
681 		return false;
682 
683 	wrmsrl(MSR_IA32_PASID, pasid | MSR_IA32_PASID_VALID);
684 	current->pasid_activated = 1;
685 
686 	return true;
687 #else
688 	return false;
689 #endif
690 }
691 
692 static bool gp_try_fixup_and_notify(struct pt_regs *regs, int trapnr,
693 				    unsigned long error_code, const char *str)
694 {
695 	if (fixup_exception(regs, trapnr, error_code, 0))
696 		return true;
697 
698 	current->thread.error_code = error_code;
699 	current->thread.trap_nr = trapnr;
700 
701 	/*
702 	 * To be potentially processing a kprobe fault and to trust the result
703 	 * from kprobe_running(), we have to be non-preemptible.
704 	 */
705 	if (!preemptible() && kprobe_running() &&
706 	    kprobe_fault_handler(regs, trapnr))
707 		return true;
708 
709 	return notify_die(DIE_GPF, str, regs, error_code, trapnr, SIGSEGV) == NOTIFY_STOP;
710 }
711 
712 static void gp_user_force_sig_segv(struct pt_regs *regs, int trapnr,
713 				   unsigned long error_code, const char *str)
714 {
715 	current->thread.error_code = error_code;
716 	current->thread.trap_nr = trapnr;
717 	show_signal(current, SIGSEGV, "", str, regs, error_code);
718 	force_sig(SIGSEGV);
719 }
720 
721 DEFINE_IDTENTRY_ERRORCODE(exc_general_protection)
722 {
723 	char desc[sizeof(GPFSTR) + 50 + 2*sizeof(unsigned long) + 1] = GPFSTR;
724 	enum kernel_gp_hint hint = GP_NO_HINT;
725 	unsigned long gp_addr;
726 
727 	if (user_mode(regs) && try_fixup_enqcmd_gp())
728 		return;
729 
730 	cond_local_irq_enable(regs);
731 
732 	if (static_cpu_has(X86_FEATURE_UMIP)) {
733 		if (user_mode(regs) && fixup_umip_exception(regs))
734 			goto exit;
735 	}
736 
737 	if (v8086_mode(regs)) {
738 		local_irq_enable();
739 		handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
740 		local_irq_disable();
741 		return;
742 	}
743 
744 	if (user_mode(regs)) {
745 		if (fixup_iopl_exception(regs))
746 			goto exit;
747 
748 		if (fixup_vdso_exception(regs, X86_TRAP_GP, error_code, 0))
749 			goto exit;
750 
751 		gp_user_force_sig_segv(regs, X86_TRAP_GP, error_code, desc);
752 		goto exit;
753 	}
754 
755 	if (gp_try_fixup_and_notify(regs, X86_TRAP_GP, error_code, desc))
756 		goto exit;
757 
758 	if (error_code)
759 		snprintf(desc, sizeof(desc), "segment-related " GPFSTR);
760 	else
761 		hint = get_kernel_gp_address(regs, &gp_addr);
762 
763 	if (hint != GP_NO_HINT)
764 		snprintf(desc, sizeof(desc), GPFSTR ", %s 0x%lx",
765 			 (hint == GP_NON_CANONICAL) ? "probably for non-canonical address"
766 						    : "maybe for address",
767 			 gp_addr);
768 
769 	/*
770 	 * KASAN is interested only in the non-canonical case, clear it
771 	 * otherwise.
772 	 */
773 	if (hint != GP_NON_CANONICAL)
774 		gp_addr = 0;
775 
776 	die_addr(desc, regs, error_code, gp_addr);
777 
778 exit:
779 	cond_local_irq_disable(regs);
780 }
781 
782 static bool do_int3(struct pt_regs *regs)
783 {
784 	int res;
785 
786 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
787 	if (kgdb_ll_trap(DIE_INT3, "int3", regs, 0, X86_TRAP_BP,
788 			 SIGTRAP) == NOTIFY_STOP)
789 		return true;
790 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
791 
792 #ifdef CONFIG_KPROBES
793 	if (kprobe_int3_handler(regs))
794 		return true;
795 #endif
796 	res = notify_die(DIE_INT3, "int3", regs, 0, X86_TRAP_BP, SIGTRAP);
797 
798 	return res == NOTIFY_STOP;
799 }
800 NOKPROBE_SYMBOL(do_int3);
801 
802 static void do_int3_user(struct pt_regs *regs)
803 {
804 	if (do_int3(regs))
805 		return;
806 
807 	cond_local_irq_enable(regs);
808 	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, 0, 0, NULL);
809 	cond_local_irq_disable(regs);
810 }
811 
812 DEFINE_IDTENTRY_RAW(exc_int3)
813 {
814 	/*
815 	 * poke_int3_handler() is completely self contained code; it does (and
816 	 * must) *NOT* call out to anything, lest it hits upon yet another
817 	 * INT3.
818 	 */
819 	if (poke_int3_handler(regs))
820 		return;
821 
822 	/*
823 	 * irqentry_enter_from_user_mode() uses static_branch_{,un}likely()
824 	 * and therefore can trigger INT3, hence poke_int3_handler() must
825 	 * be done before. If the entry came from kernel mode, then use
826 	 * nmi_enter() because the INT3 could have been hit in any context
827 	 * including NMI.
828 	 */
829 	if (user_mode(regs)) {
830 		irqentry_enter_from_user_mode(regs);
831 		instrumentation_begin();
832 		do_int3_user(regs);
833 		instrumentation_end();
834 		irqentry_exit_to_user_mode(regs);
835 	} else {
836 		irqentry_state_t irq_state = irqentry_nmi_enter(regs);
837 
838 		instrumentation_begin();
839 		if (!do_int3(regs))
840 			die("int3", regs, 0);
841 		instrumentation_end();
842 		irqentry_nmi_exit(regs, irq_state);
843 	}
844 }
845 
846 #ifdef CONFIG_X86_64
847 /*
848  * Help handler running on a per-cpu (IST or entry trampoline) stack
849  * to switch to the normal thread stack if the interrupted code was in
850  * user mode. The actual stack switch is done in entry_64.S
851  */
852 asmlinkage __visible noinstr struct pt_regs *sync_regs(struct pt_regs *eregs)
853 {
854 	struct pt_regs *regs = (struct pt_regs *)this_cpu_read(cpu_current_top_of_stack) - 1;
855 	if (regs != eregs)
856 		*regs = *eregs;
857 	return regs;
858 }
859 
860 #ifdef CONFIG_AMD_MEM_ENCRYPT
861 asmlinkage __visible noinstr struct pt_regs *vc_switch_off_ist(struct pt_regs *regs)
862 {
863 	unsigned long sp, *stack;
864 	struct stack_info info;
865 	struct pt_regs *regs_ret;
866 
867 	/*
868 	 * In the SYSCALL entry path the RSP value comes from user-space - don't
869 	 * trust it and switch to the current kernel stack
870 	 */
871 	if (ip_within_syscall_gap(regs)) {
872 		sp = this_cpu_read(cpu_current_top_of_stack);
873 		goto sync;
874 	}
875 
876 	/*
877 	 * From here on the RSP value is trusted. Now check whether entry
878 	 * happened from a safe stack. Not safe are the entry or unknown stacks,
879 	 * use the fall-back stack instead in this case.
880 	 */
881 	sp    = regs->sp;
882 	stack = (unsigned long *)sp;
883 
884 	if (!get_stack_info_noinstr(stack, current, &info) || info.type == STACK_TYPE_ENTRY ||
885 	    info.type > STACK_TYPE_EXCEPTION_LAST)
886 		sp = __this_cpu_ist_top_va(VC2);
887 
888 sync:
889 	/*
890 	 * Found a safe stack - switch to it as if the entry didn't happen via
891 	 * IST stack. The code below only copies pt_regs, the real switch happens
892 	 * in assembly code.
893 	 */
894 	sp = ALIGN_DOWN(sp, 8) - sizeof(*regs_ret);
895 
896 	regs_ret = (struct pt_regs *)sp;
897 	*regs_ret = *regs;
898 
899 	return regs_ret;
900 }
901 #endif
902 
903 asmlinkage __visible noinstr struct pt_regs *fixup_bad_iret(struct pt_regs *bad_regs)
904 {
905 	struct pt_regs tmp, *new_stack;
906 
907 	/*
908 	 * This is called from entry_64.S early in handling a fault
909 	 * caused by a bad iret to user mode.  To handle the fault
910 	 * correctly, we want to move our stack frame to where it would
911 	 * be had we entered directly on the entry stack (rather than
912 	 * just below the IRET frame) and we want to pretend that the
913 	 * exception came from the IRET target.
914 	 */
915 	new_stack = (struct pt_regs *)__this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
916 
917 	/* Copy the IRET target to the temporary storage. */
918 	__memcpy(&tmp.ip, (void *)bad_regs->sp, 5*8);
919 
920 	/* Copy the remainder of the stack from the current stack. */
921 	__memcpy(&tmp, bad_regs, offsetof(struct pt_regs, ip));
922 
923 	/* Update the entry stack */
924 	__memcpy(new_stack, &tmp, sizeof(tmp));
925 
926 	BUG_ON(!user_mode(new_stack));
927 	return new_stack;
928 }
929 #endif
930 
931 static bool is_sysenter_singlestep(struct pt_regs *regs)
932 {
933 	/*
934 	 * We don't try for precision here.  If we're anywhere in the region of
935 	 * code that can be single-stepped in the SYSENTER entry path, then
936 	 * assume that this is a useless single-step trap due to SYSENTER
937 	 * being invoked with TF set.  (We don't know in advance exactly
938 	 * which instructions will be hit because BTF could plausibly
939 	 * be set.)
940 	 */
941 #ifdef CONFIG_X86_32
942 	return (regs->ip - (unsigned long)__begin_SYSENTER_singlestep_region) <
943 		(unsigned long)__end_SYSENTER_singlestep_region -
944 		(unsigned long)__begin_SYSENTER_singlestep_region;
945 #elif defined(CONFIG_IA32_EMULATION)
946 	return (regs->ip - (unsigned long)entry_SYSENTER_compat) <
947 		(unsigned long)__end_entry_SYSENTER_compat -
948 		(unsigned long)entry_SYSENTER_compat;
949 #else
950 	return false;
951 #endif
952 }
953 
954 static __always_inline unsigned long debug_read_clear_dr6(void)
955 {
956 	unsigned long dr6;
957 
958 	/*
959 	 * The Intel SDM says:
960 	 *
961 	 *   Certain debug exceptions may clear bits 0-3. The remaining
962 	 *   contents of the DR6 register are never cleared by the
963 	 *   processor. To avoid confusion in identifying debug
964 	 *   exceptions, debug handlers should clear the register before
965 	 *   returning to the interrupted task.
966 	 *
967 	 * Keep it simple: clear DR6 immediately.
968 	 */
969 	get_debugreg(dr6, 6);
970 	set_debugreg(DR6_RESERVED, 6);
971 	dr6 ^= DR6_RESERVED; /* Flip to positive polarity */
972 
973 	return dr6;
974 }
975 
976 /*
977  * Our handling of the processor debug registers is non-trivial.
978  * We do not clear them on entry and exit from the kernel. Therefore
979  * it is possible to get a watchpoint trap here from inside the kernel.
980  * However, the code in ./ptrace.c has ensured that the user can
981  * only set watchpoints on userspace addresses. Therefore the in-kernel
982  * watchpoint trap can only occur in code which is reading/writing
983  * from user space. Such code must not hold kernel locks (since it
984  * can equally take a page fault), therefore it is safe to call
985  * force_sig_info even though that claims and releases locks.
986  *
987  * Code in ./signal.c ensures that the debug control register
988  * is restored before we deliver any signal, and therefore that
989  * user code runs with the correct debug control register even though
990  * we clear it here.
991  *
992  * Being careful here means that we don't have to be as careful in a
993  * lot of more complicated places (task switching can be a bit lazy
994  * about restoring all the debug state, and ptrace doesn't have to
995  * find every occurrence of the TF bit that could be saved away even
996  * by user code)
997  *
998  * May run on IST stack.
999  */
1000 
1001 static bool notify_debug(struct pt_regs *regs, unsigned long *dr6)
1002 {
1003 	/*
1004 	 * Notifiers will clear bits in @dr6 to indicate the event has been
1005 	 * consumed - hw_breakpoint_handler(), single_stop_cont().
1006 	 *
1007 	 * Notifiers will set bits in @virtual_dr6 to indicate the desire
1008 	 * for signals - ptrace_triggered(), kgdb_hw_overflow_handler().
1009 	 */
1010 	if (notify_die(DIE_DEBUG, "debug", regs, (long)dr6, 0, SIGTRAP) == NOTIFY_STOP)
1011 		return true;
1012 
1013 	return false;
1014 }
1015 
1016 static __always_inline void exc_debug_kernel(struct pt_regs *regs,
1017 					     unsigned long dr6)
1018 {
1019 	/*
1020 	 * Disable breakpoints during exception handling; recursive exceptions
1021 	 * are exceedingly 'fun'.
1022 	 *
1023 	 * Since this function is NOKPROBE, and that also applies to
1024 	 * HW_BREAKPOINT_X, we can't hit a breakpoint before this (XXX except a
1025 	 * HW_BREAKPOINT_W on our stack)
1026 	 *
1027 	 * Entry text is excluded for HW_BP_X and cpu_entry_area, which
1028 	 * includes the entry stack is excluded for everything.
1029 	 */
1030 	unsigned long dr7 = local_db_save();
1031 	irqentry_state_t irq_state = irqentry_nmi_enter(regs);
1032 	instrumentation_begin();
1033 
1034 	/*
1035 	 * If something gets miswired and we end up here for a user mode
1036 	 * #DB, we will malfunction.
1037 	 */
1038 	WARN_ON_ONCE(user_mode(regs));
1039 
1040 	if (test_thread_flag(TIF_BLOCKSTEP)) {
1041 		/*
1042 		 * The SDM says "The processor clears the BTF flag when it
1043 		 * generates a debug exception." but PTRACE_BLOCKSTEP requested
1044 		 * it for userspace, but we just took a kernel #DB, so re-set
1045 		 * BTF.
1046 		 */
1047 		unsigned long debugctl;
1048 
1049 		rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
1050 		debugctl |= DEBUGCTLMSR_BTF;
1051 		wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
1052 	}
1053 
1054 	/*
1055 	 * Catch SYSENTER with TF set and clear DR_STEP. If this hit a
1056 	 * watchpoint at the same time then that will still be handled.
1057 	 */
1058 	if ((dr6 & DR_STEP) && is_sysenter_singlestep(regs))
1059 		dr6 &= ~DR_STEP;
1060 
1061 	/*
1062 	 * The kernel doesn't use INT1
1063 	 */
1064 	if (!dr6)
1065 		goto out;
1066 
1067 	if (notify_debug(regs, &dr6))
1068 		goto out;
1069 
1070 	/*
1071 	 * The kernel doesn't use TF single-step outside of:
1072 	 *
1073 	 *  - Kprobes, consumed through kprobe_debug_handler()
1074 	 *  - KGDB, consumed through notify_debug()
1075 	 *
1076 	 * So if we get here with DR_STEP set, something is wonky.
1077 	 *
1078 	 * A known way to trigger this is through QEMU's GDB stub,
1079 	 * which leaks #DB into the guest and causes IST recursion.
1080 	 */
1081 	if (WARN_ON_ONCE(dr6 & DR_STEP))
1082 		regs->flags &= ~X86_EFLAGS_TF;
1083 out:
1084 	instrumentation_end();
1085 	irqentry_nmi_exit(regs, irq_state);
1086 
1087 	local_db_restore(dr7);
1088 }
1089 
1090 static __always_inline void exc_debug_user(struct pt_regs *regs,
1091 					   unsigned long dr6)
1092 {
1093 	bool icebp;
1094 
1095 	/*
1096 	 * If something gets miswired and we end up here for a kernel mode
1097 	 * #DB, we will malfunction.
1098 	 */
1099 	WARN_ON_ONCE(!user_mode(regs));
1100 
1101 	/*
1102 	 * NB: We can't easily clear DR7 here because
1103 	 * irqentry_exit_to_usermode() can invoke ptrace, schedule, access
1104 	 * user memory, etc.  This means that a recursive #DB is possible.  If
1105 	 * this happens, that #DB will hit exc_debug_kernel() and clear DR7.
1106 	 * Since we're not on the IST stack right now, everything will be
1107 	 * fine.
1108 	 */
1109 
1110 	irqentry_enter_from_user_mode(regs);
1111 	instrumentation_begin();
1112 
1113 	/*
1114 	 * Start the virtual/ptrace DR6 value with just the DR_STEP mask
1115 	 * of the real DR6. ptrace_triggered() will set the DR_TRAPn bits.
1116 	 *
1117 	 * Userspace expects DR_STEP to be visible in ptrace_get_debugreg(6)
1118 	 * even if it is not the result of PTRACE_SINGLESTEP.
1119 	 */
1120 	current->thread.virtual_dr6 = (dr6 & DR_STEP);
1121 
1122 	/*
1123 	 * The SDM says "The processor clears the BTF flag when it
1124 	 * generates a debug exception."  Clear TIF_BLOCKSTEP to keep
1125 	 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
1126 	 */
1127 	clear_thread_flag(TIF_BLOCKSTEP);
1128 
1129 	/*
1130 	 * If dr6 has no reason to give us about the origin of this trap,
1131 	 * then it's very likely the result of an icebp/int01 trap.
1132 	 * User wants a sigtrap for that.
1133 	 */
1134 	icebp = !dr6;
1135 
1136 	if (notify_debug(regs, &dr6))
1137 		goto out;
1138 
1139 	/* It's safe to allow irq's after DR6 has been saved */
1140 	local_irq_enable();
1141 
1142 	if (v8086_mode(regs)) {
1143 		handle_vm86_trap((struct kernel_vm86_regs *)regs, 0, X86_TRAP_DB);
1144 		goto out_irq;
1145 	}
1146 
1147 	/* #DB for bus lock can only be triggered from userspace. */
1148 	if (dr6 & DR_BUS_LOCK)
1149 		handle_bus_lock(regs);
1150 
1151 	/* Add the virtual_dr6 bits for signals. */
1152 	dr6 |= current->thread.virtual_dr6;
1153 	if (dr6 & (DR_STEP | DR_TRAP_BITS) || icebp)
1154 		send_sigtrap(regs, 0, get_si_code(dr6));
1155 
1156 out_irq:
1157 	local_irq_disable();
1158 out:
1159 	instrumentation_end();
1160 	irqentry_exit_to_user_mode(regs);
1161 }
1162 
1163 #ifdef CONFIG_X86_64
1164 /* IST stack entry */
1165 DEFINE_IDTENTRY_DEBUG(exc_debug)
1166 {
1167 	exc_debug_kernel(regs, debug_read_clear_dr6());
1168 }
1169 
1170 /* User entry, runs on regular task stack */
1171 DEFINE_IDTENTRY_DEBUG_USER(exc_debug)
1172 {
1173 	exc_debug_user(regs, debug_read_clear_dr6());
1174 }
1175 #else
1176 /* 32 bit does not have separate entry points. */
1177 DEFINE_IDTENTRY_RAW(exc_debug)
1178 {
1179 	unsigned long dr6 = debug_read_clear_dr6();
1180 
1181 	if (user_mode(regs))
1182 		exc_debug_user(regs, dr6);
1183 	else
1184 		exc_debug_kernel(regs, dr6);
1185 }
1186 #endif
1187 
1188 /*
1189  * Note that we play around with the 'TS' bit in an attempt to get
1190  * the correct behaviour even in the presence of the asynchronous
1191  * IRQ13 behaviour
1192  */
1193 static void math_error(struct pt_regs *regs, int trapnr)
1194 {
1195 	struct task_struct *task = current;
1196 	struct fpu *fpu = &task->thread.fpu;
1197 	int si_code;
1198 	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
1199 						"simd exception";
1200 
1201 	cond_local_irq_enable(regs);
1202 
1203 	if (!user_mode(regs)) {
1204 		if (fixup_exception(regs, trapnr, 0, 0))
1205 			goto exit;
1206 
1207 		task->thread.error_code = 0;
1208 		task->thread.trap_nr = trapnr;
1209 
1210 		if (notify_die(DIE_TRAP, str, regs, 0, trapnr,
1211 			       SIGFPE) != NOTIFY_STOP)
1212 			die(str, regs, 0);
1213 		goto exit;
1214 	}
1215 
1216 	/*
1217 	 * Synchronize the FPU register state to the memory register state
1218 	 * if necessary. This allows the exception handler to inspect it.
1219 	 */
1220 	fpu_sync_fpstate(fpu);
1221 
1222 	task->thread.trap_nr	= trapnr;
1223 	task->thread.error_code = 0;
1224 
1225 	si_code = fpu__exception_code(fpu, trapnr);
1226 	/* Retry when we get spurious exceptions: */
1227 	if (!si_code)
1228 		goto exit;
1229 
1230 	if (fixup_vdso_exception(regs, trapnr, 0, 0))
1231 		goto exit;
1232 
1233 	force_sig_fault(SIGFPE, si_code,
1234 			(void __user *)uprobe_get_trap_addr(regs));
1235 exit:
1236 	cond_local_irq_disable(regs);
1237 }
1238 
1239 DEFINE_IDTENTRY(exc_coprocessor_error)
1240 {
1241 	math_error(regs, X86_TRAP_MF);
1242 }
1243 
1244 DEFINE_IDTENTRY(exc_simd_coprocessor_error)
1245 {
1246 	if (IS_ENABLED(CONFIG_X86_INVD_BUG)) {
1247 		/* AMD 486 bug: INVD in CPL 0 raises #XF instead of #GP */
1248 		if (!static_cpu_has(X86_FEATURE_XMM)) {
1249 			__exc_general_protection(regs, 0);
1250 			return;
1251 		}
1252 	}
1253 	math_error(regs, X86_TRAP_XF);
1254 }
1255 
1256 DEFINE_IDTENTRY(exc_spurious_interrupt_bug)
1257 {
1258 	/*
1259 	 * This addresses a Pentium Pro Erratum:
1260 	 *
1261 	 * PROBLEM: If the APIC subsystem is configured in mixed mode with
1262 	 * Virtual Wire mode implemented through the local APIC, an
1263 	 * interrupt vector of 0Fh (Intel reserved encoding) may be
1264 	 * generated by the local APIC (Int 15).  This vector may be
1265 	 * generated upon receipt of a spurious interrupt (an interrupt
1266 	 * which is removed before the system receives the INTA sequence)
1267 	 * instead of the programmed 8259 spurious interrupt vector.
1268 	 *
1269 	 * IMPLICATION: The spurious interrupt vector programmed in the
1270 	 * 8259 is normally handled by an operating system's spurious
1271 	 * interrupt handler. However, a vector of 0Fh is unknown to some
1272 	 * operating systems, which would crash if this erratum occurred.
1273 	 *
1274 	 * In theory this could be limited to 32bit, but the handler is not
1275 	 * hurting and who knows which other CPUs suffer from this.
1276 	 */
1277 }
1278 
1279 static bool handle_xfd_event(struct pt_regs *regs)
1280 {
1281 	u64 xfd_err;
1282 	int err;
1283 
1284 	if (!IS_ENABLED(CONFIG_X86_64) || !cpu_feature_enabled(X86_FEATURE_XFD))
1285 		return false;
1286 
1287 	rdmsrl(MSR_IA32_XFD_ERR, xfd_err);
1288 	if (!xfd_err)
1289 		return false;
1290 
1291 	wrmsrl(MSR_IA32_XFD_ERR, 0);
1292 
1293 	/* Die if that happens in kernel space */
1294 	if (WARN_ON(!user_mode(regs)))
1295 		return false;
1296 
1297 	local_irq_enable();
1298 
1299 	err = xfd_enable_feature(xfd_err);
1300 
1301 	switch (err) {
1302 	case -EPERM:
1303 		force_sig_fault(SIGILL, ILL_ILLOPC, error_get_trap_addr(regs));
1304 		break;
1305 	case -EFAULT:
1306 		force_sig(SIGSEGV);
1307 		break;
1308 	}
1309 
1310 	local_irq_disable();
1311 	return true;
1312 }
1313 
1314 DEFINE_IDTENTRY(exc_device_not_available)
1315 {
1316 	unsigned long cr0 = read_cr0();
1317 
1318 	if (handle_xfd_event(regs))
1319 		return;
1320 
1321 #ifdef CONFIG_MATH_EMULATION
1322 	if (!boot_cpu_has(X86_FEATURE_FPU) && (cr0 & X86_CR0_EM)) {
1323 		struct math_emu_info info = { };
1324 
1325 		cond_local_irq_enable(regs);
1326 
1327 		info.regs = regs;
1328 		math_emulate(&info);
1329 
1330 		cond_local_irq_disable(regs);
1331 		return;
1332 	}
1333 #endif
1334 
1335 	/* This should not happen. */
1336 	if (WARN(cr0 & X86_CR0_TS, "CR0.TS was set")) {
1337 		/* Try to fix it up and carry on. */
1338 		write_cr0(cr0 & ~X86_CR0_TS);
1339 	} else {
1340 		/*
1341 		 * Something terrible happened, and we're better off trying
1342 		 * to kill the task than getting stuck in a never-ending
1343 		 * loop of #NM faults.
1344 		 */
1345 		die("unexpected #NM exception", regs, 0);
1346 	}
1347 }
1348 
1349 #ifdef CONFIG_INTEL_TDX_GUEST
1350 
1351 #define VE_FAULT_STR "VE fault"
1352 
1353 static void ve_raise_fault(struct pt_regs *regs, long error_code)
1354 {
1355 	if (user_mode(regs)) {
1356 		gp_user_force_sig_segv(regs, X86_TRAP_VE, error_code, VE_FAULT_STR);
1357 		return;
1358 	}
1359 
1360 	if (gp_try_fixup_and_notify(regs, X86_TRAP_VE, error_code, VE_FAULT_STR))
1361 		return;
1362 
1363 	die_addr(VE_FAULT_STR, regs, error_code, 0);
1364 }
1365 
1366 /*
1367  * Virtualization Exceptions (#VE) are delivered to TDX guests due to
1368  * specific guest actions which may happen in either user space or the
1369  * kernel:
1370  *
1371  *  * Specific instructions (WBINVD, for example)
1372  *  * Specific MSR accesses
1373  *  * Specific CPUID leaf accesses
1374  *  * Access to specific guest physical addresses
1375  *
1376  * In the settings that Linux will run in, virtualization exceptions are
1377  * never generated on accesses to normal, TD-private memory that has been
1378  * accepted (by BIOS or with tdx_enc_status_changed()).
1379  *
1380  * Syscall entry code has a critical window where the kernel stack is not
1381  * yet set up. Any exception in this window leads to hard to debug issues
1382  * and can be exploited for privilege escalation. Exceptions in the NMI
1383  * entry code also cause issues. Returning from the exception handler with
1384  * IRET will re-enable NMIs and nested NMI will corrupt the NMI stack.
1385  *
1386  * For these reasons, the kernel avoids #VEs during the syscall gap and
1387  * the NMI entry code. Entry code paths do not access TD-shared memory,
1388  * MMIO regions, use #VE triggering MSRs, instructions, or CPUID leaves
1389  * that might generate #VE. VMM can remove memory from TD at any point,
1390  * but access to unaccepted (or missing) private memory leads to VM
1391  * termination, not to #VE.
1392  *
1393  * Similarly to page faults and breakpoints, #VEs are allowed in NMI
1394  * handlers once the kernel is ready to deal with nested NMIs.
1395  *
1396  * During #VE delivery, all interrupts, including NMIs, are blocked until
1397  * TDGETVEINFO is called. It prevents #VE nesting until the kernel reads
1398  * the VE info.
1399  *
1400  * If a guest kernel action which would normally cause a #VE occurs in
1401  * the interrupt-disabled region before TDGETVEINFO, a #DF (fault
1402  * exception) is delivered to the guest which will result in an oops.
1403  *
1404  * The entry code has been audited carefully for following these expectations.
1405  * Changes in the entry code have to be audited for correctness vs. this
1406  * aspect. Similarly to #PF, #VE in these places will expose kernel to
1407  * privilege escalation or may lead to random crashes.
1408  */
1409 DEFINE_IDTENTRY(exc_virtualization_exception)
1410 {
1411 	struct ve_info ve;
1412 
1413 	/*
1414 	 * NMIs/Machine-checks/Interrupts will be in a disabled state
1415 	 * till TDGETVEINFO TDCALL is executed. This ensures that VE
1416 	 * info cannot be overwritten by a nested #VE.
1417 	 */
1418 	tdx_get_ve_info(&ve);
1419 
1420 	cond_local_irq_enable(regs);
1421 
1422 	/*
1423 	 * If tdx_handle_virt_exception() could not process
1424 	 * it successfully, treat it as #GP(0) and handle it.
1425 	 */
1426 	if (!tdx_handle_virt_exception(regs, &ve))
1427 		ve_raise_fault(regs, 0);
1428 
1429 	cond_local_irq_disable(regs);
1430 }
1431 
1432 #endif
1433 
1434 #ifdef CONFIG_X86_32
1435 DEFINE_IDTENTRY_SW(iret_error)
1436 {
1437 	local_irq_enable();
1438 	if (notify_die(DIE_TRAP, "iret exception", regs, 0,
1439 			X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
1440 		do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, 0,
1441 			ILL_BADSTK, (void __user *)NULL);
1442 	}
1443 	local_irq_disable();
1444 }
1445 #endif
1446 
1447 void __init trap_init(void)
1448 {
1449 	/* Init cpu_entry_area before IST entries are set up */
1450 	setup_cpu_entry_areas();
1451 
1452 	/* Init GHCB memory pages when running as an SEV-ES guest */
1453 	sev_es_init_vc_handling();
1454 
1455 	/* Initialize TSS before setting up traps so ISTs work */
1456 	cpu_init_exception_handling();
1457 	/* Setup traps as cpu_init() might #GP */
1458 	idt_setup_traps();
1459 	cpu_init();
1460 }
1461