2 * SPDX-License-Identifier: BSD-4-Clause
4 * Copyright (C) 1994, David Greenman
5 * Copyright (c) 1990, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * the University of Utah, and William Jolitz.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
46 * AMD64 Trap and System call handling
49 #include "opt_clock.h"
50 #include "opt_compat.h"
52 #include "opt_hwpmc_hooks.h"
55 #include "opt_stack.h"
57 #include <sys/param.h>
59 #include <sys/systm.h>
61 #include <sys/pioctl.h>
62 #include <sys/ptrace.h>
64 #include <sys/kernel.h>
67 #include <sys/mutex.h>
68 #include <sys/resourcevar.h>
69 #include <sys/signalvar.h>
70 #include <sys/syscall.h>
71 #include <sys/sysctl.h>
72 #include <sys/sysent.h>
74 #include <sys/vmmeter.h>
76 #include <sys/pmckern.h>
77 PMC_SOFT_DEFINE( , , page_fault, all);
78 PMC_SOFT_DEFINE( , , page_fault, read);
79 PMC_SOFT_DEFINE( , , page_fault, write);
83 #include <vm/vm_param.h>
85 #include <vm/vm_kern.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_extern.h>
90 #include <machine/cpu.h>
91 #include <machine/intr_machdep.h>
93 #include <machine/md_var.h>
94 #include <machine/pcb.h>
96 #include <machine/smp.h>
98 #include <machine/stack.h>
99 #include <machine/trap.h>
100 #include <machine/tss.h>
103 #include <sys/dtrace_bsd.h>
106 extern inthand_t IDTVEC(bpt), IDTVEC(bpt_pti), IDTVEC(dbg),
107 IDTVEC(fast_syscall), IDTVEC(fast_syscall_pti), IDTVEC(fast_syscall32),
108 IDTVEC(int0x80_syscall_pti), IDTVEC(int0x80_syscall);
110 void __noinline trap(struct trapframe *frame);
111 void trap_check(struct trapframe *frame);
112 void dblfault_handler(struct trapframe *frame);
114 static int trap_pfault(struct trapframe *, int);
115 static void trap_fatal(struct trapframe *, vm_offset_t);
117 static bool trap_user_dtrace(struct trapframe *,
118 int (**hook)(struct trapframe *));
121 static const char UNKNOWN[] = "unknown";
122 static const char *const trap_msg[] = {
123 [0] = UNKNOWN, /* unused */
124 [T_PRIVINFLT] = "privileged instruction fault",
125 [2] = UNKNOWN, /* unused */
126 [T_BPTFLT] = "breakpoint instruction fault",
127 [4] = UNKNOWN, /* unused */
128 [5] = UNKNOWN, /* unused */
129 [T_ARITHTRAP] = "arithmetic trap",
130 [7] = UNKNOWN, /* unused */
131 [8] = UNKNOWN, /* unused */
132 [T_PROTFLT] = "general protection fault",
133 [T_TRCTRAP] = "debug exception",
134 [11] = UNKNOWN, /* unused */
135 [T_PAGEFLT] = "page fault",
136 [13] = UNKNOWN, /* unused */
137 [T_ALIGNFLT] = "alignment fault",
138 [15] = UNKNOWN, /* unused */
139 [16] = UNKNOWN, /* unused */
140 [17] = UNKNOWN, /* unused */
141 [T_DIVIDE] = "integer divide fault",
142 [T_NMI] = "non-maskable interrupt trap",
143 [T_OFLOW] = "overflow trap",
144 [T_BOUND] = "FPU bounds check fault",
145 [T_DNA] = "FPU device not available",
146 [T_DOUBLEFLT] = "double fault",
147 [T_FPOPFLT] = "FPU operand fetch fault",
148 [T_TSSFLT] = "invalid TSS fault",
149 [T_SEGNPFLT] = "segment not present fault",
150 [T_STKFLT] = "stack fault",
151 [T_MCHK] = "machine check trap",
152 [T_XMMFLT] = "SIMD floating-point exception",
153 [T_RESERVED] = "reserved (unknown) fault",
154 [31] = UNKNOWN, /* reserved */
155 [T_DTRACE_RET] = "DTrace pid return trap",
158 static int prot_fault_translation;
159 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RWTUN,
160 &prot_fault_translation, 0,
161 "Select signal to deliver on protection fault");
162 static int uprintf_signal;
163 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RWTUN,
165 "Print debugging information on trap signal to ctty");
168 * Control L1D flush on return from NMI.
170 * Tunable can be set to the following values:
171 * 0 - only enable flush on return from NMI if required by vmm.ko (default)
172 * >1 - always flush on return from NMI.
174 * Post-boot, the sysctl indicates if flushing is currently enabled.
176 int nmi_flush_l1d_sw;
177 SYSCTL_INT(_machdep, OID_AUTO, nmi_flush_l1d_sw, CTLFLAG_RWTUN,
178 &nmi_flush_l1d_sw, 0,
179 "Flush L1 Data Cache on NMI exit, software bhyve L1TF mitigation assist");
182 * Exception, fault, and trap interface to the FreeBSD kernel.
183 * This common code is called from assembly language IDT gate entry
184 * routines that prepare a suitable stack frame, and restore this
185 * frame after the exception has been processed.
189 trap(struct trapframe *frame)
194 register_t addr, dr6;
206 type = frame->tf_trapno;
209 /* Handler for NMI IPIs used for stopping CPUs. */
210 if (type == T_NMI && ipi_nmi_handler() == 0)
221 if (type == T_RESERVED) {
222 trap_fatal(frame, 0);
229 * CPU PMCs interrupt using an NMI. If the PMC module is
230 * active, pass the 'rip' value to the PMC module's interrupt
231 * handler. A non-zero return value from the handler means that
232 * the NMI was consumed by it and we can return immediately.
234 if (pmc_intr != NULL &&
235 (*pmc_intr)(frame) != 0)
240 if (stack_nmi_handler(frame) != 0)
245 if ((frame->tf_rflags & PSL_I) == 0) {
247 * Buggy application or kernel code has disabled
248 * interrupts and then trapped. Enabling interrupts
249 * now is wrong, but it is better than running with
250 * interrupts disabled until they are accidentally
253 if (TRAPF_USERMODE(frame))
255 "pid %ld (%s): trap %d with interrupts disabled\n",
256 (long)curproc->p_pid, curthread->td_name, type);
257 else if (type != T_NMI && type != T_BPTFLT &&
260 * XXX not quite right, since this may be for a
261 * multiple fault in user mode.
263 printf("kernel trap %d with interrupts disabled\n",
267 * We shouldn't enable interrupts while holding a
270 if (td->td_md.md_spinlock_count == 0)
275 if (TRAPF_USERMODE(frame)) {
279 td->td_frame = frame;
280 addr = frame->tf_rip;
281 if (td->td_cowgen != p->p_cowgen)
282 thread_cow_update(td);
285 case T_PRIVINFLT: /* privileged instruction fault */
290 case T_BPTFLT: /* bpt instruction fault */
292 if (trap_user_dtrace(frame, &dtrace_pid_probe_ptr))
301 case T_TRCTRAP: /* debug exception */
306 if ((dr6 & DBREG_DR6_BS) != 0) {
307 PROC_LOCK(td->td_proc);
308 if ((td->td_dbgflags & TDB_STEP) != 0) {
309 td->td_frame->tf_rflags &= ~PSL_T;
310 td->td_dbgflags &= ~TDB_STEP;
312 PROC_UNLOCK(td->td_proc);
316 case T_ARITHTRAP: /* arithmetic trap */
317 ucode = fputrap_x87();
323 case T_PROTFLT: /* general protection fault */
327 case T_STKFLT: /* stack fault */
328 case T_SEGNPFLT: /* segment not present fault */
332 case T_TSSFLT: /* invalid TSS fault */
340 case T_DOUBLEFLT: /* double fault */
346 case T_PAGEFLT: /* page fault */
348 * Emulator can take care about this trap?
350 if (*p->p_sysent->sv_trap != NULL &&
351 (*p->p_sysent->sv_trap)(td) == 0)
354 addr = frame->tf_addr;
355 signo = trap_pfault(frame, TRUE);
360 if (signo == SIGSEGV) {
362 } else if (prot_fault_translation == 0) {
364 * Autodetect. This check also covers
365 * the images without the ABI-tag ELF
368 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD &&
369 p->p_osrel >= P_OSREL_SIGSEGV) {
376 } else if (prot_fault_translation == 1) {
378 * Always compat mode.
384 * Always SIGSEGV mode.
391 case T_DIVIDE: /* integer divide fault */
398 nmi_handle_intr(type, frame);
402 case T_OFLOW: /* integer overflow fault */
407 case T_BOUND: /* bounds check fault */
413 /* transparent fault (due to context switch "late") */
414 KASSERT(PCB_USER_FPU(td->td_pcb),
415 ("kernel FPU ctx has leaked"));
419 case T_FPOPFLT: /* FPU operand fetch fault */
424 case T_XMMFLT: /* SIMD floating-point exception */
425 ucode = fputrap_sse();
432 (void)trap_user_dtrace(frame, &dtrace_return_probe_ptr);
439 KASSERT(cold || td->td_ucred != NULL,
440 ("kernel trap doesn't have ucred"));
442 case T_PAGEFLT: /* page fault */
443 (void) trap_pfault(frame, FALSE);
447 if (PCB_USER_FPU(td->td_pcb))
448 panic("Unregistered use of FPU in kernel");
452 case T_ARITHTRAP: /* arithmetic trap */
453 case T_XMMFLT: /* SIMD floating-point exception */
454 case T_FPOPFLT: /* FPU operand fetch fault */
456 * For now, supporting kernel handler
457 * registration for FPU traps is overkill.
459 trap_fatal(frame, 0);
462 case T_STKFLT: /* stack fault */
463 case T_PROTFLT: /* general protection fault */
464 case T_SEGNPFLT: /* segment not present fault */
465 if (td->td_intr_nesting_level != 0)
469 * Invalid segment selectors and out of bounds
470 * %rip's and %rsp's can be set up in user mode.
471 * This causes a fault in kernel mode when the
472 * kernel tries to return to user mode. We want
473 * to get this fault so that we can fix the
474 * problem here and not have to check all the
475 * selectors and pointers when the user changes
478 * In case of PTI, the IRETQ faulted while the
479 * kernel used the pti stack, and exception
480 * frame records %rsp value pointing to that
481 * stack. If we return normally to
482 * doreti_iret_fault, the trapframe is
483 * reconstructed on pti stack, and calltrap()
484 * called on it as well. Due to the very
485 * limited pti stack size, kernel does not
486 * survive for too long. Switch to the normal
487 * thread stack for the trap handling.
489 * Magic '5' is the number of qwords occupied by
490 * the hardware trap frame.
492 if (frame->tf_rip == (long)doreti_iret) {
493 frame->tf_rip = (long)doreti_iret_fault;
494 if ((PCPU_GET(curpmap)->pm_ucr3 !=
496 (frame->tf_rsp == (uintptr_t)PCPU_GET(
497 pti_rsp0) - 5 * sizeof(register_t))) {
498 frame->tf_rsp = PCPU_GET(rsp0) - 5 *
503 if (frame->tf_rip == (long)ld_ds) {
504 frame->tf_rip = (long)ds_load_fault;
507 if (frame->tf_rip == (long)ld_es) {
508 frame->tf_rip = (long)es_load_fault;
511 if (frame->tf_rip == (long)ld_fs) {
512 frame->tf_rip = (long)fs_load_fault;
515 if (frame->tf_rip == (long)ld_gs) {
516 frame->tf_rip = (long)gs_load_fault;
519 if (frame->tf_rip == (long)ld_gsbase) {
520 frame->tf_rip = (long)gsbase_load_fault;
523 if (frame->tf_rip == (long)ld_fsbase) {
524 frame->tf_rip = (long)fsbase_load_fault;
527 if (curpcb->pcb_onfault != NULL) {
528 frame->tf_rip = (long)curpcb->pcb_onfault;
535 * PSL_NT can be set in user mode and isn't cleared
536 * automatically when the kernel is entered. This
537 * causes a TSS fault when the kernel attempts to
538 * `iret' because the TSS link is uninitialized. We
539 * want to get this fault so that we can fix the
540 * problem here and not every time the kernel is
543 if (frame->tf_rflags & PSL_NT) {
544 frame->tf_rflags &= ~PSL_NT;
549 case T_TRCTRAP: /* debug exception */
550 /* Clear any pending debug events. */
555 * Ignore debug register exceptions due to
556 * accesses in the user's address space, which
557 * can happen under several conditions such as
558 * if a user sets a watchpoint on a buffer and
559 * then passes that buffer to a system call.
560 * We still want to get TRCTRAPS for addresses
561 * in kernel space because that is useful when
562 * debugging the kernel.
564 if (user_dbreg_trap(dr6))
568 * Malicious user code can configure a debug
569 * register watchpoint to trap on data access
570 * to the top of stack and then execute 'pop
571 * %ss; int 3'. Due to exception deferral for
572 * 'pop %ss', the CPU will not interrupt 'int
573 * 3' to raise the DB# exception for the debug
574 * register but will postpone the DB# until
575 * execution of the first instruction of the
576 * BP# handler (in kernel mode). Normally the
577 * previous check would ignore DB# exceptions
578 * for watchpoints on user addresses raised in
579 * kernel mode. However, some CPU errata
580 * include cases where DB# exceptions do not
581 * properly set bits in %dr6, e.g. Haswell
582 * HSD23 and Skylake-X SKZ24.
584 * A deferred DB# can also be raised on the
585 * first instructions of system call entry
586 * points or single-step traps via similar use
587 * of 'pop %ss' or 'mov xxx, %ss'.
591 (uintptr_t)IDTVEC(fast_syscall_pti) ||
592 #ifdef COMPAT_FREEBSD32
594 (uintptr_t)IDTVEC(int0x80_syscall_pti) ||
596 frame->tf_rip == (uintptr_t)IDTVEC(bpt_pti))
600 (uintptr_t)IDTVEC(fast_syscall) ||
601 #ifdef COMPAT_FREEBSD32
603 (uintptr_t)IDTVEC(int0x80_syscall) ||
605 frame->tf_rip == (uintptr_t)IDTVEC(bpt))
608 if (frame->tf_rip == (uintptr_t)IDTVEC(dbg) ||
609 /* Needed for AMD. */
610 frame->tf_rip == (uintptr_t)IDTVEC(fast_syscall32))
613 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
617 * If KDB is enabled, let it handle the debugger trap.
618 * Otherwise, debugger traps "can't happen".
621 if (kdb_trap(type, dr6, frame))
628 nmi_handle_intr(type, frame);
633 trap_fatal(frame, 0);
637 /* Translate fault for emulators (e.g. Linux) */
638 if (*p->p_sysent->sv_transtrap != NULL)
639 signo = (*p->p_sysent->sv_transtrap)(signo, type);
641 ksiginfo_init_trap(&ksi);
642 ksi.ksi_signo = signo;
643 ksi.ksi_code = ucode;
644 ksi.ksi_trapno = type;
645 ksi.ksi_addr = (void *)addr;
646 if (uprintf_signal) {
647 uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
648 "addr 0x%lx rsp 0x%lx rip 0x%lx "
649 "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
650 p->p_pid, p->p_comm, signo, frame->tf_err, ucode, type,
651 addr, frame->tf_rsp, frame->tf_rip,
652 fubyte((void *)(frame->tf_rip + 0)),
653 fubyte((void *)(frame->tf_rip + 1)),
654 fubyte((void *)(frame->tf_rip + 2)),
655 fubyte((void *)(frame->tf_rip + 3)),
656 fubyte((void *)(frame->tf_rip + 4)),
657 fubyte((void *)(frame->tf_rip + 5)),
658 fubyte((void *)(frame->tf_rip + 6)),
659 fubyte((void *)(frame->tf_rip + 7)));
661 KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
662 trapsignal(td, &ksi);
666 KASSERT(PCB_USER_FPU(td->td_pcb),
667 ("Return from trap with kernel FPU ctx leaked"));
671 * Ensure that we ignore any DTrace-induced faults. This function cannot
672 * be instrumented, so it cannot generate such faults itself.
675 trap_check(struct trapframe *frame)
679 if (dtrace_trap_func != NULL &&
680 (*dtrace_trap_func)(frame, frame->tf_trapno) != 0)
687 trap_is_smap(struct trapframe *frame)
691 * A page fault on a userspace address is classified as
693 * - SMAP is supported;
694 * - kernel mode accessed present data page;
695 * - rflags.AC was cleared.
696 * Kernel must never access user space with rflags.AC cleared
697 * if SMAP is enabled.
699 return ((cpu_stdext_feature & CPUID_STDEXT_SMAP) != 0 &&
700 (frame->tf_err & (PGEX_P | PGEX_U | PGEX_I | PGEX_RSV)) ==
701 PGEX_P && (frame->tf_rflags & PSL_AC) == 0);
705 trap_is_pti(struct trapframe *frame)
708 return (PCPU_GET(curpmap)->pm_ucr3 != PMAP_NO_CR3 &&
709 pg_nx != 0 && (frame->tf_err & (PGEX_P | PGEX_W |
710 PGEX_U | PGEX_I)) == (PGEX_P | PGEX_U | PGEX_I) &&
711 (curpcb->pcb_saved_ucr3 & ~CR3_PCID_MASK) ==
712 (PCPU_GET(curpmap)->pm_cr3 & ~CR3_PCID_MASK));
716 trap_pfault(struct trapframe *frame, int usermode)
728 eva = frame->tf_addr;
730 if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
732 * Due to both processor errata and lazy TLB invalidation when
733 * access restrictions are removed from virtual pages, memory
734 * accesses that are allowed by the physical mapping layer may
735 * nonetheless cause one spurious page fault per virtual page.
736 * When the thread is executing a "no faulting" section that
737 * is bracketed by vm_fault_{disable,enable}_pagefaults(),
738 * every page fault is treated as a spurious page fault,
739 * unless it accesses the same virtual address as the most
740 * recent page fault within the same "no faulting" section.
742 if (td->td_md.md_spurflt_addr != eva ||
743 (td->td_pflags & TDP_RESETSPUR) != 0) {
745 * Do nothing to the TLB. A stale TLB entry is
746 * flushed automatically by a page fault.
748 td->td_md.md_spurflt_addr = eva;
749 td->td_pflags &= ~TDP_RESETSPUR;
754 * If we get a page fault while in a critical section, then
755 * it is most likely a fatal kernel page fault. The kernel
756 * is already going to panic trying to get a sleep lock to
757 * do the VM lookup, so just consider it a fatal trap so the
758 * kernel can print out a useful trap message and even get
761 * If we get a page fault while holding a non-sleepable
762 * lock, then it is most likely a fatal kernel page fault.
763 * If WITNESS is enabled, then it's going to whine about
764 * bogus LORs with various VM locks, so just skip to the
765 * fatal trap handling directly.
767 if (td->td_critnest != 0 ||
768 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
769 "Kernel page fault") != 0) {
770 trap_fatal(frame, eva);
774 va = trunc_page(eva);
775 if (va >= VM_MIN_KERNEL_ADDRESS) {
777 * Don't allow user-mode faults in kernel address space.
784 map = &p->p_vmspace->vm_map;
787 * When accessing a usermode address, kernel must be
788 * ready to accept the page fault, and provide a
789 * handling routine. Since accessing the address
790 * without the handler is a bug, do not try to handle
791 * it normally, and panic immediately.
793 * If SMAP is enabled, filter SMAP faults also,
794 * because illegal access might occur to the mapped
795 * user address, causing infinite loop.
797 if (!usermode && (td->td_intr_nesting_level != 0 ||
798 trap_is_smap(frame) || curpcb->pcb_onfault == NULL)) {
799 trap_fatal(frame, eva);
805 * If the trap was caused by errant bits in the PTE then panic.
807 if (frame->tf_err & PGEX_RSV) {
808 trap_fatal(frame, eva);
813 * User-mode protection key violation (PKU). May happen
814 * either from usermode or from kernel if copyin accessed
815 * key-protected mapping.
817 if ((frame->tf_err & PGEX_PK) != 0) {
818 if (eva > VM_MAXUSER_ADDRESS) {
819 trap_fatal(frame, eva);
822 rv = KERN_PROTECTION_FAILURE;
827 * If nx protection of the usermode portion of kernel page
828 * tables caused trap, panic.
830 if (usermode && trap_is_pti(frame))
831 panic("PTI: pid %d comm %s tf_err %#lx", p->p_pid,
832 p->p_comm, frame->tf_err);
835 * PGEX_I is defined only if the execute disable bit capability is
836 * supported and enabled.
838 if (frame->tf_err & PGEX_W)
839 ftype = VM_PROT_WRITE;
840 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
841 ftype = VM_PROT_EXECUTE;
843 ftype = VM_PROT_READ;
845 /* Fault in the page. */
846 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
847 if (rv == KERN_SUCCESS) {
849 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
850 PMC_SOFT_CALL_TF( , , page_fault, all, frame);
851 if (ftype == VM_PROT_READ)
852 PMC_SOFT_CALL_TF( , , page_fault, read,
855 PMC_SOFT_CALL_TF( , , page_fault, write,
863 if (td->td_intr_nesting_level == 0 &&
864 curpcb->pcb_onfault != NULL) {
865 frame->tf_rip = (long)curpcb->pcb_onfault;
868 trap_fatal(frame, eva);
871 return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
875 trap_fatal(frame, eva)
876 struct trapframe *frame;
881 struct soft_segment_descriptor softseg;
886 code = frame->tf_err;
887 type = frame->tf_trapno;
888 sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
891 printf("\n\nFatal trap %d: %s while in %s mode\n", type,
892 type < nitems(trap_msg) ? trap_msg[type] : UNKNOWN,
893 TRAPF_USERMODE(frame) ? "user" : "kernel");
895 /* two separate prints in case of a trap on an unmapped page */
896 printf("cpuid = %d; ", PCPU_GET(cpuid));
897 printf("apic id = %02x\n", PCPU_GET(apic_id));
899 if (type == T_PAGEFLT) {
900 printf("fault virtual address = 0x%lx\n", eva);
901 printf("fault code = %s %s %s%s%s, %s\n",
902 code & PGEX_U ? "user" : "supervisor",
903 code & PGEX_W ? "write" : "read",
904 code & PGEX_I ? "instruction" : "data",
905 code & PGEX_PK ? " prot key" : "",
906 code & PGEX_SGX ? " SGX" : "",
907 code & PGEX_RSV ? "reserved bits in PTE" :
908 code & PGEX_P ? "protection violation" : "page not present");
910 printf("instruction pointer = 0x%lx:0x%lx\n",
911 frame->tf_cs & 0xffff, frame->tf_rip);
912 ss = frame->tf_ss & 0xffff;
913 printf("stack pointer = 0x%x:0x%lx\n", ss, frame->tf_rsp);
914 printf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
915 printf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
916 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
917 printf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
918 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
920 printf("processor eflags = ");
921 if (frame->tf_rflags & PSL_T)
922 printf("trace trap, ");
923 if (frame->tf_rflags & PSL_I)
924 printf("interrupt enabled, ");
925 if (frame->tf_rflags & PSL_NT)
926 printf("nested task, ");
927 if (frame->tf_rflags & PSL_RF)
929 printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
930 printf("current process = %d (%s)\n",
931 curproc->p_pid, curthread->td_name);
934 if (debugger_on_trap) {
935 kdb_why = KDB_WHY_TRAP;
936 handled = kdb_trap(type, 0, frame);
937 kdb_why = KDB_WHY_UNSET;
942 printf("trap number = %d\n", type);
943 panic("%s", type < nitems(trap_msg) ? trap_msg[type] :
944 "unknown/reserved trap");
949 * Invoke a userspace DTrace hook. The hook pointer is cleared when no
950 * userspace probes are enabled, so we must synchronize with DTrace to ensure
951 * that a trapping thread is able to call the hook before it is cleared.
954 trap_user_dtrace(struct trapframe *frame, int (**hookp)(struct trapframe *))
956 int (*hook)(struct trapframe *);
958 hook = (int (*)(struct trapframe *))atomic_load_ptr(hookp);
961 return ((hook)(frame) == 0);
967 * Double fault handler. Called when a fault occurs while writing
968 * a frame for a trap/exception onto the stack. This usually occurs
969 * when the stack overflows (such is the case with infinite recursion,
973 dblfault_handler(struct trapframe *frame)
976 if (dtrace_doubletrap_func != NULL)
977 (*dtrace_doubletrap_func)();
979 printf("\nFatal double fault\n"
980 "rip %#lx rsp %#lx rbp %#lx\n"
981 "rax %#lx rdx %#lx rbx %#lx\n"
982 "rcx %#lx rsi %#lx rdi %#lx\n"
983 "r8 %#lx r9 %#lx r10 %#lx\n"
984 "r11 %#lx r12 %#lx r13 %#lx\n"
985 "r14 %#lx r15 %#lx rflags %#lx\n"
986 "cs %#lx ss %#lx ds %#hx es %#hx fs %#hx gs %#hx\n"
987 "fsbase %#lx gsbase %#lx kgsbase %#lx\n",
988 frame->tf_rip, frame->tf_rsp, frame->tf_rbp,
989 frame->tf_rax, frame->tf_rdx, frame->tf_rbx,
990 frame->tf_rcx, frame->tf_rdi, frame->tf_rsi,
991 frame->tf_r8, frame->tf_r9, frame->tf_r10,
992 frame->tf_r11, frame->tf_r12, frame->tf_r13,
993 frame->tf_r14, frame->tf_r15, frame->tf_rflags,
994 frame->tf_cs, frame->tf_ss, frame->tf_ds, frame->tf_es,
995 frame->tf_fs, frame->tf_gs,
996 rdmsr(MSR_FSBASE), rdmsr(MSR_GSBASE), rdmsr(MSR_KGSBASE));
998 /* two separate prints in case of a trap on an unmapped page */
999 printf("cpuid = %d; ", PCPU_GET(cpuid));
1000 printf("apic id = %02x\n", PCPU_GET(apic_id));
1002 panic("double fault");
1005 static int __noinline
1006 cpu_fetch_syscall_args_fallback(struct thread *td, struct syscall_args *sa)
1009 struct trapframe *frame;
1012 int reg, regcnt, error;
1015 frame = td->td_frame;
1019 sa->code = frame->tf_rax;
1021 if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
1022 sa->code = frame->tf_rdi;
1027 if (sa->code >= p->p_sysent->sv_size)
1028 sa->callp = &p->p_sysent->sv_table[0];
1030 sa->callp = &p->p_sysent->sv_table[sa->code];
1032 sa->narg = sa->callp->sy_narg;
1033 KASSERT(sa->narg <= nitems(sa->args), ("Too many syscall arguments!"));
1034 argp = &frame->tf_rdi;
1036 memcpy(sa->args, argp, sizeof(sa->args[0]) * NARGREGS);
1037 if (sa->narg > regcnt) {
1038 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1039 error = copyin(params, &sa->args[regcnt],
1040 (sa->narg - regcnt) * sizeof(sa->args[0]));
1041 if (__predict_false(error != 0))
1045 td->td_retval[0] = 0;
1046 td->td_retval[1] = frame->tf_rdx;
1052 cpu_fetch_syscall_args(struct thread *td)
1055 struct trapframe *frame;
1056 struct syscall_args *sa;
1059 frame = td->td_frame;
1062 sa->code = frame->tf_rax;
1064 if (__predict_false(sa->code == SYS_syscall ||
1065 sa->code == SYS___syscall ||
1066 sa->code >= p->p_sysent->sv_size))
1067 return (cpu_fetch_syscall_args_fallback(td, sa));
1069 sa->callp = &p->p_sysent->sv_table[sa->code];
1070 sa->narg = sa->callp->sy_narg;
1071 KASSERT(sa->narg <= nitems(sa->args), ("Too many syscall arguments!"));
1073 if (__predict_false(sa->narg > NARGREGS))
1074 return (cpu_fetch_syscall_args_fallback(td, sa));
1076 memcpy(sa->args, &frame->tf_rdi, sizeof(sa->args[0]) * NARGREGS);
1078 td->td_retval[0] = 0;
1079 td->td_retval[1] = frame->tf_rdx;
1084 #include "../../kern/subr_syscall.c"
1086 static void (*syscall_ret_l1d_flush)(void);
1087 int syscall_ret_l1d_flush_mode;
1093 wrmsr(MSR_IA32_FLUSH_CMD, IA32_FLUSH_CMD_L1D);
1096 static void __inline
1097 amd64_syscall_ret_flush_l1d_inline(int error)
1101 if (error != 0 && error != EEXIST && error != EAGAIN &&
1102 error != EXDEV && error != ENOENT && error != ENOTCONN &&
1103 error != EINPROGRESS) {
1104 p = syscall_ret_l1d_flush;
1111 amd64_syscall_ret_flush_l1d(int error)
1114 amd64_syscall_ret_flush_l1d_inline(error);
1118 amd64_syscall_ret_flush_l1d_recalc(void)
1122 l1d_hw = (cpu_stdext_feature3 & CPUID_STDEXT3_L1D_FLUSH) != 0;
1124 switch (syscall_ret_l1d_flush_mode) {
1126 syscall_ret_l1d_flush = NULL;
1129 syscall_ret_l1d_flush = l1d_hw ? flush_l1d_hw :
1133 syscall_ret_l1d_flush = l1d_hw ? flush_l1d_hw : NULL;
1136 syscall_ret_l1d_flush = flush_l1d_sw_abi;
1139 syscall_ret_l1d_flush_mode = 1;
1145 machdep_syscall_ret_flush_l1d(SYSCTL_HANDLER_ARGS)
1149 val = syscall_ret_l1d_flush_mode;
1150 error = sysctl_handle_int(oidp, &val, 0, req);
1151 if (error != 0 || req->newptr == NULL)
1153 syscall_ret_l1d_flush_mode = val;
1154 amd64_syscall_ret_flush_l1d_recalc();
1157 SYSCTL_PROC(_machdep, OID_AUTO, syscall_ret_flush_l1d, CTLTYPE_INT |
1158 CTLFLAG_RWTUN | CTLFLAG_NOFETCH | CTLFLAG_MPSAFE, NULL, 0,
1159 machdep_syscall_ret_flush_l1d, "I",
1160 "Flush L1D on syscall return with error (0 - off, 1 - on, "
1161 "2 - use hw only, 3 - use sw only");
1165 * System call handler for native binaries. The trap frame is already
1166 * set up by the assembler trampoline and a pointer to it is saved in
1170 amd64_syscall(struct thread *td, int traced)
1176 if (!TRAPF_USERMODE(td->td_frame)) {
1181 error = syscallenter(td);
1186 if (__predict_false(traced)) {
1187 td->td_frame->tf_rflags &= ~PSL_T;
1188 ksiginfo_init_trap(&ksi);
1189 ksi.ksi_signo = SIGTRAP;
1190 ksi.ksi_code = TRAP_TRACE;
1191 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
1192 trapsignal(td, &ksi);
1195 KASSERT(PCB_USER_FPU(td->td_pcb),
1196 ("System call %s returning with kernel FPU ctx leaked",
1197 syscallname(td->td_proc, td->td_sa.code)));
1198 KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
1199 ("System call %s returning with mangled pcb_save",
1200 syscallname(td->td_proc, td->td_sa.code)));
1201 KASSERT(pmap_not_in_di(),
1202 ("System call %s returning with leaked invl_gen %lu",
1203 syscallname(td->td_proc, td->td_sa.code),
1204 td->td_md.md_invl_gen.gen));
1206 syscallret(td, error);
1209 * If the user-supplied value of %rip is not a canonical
1210 * address, then some CPUs will trigger a ring 0 #GP during
1211 * the sysret instruction. However, the fault handler would
1212 * execute in ring 0 with the user's %gs and %rsp which would
1213 * not be safe. Instead, use the full return path which
1214 * catches the problem safely.
1216 if (__predict_false(td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS))
1217 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
1219 amd64_syscall_ret_flush_l1d_inline(error);