X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/0c530ab8987f0ae6a1a3d9284f40182b88852816..008676633c2ad2c325837c2b64915f7ded690a8f:/osfmk/i386/trap.c diff --git a/osfmk/i386/trap.c b/osfmk/i386/trap.c index 7c9aa04f1..7924f4f7d 100644 --- a/osfmk/i386/trap.c +++ b/osfmk/i386/trap.c @@ -1,60 +1,65 @@ /* - * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2016 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * - * The contents of this file constitute Original Code as defined in and - * are subject to the Apple Public Source License Version 1.1 (the - * "License"). You may not use this file except in compliance with the - * License. Please obtain a copy of the License at - * http://www.apple.com/publicsource and read it before using this file. + * This file contains Original Code and/or Modifications of Original Code + * as defined in and that are subject to the Apple Public Source License + * Version 2.0 (the 'License'). You may not use this file except in + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. * - * This Original Code and all software distributed under the License are - * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. + * + * The Original Code and all software distributed under the License are + * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the - * License for the specific language governing rights and limitations - * under the License. + * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. + * Please see the License for the specific language governing rights and + * limitations under the License. * - * @APPLE_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* - * @OSF_COPYRIGHT@ - */ +* @OSF_COPYRIGHT@ +*/ /* - * Mach Operating System - * Copyright (c) 1991,1990,1989,1988 Carnegie Mellon University - * All Rights Reserved. - * - * Permission to use, copy, modify and distribute this software and its - * documentation is hereby granted, provided that both the copyright - * notice and this permission notice appear in all copies of the - * software, derivative works or modified versions, and any portions - * thereof, and that both notices appear in supporting documentation. - * - * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" - * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR - * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. - * - * Carnegie Mellon requests users of this software to return to - * - * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU - * School of Computer Science - * Carnegie Mellon University - * Pittsburgh PA 15213-3890 - * - * any improvements or extensions that they make and grant Carnegie Mellon - * the rights to redistribute these changes. - */ +* Mach Operating System +* Copyright (c) 1991,1990,1989,1988 Carnegie Mellon University +* All Rights Reserved. +* +* Permission to use, copy, modify and distribute this software and its +* documentation is hereby granted, provided that both the copyright +* notice and this permission notice appear in all copies of the +* software, derivative works or modified versions, and any portions +* thereof, and that both notices appear in supporting documentation. +* +* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" +* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR +* ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. +* +* Carnegie Mellon requests users of this software to return to +* +* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU +* School of Computer Science +* Carnegie Mellon University +* Pittsburgh PA 15213-3890 +* +* any improvements or extensions that they make and grant Carnegie Mellon +* the rights to redistribute these changes. +*/ /* - */ +*/ + /* - * Hardware trap/fault handler. +* Hardware trap/fault handler. */ -#include -#include #include #include @@ -64,6 +69,7 @@ #include #include #include /* panic_io_port_read() */ +#include #include #include @@ -82,187 +88,141 @@ #include #include #include - +#include +#if CONFIG_TELEMETRY +#include +#endif #include - -#if MACH_KGDB -#include -#endif /* MACH_KGDB */ - -#if MACH_KDB -#include -#include -#include -#include -#include -#endif /* MACH_KDB */ +#include #include -#include #include #include #include +#if CONFIG_MCA #include +#endif #include +#include +#include +#include + +extern void throttle_lowpri_io(int); +extern void kprint_state(x86_saved_state64_t *saved_state); + /* * Forward declarations */ static void user_page_fault_continue(kern_return_t kret); -static void panic_trap(x86_saved_state32_t *saved_state); -static void set_recovery_ip(x86_saved_state32_t *saved_state, vm_offset_t ip); +static void panic_trap(x86_saved_state64_t *saved_state, uint32_t pl, kern_return_t fault_result); +static void set_recovery_ip(x86_saved_state64_t *saved_state, vm_offset_t ip); + +volatile perfCallback perfTrapHook = NULL; /* Pointer to CHUD trap hook routine */ -perfCallback perfTrapHook = NULL; /* Pointer to CHUD trap hook routine */ -perfCallback perfASTHook = NULL; /* Pointer to CHUD AST hook routine */ +#if CONFIG_DTRACE +/* See */ +perfCallback tempDTraceTrapHook = NULL; /* Pointer to DTrace fbt trap hook routine */ +extern boolean_t dtrace_tally_fault(user_addr_t); +#endif + +extern boolean_t pmap_smep_enabled; +extern boolean_t pmap_smap_enabled; + +__attribute__((noreturn)) void thread_syscall_return( kern_return_t ret) { thread_t thr_act = current_thread(); + boolean_t is_mach; + int code; + + pal_register_cache_state(thr_act, DIRTY); if (thread_is_64bit(thr_act)) { x86_saved_state64_t *regs; regs = USER_REGS64(thr_act); - if (kdebug_enable && ((regs->rax & SYSCALL_CLASS_MASK) == (SYSCALL_CLASS_MACH << SYSCALL_CLASS_SHIFT))) { + code = (int) (regs->rax & SYSCALL_NUMBER_MASK); + is_mach = (regs->rax & SYSCALL_CLASS_MASK) + == (SYSCALL_CLASS_MACH << SYSCALL_CLASS_SHIFT); + if (kdebug_enable && is_mach) { /* Mach trap */ - KERNEL_DEBUG_CONSTANT( - MACHDBG_CODE(DBG_MACH_EXCP_SC, ((int) (regs->rax & SYSCALL_NUMBER_MASK))) - | DBG_FUNC_END, - ret, 0, 0, 0, 0); + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + MACHDBG_CODE(DBG_MACH_EXCP_SC,code)|DBG_FUNC_END, + ret, 0, 0, 0, 0); } regs->rax = ret; - +#if DEBUG + if (is_mach) + DEBUG_KPRINT_SYSCALL_MACH( + "thread_syscall_return: 64-bit mach ret=%u\n", + ret); + else + DEBUG_KPRINT_SYSCALL_UNIX( + "thread_syscall_return: 64-bit unix ret=%u\n", + ret); +#endif } else { x86_saved_state32_t *regs; regs = USER_REGS32(thr_act); - if (kdebug_enable && ((int) regs->eax < 0)) { + code = ((int) regs->eax); + is_mach = (code < 0); + if (kdebug_enable && is_mach) { /* Mach trap */ - KERNEL_DEBUG_CONSTANT( - MACHDBG_CODE(DBG_MACH_EXCP_SC, -((int) regs->eax)) - | DBG_FUNC_END, - ret, 0, 0, 0, 0); + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + MACHDBG_CODE(DBG_MACH_EXCP_SC,-code)|DBG_FUNC_END, + ret, 0, 0, 0, 0); } regs->eax = ret; +#if DEBUG + if (is_mach) + DEBUG_KPRINT_SYSCALL_MACH( + "thread_syscall_return: 32-bit mach ret=%u\n", + ret); + else + DEBUG_KPRINT_SYSCALL_UNIX( + "thread_syscall_return: 32-bit unix ret=%u\n", + ret); +#endif } - thread_exception_return(); - /*NOTREACHED*/ -} - - -#if MACH_KDB -boolean_t debug_all_traps_with_kdb = FALSE; -extern struct db_watchpoint *db_watchpoint_list; -extern boolean_t db_watchpoints_inserted; -extern boolean_t db_breakpoints_inserted; - -void -thread_kdb_return(void) -{ - thread_t thr_act = current_thread(); - x86_saved_state_t *iss = USER_STATE(thr_act); + throttle_lowpri_io(1); - if (is_saved_state64(iss)) { - x86_saved_state64_t *regs; - - regs = saved_state64(iss); - - if (kdb_trap(regs->isf.trapno, (int)regs->isf.err, (void *)regs)) { - thread_exception_return(); - /*NOTREACHED*/ - } - - } else { - x86_saved_state32_t *regs; - - regs = saved_state32(iss); - - if (kdb_trap(regs->trapno, regs->err, (void *)regs)) { - thread_exception_return(); - /*NOTREACHED*/ - } - } + thread_exception_return(); + /*NOTREACHED*/ } -#endif /* MACH_KDB */ -void +static inline void user_page_fault_continue( kern_return_t kr) { thread_t thread = current_thread(); - x86_saved_state_t *regs = USER_STATE(thread); - ast_t *myast; - boolean_t intr; user_addr_t vaddr; -#if MACH_KDB - int err; - int trapno; -#endif - assert((is_saved_state32(regs) && !thread_is_64bit(thread)) || - (is_saved_state64(regs) && thread_is_64bit(thread))); - - if (thread_is_64bit(thread)) { - x86_saved_state64_t *uregs; + if (thread_is_64bit(thread)) { + x86_saved_state64_t *uregs; uregs = USER_REGS64(thread); -#if MACH_KDB - trapno = uregs->isf.trapno; - err = uregs->isf.err; -#endif vaddr = (user_addr_t)uregs->cr2; } else { x86_saved_state32_t *uregs; uregs = USER_REGS32(thread); -#if MACH_KDB - trapno = uregs->trapno; - err = uregs->err; -#endif vaddr = uregs->cr2; } - if ((kr == KERN_SUCCESS) || (kr == KERN_ABORTED)) { -#if MACH_KDB - if (!db_breakpoints_inserted) { - db_set_breakpoints(); - } - if (db_watchpoint_list && - db_watchpoints_inserted && - (err & T_PF_WRITE) && - db_find_watchpoint(thread->map, - (vm_offset_t)vaddr, - regs)) - kdb_trap(T_WATCHPOINT, 0, regs); -#endif /* MACH_KDB */ - intr = ml_set_interrupts_enabled(FALSE); - myast = ast_pending(); - while (*myast & AST_ALL) { - ast_taken(AST_ALL, intr); - ml_set_interrupts_enabled(FALSE); - myast = ast_pending(); - } - ml_set_interrupts_enabled(intr); - thread_exception_return(); - /*NOTREACHED*/ - } - -#if MACH_KDB - if (debug_all_traps_with_kdb && - kdb_trap(trapno, err, regs)) { - thread_exception_return(); - /*NOTREACHED*/ - } -#endif /* MACH_KDB */ + /* PAL debug hook */ + pal_dbg_page_fault( thread, vaddr, kr ); i386_exception(EXC_BAD_ACCESS, kr, vaddr); /*NOTREACHED*/ @@ -272,8 +232,8 @@ user_page_fault_continue( * Fault recovery in copyin/copyout routines. */ struct recovery { - uint32_t fault_addr; - uint32_t recover_addr; + uintptr_t fault_addr; + uintptr_t recover_addr; }; extern struct recovery recover_table[]; @@ -282,112 +242,368 @@ extern struct recovery recover_table_end[]; const char * trap_type[] = {TRAP_NAMES}; unsigned TRAP_TYPES = sizeof(trap_type)/sizeof(trap_type[0]); +extern void PE_incoming_interrupt(int interrupt); + +#if defined(__x86_64__) && DEBUG +void +kprint_state(x86_saved_state64_t *saved_state) +{ + kprintf("current_cpu_datap() 0x%lx\n", (uintptr_t)current_cpu_datap()); + kprintf("Current GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_GS_BASE)); + kprintf("Kernel GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_KERNEL_GS_BASE)); + kprintf("state at 0x%lx:\n", (uintptr_t) saved_state); + + kprintf(" rdi 0x%llx\n", saved_state->rdi); + kprintf(" rsi 0x%llx\n", saved_state->rsi); + kprintf(" rdx 0x%llx\n", saved_state->rdx); + kprintf(" r10 0x%llx\n", saved_state->r10); + kprintf(" r8 0x%llx\n", saved_state->r8); + kprintf(" r9 0x%llx\n", saved_state->r9); + + kprintf(" cr2 0x%llx\n", saved_state->cr2); + kprintf("real cr2 0x%lx\n", get_cr2()); + kprintf(" r15 0x%llx\n", saved_state->r15); + kprintf(" r14 0x%llx\n", saved_state->r14); + kprintf(" r13 0x%llx\n", saved_state->r13); + kprintf(" r12 0x%llx\n", saved_state->r12); + kprintf(" r11 0x%llx\n", saved_state->r11); + kprintf(" rbp 0x%llx\n", saved_state->rbp); + kprintf(" rbx 0x%llx\n", saved_state->rbx); + kprintf(" rcx 0x%llx\n", saved_state->rcx); + kprintf(" rax 0x%llx\n", saved_state->rax); + + kprintf(" gs 0x%x\n", saved_state->gs); + kprintf(" fs 0x%x\n", saved_state->fs); + + kprintf(" isf.trapno 0x%x\n", saved_state->isf.trapno); + kprintf(" isf._pad 0x%x\n", saved_state->isf._pad); + kprintf(" isf.trapfn 0x%llx\n", saved_state->isf.trapfn); + kprintf(" isf.err 0x%llx\n", saved_state->isf.err); + kprintf(" isf.rip 0x%llx\n", saved_state->isf.rip); + kprintf(" isf.cs 0x%llx\n", saved_state->isf.cs); + kprintf(" isf.rflags 0x%llx\n", saved_state->isf.rflags); + kprintf(" isf.rsp 0x%llx\n", saved_state->isf.rsp); + kprintf(" isf.ss 0x%llx\n", saved_state->isf.ss); +} +#endif + + +/* + * Non-zero indicates latency assert is enabled and capped at valued + * absolute time units. + */ + +uint64_t interrupt_latency_cap = 0; +boolean_t ilat_assert = FALSE; + +void +interrupt_latency_tracker_setup(void) { + uint32_t ilat_cap_us; + if (PE_parse_boot_argn("interrupt_latency_cap_us", &ilat_cap_us, sizeof(ilat_cap_us))) { + interrupt_latency_cap = ilat_cap_us * NSEC_PER_USEC; + nanoseconds_to_absolutetime(interrupt_latency_cap, &interrupt_latency_cap); + } else { + interrupt_latency_cap = LockTimeOut; + } + PE_parse_boot_argn("-interrupt_latency_assert_enable", &ilat_assert, sizeof(ilat_assert)); +} + +void interrupt_reset_latency_stats(void) { + uint32_t i; + for (i = 0; i < real_ncpus; i++) { + cpu_data_ptr[i]->cpu_max_observed_int_latency = + cpu_data_ptr[i]->cpu_max_observed_int_latency_vector = 0; + } +} + +void interrupt_populate_latency_stats(char *buf, unsigned bufsize) { + uint32_t i, tcpu = ~0; + uint64_t cur_max = 0; + + for (i = 0; i < real_ncpus; i++) { + if (cur_max < cpu_data_ptr[i]->cpu_max_observed_int_latency) { + cur_max = cpu_data_ptr[i]->cpu_max_observed_int_latency; + tcpu = i; + } + } + + if (tcpu < real_ncpus) + snprintf(buf, bufsize, "0x%x 0x%x 0x%llx", tcpu, cpu_data_ptr[tcpu]->cpu_max_observed_int_latency_vector, cpu_data_ptr[tcpu]->cpu_max_observed_int_latency); +} + +uint32_t interrupt_timer_coalescing_enabled = 1; +uint64_t interrupt_coalesced_timers; + +/* + * Handle interrupts: + * - local APIC interrupts (IPIs, timers, etc) are handled by the kernel, + * - device interrupts go to the platform expert. + */ +void +interrupt(x86_saved_state_t *state) +{ + uint64_t rip; + uint64_t rsp; + int interrupt_num; + boolean_t user_mode = FALSE; + int ipl; + int cnum = cpu_number(); + cpu_data_t *cdp = cpu_data_ptr[cnum]; + int itype = 0; + + if (is_saved_state64(state) == TRUE) { + x86_saved_state64_t *state64; + + state64 = saved_state64(state); + rip = state64->isf.rip; + rsp = state64->isf.rsp; + interrupt_num = state64->isf.trapno; +#ifdef __x86_64__ + if(state64->isf.cs & 0x03) +#endif + user_mode = TRUE; + } else { + x86_saved_state32_t *state32; + + state32 = saved_state32(state); + if (state32->cs & 0x03) + user_mode = TRUE; + rip = state32->eip; + rsp = state32->uesp; + interrupt_num = state32->trapno; + } + + if (cpu_data_ptr[cnum]->lcpu.package->num_idle == topoParms.nLThreadsPerPackage) + cpu_data_ptr[cnum]->cpu_hwIntpexits[interrupt_num]++; + + if (interrupt_num == (LAPIC_DEFAULT_INTERRUPT_BASE + LAPIC_INTERPROCESSOR_INTERRUPT)) + itype = 1; + else if (interrupt_num == (LAPIC_DEFAULT_INTERRUPT_BASE + LAPIC_TIMER_INTERRUPT)) + itype = 2; + else + itype = 3; + + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_START, + interrupt_num, + (user_mode ? rip : VM_KERNEL_UNSLIDE(rip)), + user_mode, itype, 0); + + SCHED_STATS_INTERRUPT(current_processor()); + +#if CONFIG_TELEMETRY + if (telemetry_needs_record) { + telemetry_mark_curthread(user_mode); + } +#endif + + ipl = get_preemption_level(); + + /* + * Handle local APIC interrupts + * else call platform expert for devices. + */ + if (!lapic_interrupt(interrupt_num, state)) { + PE_incoming_interrupt(interrupt_num); + } + + if (__improbable(get_preemption_level() != ipl)) { + panic("Preemption level altered by interrupt vector 0x%x: initial 0x%x, final: 0x%x\n", interrupt_num, ipl, get_preemption_level()); + } + + + if (__improbable(cdp->cpu_nested_istack)) { + cdp->cpu_nested_istack_events++; + } + else { + uint64_t ctime = mach_absolute_time(); + uint64_t int_latency = ctime - cdp->cpu_int_event_time; + uint64_t esdeadline, ehdeadline; + /* Attempt to process deferred timers in the context of + * this interrupt, unless interrupt time has already exceeded + * TCOAL_ILAT_THRESHOLD. + */ +#define TCOAL_ILAT_THRESHOLD (30000ULL) + + if ((int_latency < TCOAL_ILAT_THRESHOLD) && + interrupt_timer_coalescing_enabled) { + esdeadline = cdp->rtclock_timer.queue.earliest_soft_deadline; + ehdeadline = cdp->rtclock_timer.deadline; + if ((ctime >= esdeadline) && (ctime < ehdeadline)) { + interrupt_coalesced_timers++; + TCOAL_DEBUG(0x88880000 | DBG_FUNC_START, ctime, esdeadline, ehdeadline, interrupt_coalesced_timers, 0); + rtclock_intr(state); + TCOAL_DEBUG(0x88880000 | DBG_FUNC_END, ctime, esdeadline, interrupt_coalesced_timers, 0, 0); + } else { + TCOAL_DEBUG(0x77770000, ctime, cdp->rtclock_timer.queue.earliest_soft_deadline, cdp->rtclock_timer.deadline, interrupt_coalesced_timers, 0); + } + } + + if (__improbable(ilat_assert && (int_latency > interrupt_latency_cap) && !machine_timeout_suspended())) { + panic("Interrupt vector 0x%x exceeded interrupt latency threshold, 0x%llx absolute time delta, prior signals: 0x%x, current signals: 0x%x", interrupt_num, int_latency, cdp->cpu_prior_signals, cdp->cpu_signals); + } + + if (__improbable(int_latency > cdp->cpu_max_observed_int_latency)) { + cdp->cpu_max_observed_int_latency = int_latency; + cdp->cpu_max_observed_int_latency_vector = interrupt_num; + } + } + + /* + * Having serviced the interrupt first, look at the interrupted stack depth. + */ + if (!user_mode) { + uint64_t depth = cdp->cpu_kernel_stack + + sizeof(struct x86_kernel_state) + + sizeof(struct i386_exception_link *) + - rsp; + if (__improbable(depth > kernel_stack_depth_max)) { + kernel_stack_depth_max = (vm_offset_t)depth; + KERNEL_DEBUG_CONSTANT( + MACHDBG_CODE(DBG_MACH_SCHED, MACH_STACK_DEPTH), + (long) depth, (long) VM_KERNEL_UNSLIDE(rip), 0, 0, 0); + } + } + + if (cnum == master_cpu) + ml_entropy_collect(); + + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_END, + interrupt_num, 0, 0, 0, 0); + + assert(ml_get_interrupts_enabled() == FALSE); +} + static inline void reset_dr7(void) { - uint32_t dr7 = 0x400; /* magic dr7 reset value */ - __asm__ volatile("movl %0,%%dr7" : : "r" (dr7)); + long dr7 = 0x400; /* magic dr7 reset value; 32 bit on i386, 64 bit on x86_64 */ + __asm__ volatile("mov %0,%%dr7" : : "r" (dr7)); } #if MACH_KDP unsigned kdp_has_active_watchpoints = 0; +#define NO_WATCHPOINTS (!kdp_has_active_watchpoints) +#else +#define NO_WATCHPOINTS 1 #endif /* * Trap from kernel mode. Only page-fault errors are recoverable, * and then only in special circumstances. All other errors are * fatal. Return value indicates if trap was handled. */ + void kernel_trap( - x86_saved_state_t *state) + x86_saved_state_t *state, + uintptr_t *lo_spp) { - x86_saved_state32_t *saved_state; + x86_saved_state64_t *saved_state; int code; user_addr_t vaddr; int type; - vm_map_t map; + vm_map_t map = 0; /* protected by T_PAGE_FAULT */ kern_return_t result = KERN_FAILURE; + kern_return_t fault_result = KERN_SUCCESS; thread_t thread; ast_t *myast; boolean_t intr; vm_prot_t prot; struct recovery *rp; vm_offset_t kern_ip; +#if NCOPY_WINDOWS > 0 int fault_in_copy_window = -1; - int is_user = 0; -#if MACH_KDB - pt_entry_t *pte; -#endif /* MACH_KDB */ +#endif + int is_user; + int trap_pl = get_preemption_level(); thread = current_thread(); - if (is_saved_state64(state)) - panic("kernel_trap(%p) with 64-bit state", state); - saved_state = saved_state32(state); + if (__improbable(is_saved_state32(state))) + panic("kernel_trap(%p) with 32-bit state", state); + saved_state = saved_state64(state); - vaddr = (user_addr_t)saved_state->cr2; - type = saved_state->trapno; - code = saved_state->err & 0xffff; - intr = (saved_state->efl & EFL_IF) != 0; /* state of ints at trap */ + /* Record cpu where state was captured */ + saved_state->isf.cpu = cpu_number(); - kern_ip = (vm_offset_t)saved_state->eip; + vaddr = (user_addr_t)saved_state->cr2; + type = saved_state->isf.trapno; + code = (int)(saved_state->isf.err & 0xffff); + intr = (saved_state->isf.rflags & EFL_IF) != 0; /* state of ints at trap */ + kern_ip = (vm_offset_t)saved_state->isf.rip; myast = ast_pending(); - if (perfASTHook) { + is_user = (vaddr < VM_MAX_USER_PAGE_ADDRESS); + + perfASTCallback astfn = perfASTHook; + if (__improbable(astfn != NULL)) { if (*myast & AST_CHUD_ALL) - perfASTHook(type, NULL, 0, 0); + astfn(AST_CHUD_ALL, myast); } else *myast &= ~AST_CHUD_ALL; + +#if CONFIG_DTRACE /* - * Is there a hook? - */ - if (perfTrapHook) { - if (perfTrapHook(type, NULL, 0, 0) == KERN_SUCCESS) { - /* + * Is there a DTrace hook? + */ + if (__improbable(tempDTraceTrapHook != NULL)) { + if (tempDTraceTrapHook(type, state, lo_spp, 0) == KERN_SUCCESS) { + /* * If it succeeds, we are done... */ return; } } +#endif /* CONFIG_DTRACE */ + /* * we come here with interrupts off as we don't want to recurse * on preemption below. but we do want to re-enable interrupts * as soon we possibly can to hold latency down */ - if (T_PREEMPT == type) { - - KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86, type)) | DBG_FUNC_NONE, - 0, 0, 0, kern_ip, 0); - + if (__improbable(T_PREEMPT == type)) { ast_taken(AST_PREEMPTION, FALSE); + + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86, type)) | DBG_FUNC_NONE, + 0, 0, 0, VM_KERNEL_UNSLIDE(kern_ip), 0); return; } - + + user_addr_t kd_vaddr = is_user ? vaddr : VM_KERNEL_UNSLIDE(vaddr); + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86, type)) | DBG_FUNC_NONE, + (unsigned)(kd_vaddr >> 32), (unsigned)kd_vaddr, is_user, + VM_KERNEL_UNSLIDE(kern_ip), 0); + + if (T_PAGE_FAULT == type) { /* * assume we're faulting in the kernel map */ map = kernel_map; - if (thread != THREAD_NULL && thread->map != kernel_map) { - vm_offset_t copy_window_base; + if (__probable(thread != THREAD_NULL && thread->map != kernel_map)) { +#if NCOPY_WINDOWS > 0 + vm_offset_t copy_window_base; vm_offset_t kvaddr; int window_index; kvaddr = (vm_offset_t)vaddr; - /* + /* * must determine if fault occurred in * the copy window while pre-emption is * disabled for this processor so that * we only need to look at the window * associated with this processor */ - copy_window_base = current_cpu_datap()->cpu_copywindow_base; + copy_window_base = current_cpu_datap()->cpu_copywindow_base; if (kvaddr >= copy_window_base && kvaddr < (copy_window_base + (NBPDE * NCOPY_WINDOWS)) ) { - window_index = (kvaddr - copy_window_base) / NBPDE; + window_index = (int)((kvaddr - copy_window_base) / NBPDE); if (thread->machine.copy_window[window_index].user_base != (user_addr_t)-1) { @@ -397,13 +613,55 @@ kernel_trap( map = thread->map; fault_in_copy_window = window_index; } - is_user = -1; } +#else + if (__probable(vaddr < VM_MAX_USER_PAGE_ADDRESS)) { + /* fault occurred in userspace */ + map = thread->map; + + /* Intercept a potential Supervisor Mode Execute + * Protection fault. These criteria identify + * both NX faults and SMEP faults, but both + * are fatal. We avoid checking PTEs (racy). + * (The VM could just redrive a SMEP fault, hence + * the intercept). + */ + if (__improbable((code == (T_PF_PROT | T_PF_EXECUTE)) && + (pmap_smep_enabled) && (saved_state->isf.rip == vaddr))) { + goto debugger_entry; + } + + /* + * Additionally check for SMAP faults... + * which are characterized by page-present and + * the AC bit unset (i.e. not from copyin/out path). + */ + if (__improbable(code & T_PF_PROT && + pmap_smap_enabled && + (saved_state->isf.rflags & EFL_AC) == 0)) { + goto debugger_entry; + } + + /* + * If we're not sharing cr3 with the user + * and we faulted in copyio, + * then switch cr3 here and dismiss the fault. + */ + if (no_shared_cr3 && + (thread->machine.specFlags&CopyIOActive) && + map->pmap->pm_cr3 != get_cr3_base()) { + pmap_assert(current_cpu_datap()->cpu_pmap_pcid_enabled == FALSE); + set_cr3_raw(map->pmap->pm_cr3); + return; + } + if (__improbable(vaddr < PAGE_SIZE) && + ((thread->machine.specFlags & CopyIOActive) == 0)) { + goto debugger_entry; + } + } +#endif } } - KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86, type)) | DBG_FUNC_NONE, - (int)(vaddr >> 32), (int)vaddr, is_user, kern_ip, 0); - (void) ml_set_interrupts_enabled(intr); @@ -425,8 +683,8 @@ kernel_trap( fpSSEexterrflt(); return; case T_DEBUG: - if ((saved_state->efl & EFL_TF) == 0 - && !kdp_has_active_watchpoints) { + if ((saved_state->isf.rflags & EFL_TF) == 0 && NO_WATCHPOINTS) + { /* We've somehow encountered a debug * register match that does not belong * to the kernel debugger. @@ -434,99 +692,57 @@ kernel_trap( */ reset_dr7(); return; - } + } goto debugger_entry; +#ifdef __x86_64__ + case T_INT3: + goto debugger_entry; +#endif case T_PAGE_FAULT: - /* - * If the current map is a submap of the kernel map, - * and the address is within that map, fault on that - * map. If the same check is done in vm_fault - * (vm_map_lookup), we may deadlock on the kernel map - * lock. - */ +#if CONFIG_DTRACE + if (thread != THREAD_NULL && thread->options & TH_OPT_DTRACE) { /* Executing under dtrace_probe? */ + if (dtrace_tally_fault(vaddr)) { /* Should a fault under dtrace be ignored? */ + /* + * DTrace has "anticipated" the possibility of this fault, and has + * established the suitable recovery state. Drop down now into the + * recovery handling code in "case T_GENERAL_PROTECTION:". + */ + goto FALL_THROUGH; + } + } +#endif /* CONFIG_DTRACE */ + prot = VM_PROT_READ; if (code & T_PF_WRITE) prot |= VM_PROT_WRITE; -#if PAE if (code & T_PF_EXECUTE) prot |= VM_PROT_EXECUTE; -#endif - -#if MACH_KDB - /* - * Check for watchpoint on kernel static data. - * vm_fault would fail in this case - */ - if (map == kernel_map && db_watchpoint_list && db_watchpoints_inserted && - (code & T_PF_WRITE) && vaddr < vm_map_max(map) && - ((*(pte = pmap_pte(kernel_pmap, (vm_map_offset_t)vaddr))) & INTEL_PTE_WRITE) == 0) { - pmap_store_pte( - pte, - *pte | INTEL_PTE_VALID | INTEL_PTE_WRITE); - /* XXX need invltlb here? */ - - result = KERN_SUCCESS; - goto look_for_watchpoints; - } -#endif /* MACH_KDB */ - result = vm_fault(map, - vm_map_trunc_page(vaddr), + fault_result = result = vm_fault(map, + vaddr, prot, FALSE, THREAD_UNINT, NULL, 0); -#if MACH_KDB - if (result == KERN_SUCCESS) { - /* - * Look for watchpoints - */ -look_for_watchpoints: - if (map == kernel_map && db_watchpoint_list && db_watchpoints_inserted && (code & T_PF_WRITE) && - db_find_watchpoint(map, vaddr, saved_state)) - kdb_trap(T_WATCHPOINT, 0, saved_state); - } -#endif /* MACH_KDB */ - if (result == KERN_SUCCESS) { - +#if NCOPY_WINDOWS > 0 if (fault_in_copy_window != -1) { - pt_entry_t *updp; - pt_entry_t *kpdp; - - /* - * in case there was no page table assigned - * for the user base address and the pmap - * got 'expanded' due to this fault, we'll - * copy in the descriptor - * - * we're either setting the page table descriptor - * to the same value or it was 0... no need - * for a TLB flush in either case - */ - - ml_set_interrupts_enabled(FALSE); - updp = pmap_pde(map->pmap, thread->machine.copy_window[fault_in_copy_window].user_base); - assert(updp); - if (0 == updp) panic("trap: updp 0"); /* XXX DEBUG */ - kpdp = current_cpu_datap()->cpu_copywindow_pdp; - kpdp += fault_in_copy_window; - -#if JOE_DEBUG - if (*kpdp && (*kpdp & PG_FRAME) != (*updp & PG_FRAME)) - panic("kernel_fault: user pdp doesn't match - updp = 0x%x, kpdp = 0x%x\n", updp, kpdp); -#endif - pmap_store_pte(kpdp, *updp); - + ml_set_interrupts_enabled(FALSE); + copy_window_fault(thread, map, + fault_in_copy_window); (void) ml_set_interrupts_enabled(intr); } +#endif /* NCOPY_WINDOWS > 0 */ return; } /* * fall through */ +#if CONFIG_DTRACE +FALL_THROUGH: +#endif /* CONFIG_DTRACE */ case T_GENERAL_PROTECTION: /* @@ -543,8 +759,8 @@ look_for_watchpoints: /* * Check thread recovery address also. */ - if (thread->recover) { - set_recovery_ip(saved_state, thread->recover); + if (thread != THREAD_NULL && thread->recover) { + set_recovery_ip(saved_state, thread->recover); thread->recover = 0; return; } @@ -554,7 +770,6 @@ look_for_watchpoints: * * fall through... */ - default: /* * Exception 15 is reserved but some chips may generate it @@ -570,30 +785,14 @@ debugger_entry: * context at the moment of the trap, to facilitate * access through the debugger. */ - sync_iss_to_iks(saved_state); -#if MACH_KDB -restart_debugger: -#endif /* MACH_KDB */ + sync_iss_to_iks(state); #if MACH_KDP - if (current_debugger != KDB_CUR_DB) { - if (kdp_i386_trap(type, saved_state, result, vaddr)) - return; - } -#endif /* MACH_KDP */ -#if MACH_KDB - else - if (kdb_trap(type, code, saved_state)) { - if (switch_debugger) { - current_debugger = KDP_CUR_DB; - switch_debugger = 0; - goto restart_debugger; - } - return; - } -#endif /* MACH_KDB */ + if (kdp_i386_trap(type, saved_state, result, (vm_offset_t)vaddr)) + return; +#endif } - - panic_trap(saved_state); + pal_cli(); + panic_trap(saved_state, trap_pl, fault_result); /* * NO RETURN */ @@ -601,39 +800,73 @@ restart_debugger: static void -set_recovery_ip(x86_saved_state32_t *saved_state, vm_offset_t ip) +set_recovery_ip(x86_saved_state64_t *saved_state, vm_offset_t ip) { - saved_state->eip = ip; + saved_state->isf.rip = ip; } - static void -panic_trap(x86_saved_state32_t *regs) +panic_trap(x86_saved_state64_t *regs, uint32_t pl, kern_return_t fault_result) { - const char *trapname = "Unknown"; - uint32_t cr0 = get_cr0(); - uint32_t cr2 = get_cr2(); - uint32_t cr3 = get_cr3(); - uint32_t cr4 = get_cr4(); - + const char *trapname = "Unknown"; + pal_cr_t cr0, cr2, cr3, cr4; + boolean_t potential_smep_fault = FALSE, potential_kernel_NX_fault = FALSE; + boolean_t potential_smap_fault = FALSE; + + pal_get_control_registers( &cr0, &cr2, &cr3, &cr4 ); + assert(ml_get_interrupts_enabled() == FALSE); + current_cpu_datap()->cpu_fatal_trap_state = regs; + /* + * Issue an I/O port read if one has been requested - this is an + * event logic analyzers can use as a trigger point. + */ panic_io_port_read(); - kprintf("panic trap number 0x%x, eip 0x%x\n", regs->trapno, regs->eip); - kprintf("cr0 0x%08x cr2 0x%08x cr3 0x%08x cr4 0x%08x\n", + kprintf("panic trap number 0x%x, rip 0x%016llx\n", + regs->isf.trapno, regs->isf.rip); + kprintf("cr0 0x%016llx cr2 0x%016llx cr3 0x%016llx cr4 0x%016llx\n", cr0, cr2, cr3, cr4); - if (regs->trapno < TRAP_TYPES) - trapname = trap_type[regs->trapno]; - - panic("Unresolved kernel trap (CPU %d, Type %d=%s), registers:\n" - "CR0: 0x%08x, CR2: 0x%08x, CR3: 0x%08x, CR4: 0x%08x\n" - "EAX: 0x%08x, EBX: 0x%08x, ECX: 0x%08x, EDX: 0x%08x\n" - "CR2: 0x%08x, EBP: 0x%08x, ESI: 0x%08x, EDI: 0x%08x\n" - "EFL: 0x%08x, EIP: 0x%08x, CS: 0x%08x, DS: 0x%08x\n", - cpu_number(), regs->trapno, trapname, cr0, cr2, cr3, cr4, - regs->eax,regs->ebx,regs->ecx,regs->edx, - regs->cr2,regs->ebp,regs->esi,regs->edi, - regs->efl,regs->eip,regs->cs, regs->ds); + if (regs->isf.trapno < TRAP_TYPES) + trapname = trap_type[regs->isf.trapno]; + + if ((regs->isf.trapno == T_PAGE_FAULT) && (regs->isf.err == (T_PF_PROT | T_PF_EXECUTE)) && (regs->isf.rip == regs->cr2)) { + if (pmap_smep_enabled && (regs->isf.rip < VM_MAX_USER_PAGE_ADDRESS)) { + potential_smep_fault = TRUE; + } else if (regs->isf.rip >= VM_MIN_KERNEL_AND_KEXT_ADDRESS) { + potential_kernel_NX_fault = TRUE; + } + } else if (pmap_smap_enabled && + regs->isf.trapno == T_PAGE_FAULT && + regs->isf.err & T_PF_PROT && + regs->cr2 < VM_MAX_USER_PAGE_ADDRESS && + regs->isf.rip >= VM_MIN_KERNEL_AND_KEXT_ADDRESS) { + potential_smap_fault = TRUE; + } + +#undef panic + panic("Kernel trap at 0x%016llx, type %d=%s, registers:\n" + "CR0: 0x%016llx, CR2: 0x%016llx, CR3: 0x%016llx, CR4: 0x%016llx\n" + "RAX: 0x%016llx, RBX: 0x%016llx, RCX: 0x%016llx, RDX: 0x%016llx\n" + "RSP: 0x%016llx, RBP: 0x%016llx, RSI: 0x%016llx, RDI: 0x%016llx\n" + "R8: 0x%016llx, R9: 0x%016llx, R10: 0x%016llx, R11: 0x%016llx\n" + "R12: 0x%016llx, R13: 0x%016llx, R14: 0x%016llx, R15: 0x%016llx\n" + "RFL: 0x%016llx, RIP: 0x%016llx, CS: 0x%016llx, SS: 0x%016llx\n" + "Fault CR2: 0x%016llx, Error code: 0x%016llx, Fault CPU: 0x%x%s%s%s%s, PL: %d, VF: %d\n", + regs->isf.rip, regs->isf.trapno, trapname, + cr0, cr2, cr3, cr4, + regs->rax, regs->rbx, regs->rcx, regs->rdx, + regs->isf.rsp, regs->rbp, regs->rsi, regs->rdi, + regs->r8, regs->r9, regs->r10, regs->r11, + regs->r12, regs->r13, regs->r14, regs->r15, + regs->isf.rflags, regs->isf.rip, regs->isf.cs & 0xFFFF, + regs->isf.ss & 0xFFFF,regs->cr2, regs->isf.err, regs->isf.cpu, + virtualized ? " VMM" : "", + potential_kernel_NX_fault ? " Kernel NX fault" : "", + potential_smep_fault ? " SMEP/User NX fault" : "", + potential_smap_fault ? " SMAP fault" : "", + pl, + fault_result); /* * This next statement is not executed, * but it's needed to stop the compiler using tail call optimization @@ -642,196 +875,10 @@ panic_trap(x86_saved_state32_t *regs) cr0 = 0; } -extern void kprintf_break_lock(void); - - -/* - * Called from locore on a special reserved stack after a double-fault - * is taken in kernel space. - * Kernel stack overflow is one route here. - */ -void -panic_double_fault(int code) -{ - struct i386_tss *my_ktss = current_ktss(); - - /* Set postcode (DEBUG only) */ - postcode(PANIC_DOUBLE_FAULT); - -/* Issue an I/O port read if one has been requested - this is an event logic - * analyzers can use as a trigger point. - */ - panic_io_port_read(); - - /* - * Break kprintf lock in case of recursion, - * and record originally faulted instruction address. - */ - kprintf_break_lock(); - -#if MACH_KDP - /* - * Print backtrace leading to first fault: - */ - panic_i386_backtrace((void *) my_ktss->ebp, 10); +#if CONFIG_DTRACE +extern kern_return_t dtrace_user_probe(x86_saved_state_t *); #endif - panic("Double fault (CPU:%d, thread:%p, code:0x%x)," - "registers:\n" - "CR0: 0x%08x, CR2: 0x%08x, CR3: 0x%08x, CR4: 0x%08x\n" - "EAX: 0x%08x, EBX: 0x%08x, ECX: 0x%08x, EDX: 0x%08x\n" - "ESP: 0x%08x, EBP: 0x%08x, ESI: 0x%08x, EDI: 0x%08x\n" - "EFL: 0x%08x, EIP: 0x%08x\n", - cpu_number(), current_thread(), code, - get_cr0(), get_cr2(), get_cr3(), get_cr4(), - my_ktss->eax, my_ktss->ebx, my_ktss->ecx, my_ktss->edx, - my_ktss->esp, my_ktss->ebp, my_ktss->esi, my_ktss->edi, - my_ktss->eflags, my_ktss->eip); -} - - -/* - * Called from locore on a special reserved stack after a machine-check - */ -void -panic_machine_check(int code) -{ - struct i386_tss *my_ktss = current_ktss(); - - /* Set postcode (DEBUG only) */ - postcode(PANIC_MACHINE_CHECK); - - /* - * Break kprintf lock in case of recursion, - * and record originally faulted instruction address. - */ - kprintf_break_lock(); - - /* - * Dump the contents of the machine check MSRs (if any). - */ - mca_dump(); - - /* - * And that's all folks, we don't attempt recovery... - */ - panic("Machine-check (CPU:%d, thread:%p, code:0x%x)," - "registers:\n" - "CR0: 0x%08x, CR2: 0x%08x, CR3: 0x%08x, CR4: 0x%08x\n" - "EAX: 0x%08x, EBX: 0x%08x, ECX: 0x%08x, EDX: 0x%08x\n" - "ESP: 0x%08x, EBP: 0x%08x, ESI: 0x%08x, EDI: 0x%08x\n" - "EFL: 0x%08x, EIP: 0x%08x\n", - cpu_number(), current_thread(), code, - get_cr0(), get_cr2(), get_cr3(), get_cr4(), - my_ktss->eax, my_ktss->ebx, my_ktss->ecx, my_ktss->edx, - my_ktss->esp, my_ktss->ebp, my_ktss->esi, my_ktss->edi, - my_ktss->eflags, my_ktss->eip); -} - -void -panic_double_fault64(x86_saved_state_t *esp) -{ - /* Set postcode (DEBUG only) */ - postcode(PANIC_DOUBLE_FAULT); - - /* - * Break kprintf lock in case of recursion, - * and record originally faulted instruction address. - */ - kprintf_break_lock(); - - /* - * Dump the interrupt stack frame at last kernel entry. - */ - if (is_saved_state64(esp)) { - x86_saved_state64_t *ss64p = saved_state64(esp); - panic("Double fault (CPU:%d, thread:%p, trapno:0x%x, err:0x%qx)," - "registers:\n" - "CR0: 0x%08x, CR2: 0x%08x, CR3: 0x%08x, CR4: 0x%08x\n" - "RAX: 0x%016qx, RBX: 0x%016qx, RCX: 0x%016qx, RDX: 0x%016qx\n" - "RSP: 0x%016qx, RBP: 0x%016qx, RSI: 0x%016qx, RDI: 0x%016qx\n" - "R8: 0x%016qx, R9: 0x%016qx, R10: 0x%016qx, R11: 0x%016qx\n" - "R12: 0x%016qx, R13: 0x%016qx, R14: 0x%016qx, R15: 0x%016qx\n" - "RFL: 0x%016qx, RIP: 0x%016qx, CR2: 0x%016qx\n", - cpu_number(), current_thread(), ss64p->isf.trapno, ss64p->isf.err, - get_cr0(), get_cr2(), get_cr3(), get_cr4(), - ss64p->rax, ss64p->rbx, ss64p->rcx, ss64p->rdx, - ss64p->isf.rsp, ss64p->rbp, ss64p->rsi, ss64p->rdi, - ss64p->r8, ss64p->r9, ss64p->r10, ss64p->r11, - ss64p->r12, ss64p->r13, ss64p->r14, ss64p->r15, - ss64p->isf.rflags, ss64p->isf.rip, ss64p->cr2); - } else { - x86_saved_state32_t *ss32p = saved_state32(esp); - panic("Double fault (CPU:%d, thread:%p, trapno:0x%x, err:0x%x)," - "registers:\n" - "CR0: 0x%08x, CR2: 0x%08x, CR3: 0x%08x, CR4: 0x%08x\n" - "EAX: 0x%08x, EBX: 0x%08x, ECX: 0x%08x, EDX: 0x%08x\n" - "ESP: 0x%08x, EBP: 0x%08x, ESI: 0x%08x, EDI: 0x%08x\n" - "EFL: 0x%08x, EIP: 0x%08x\n", - cpu_number(), current_thread(), ss32p->trapno, ss32p->err, - get_cr0(), get_cr2(), get_cr3(), get_cr4(), - ss32p->eax, ss32p->ebx, ss32p->ecx, ss32p->edx, - ss32p->uesp, ss32p->ebp, ss32p->esi, ss32p->edi, - ss32p->efl, ss32p->eip); - } -} - -/* - * Machine check handler for 64-bit. - */ -void -panic_machine_check64(x86_saved_state_t *esp) -{ - /* Set postcode (DEBUG only) */ - postcode(PANIC_MACHINE_CHECK); - - /* - * Break kprintf lock in case of recursion, - * and record originally faulted instruction address. - */ - kprintf_break_lock(); - - /* - * Dump the contents of the machine check MSRs (if any). - */ - mca_dump(); - - /* - * And that's all folks, we don't attempt recovery... - */ - if (is_saved_state64(esp)) { - x86_saved_state64_t *ss64p = saved_state64(esp); - panic("Machine Check (CPU:%d, thread:%p, trapno:0x%x, err:0x%qx)," - "registers:\n" - "CR0: 0x%08x, CR2: 0x%08x, CR3: 0x%08x, CR4: 0x%08x\n" - "RAX: 0x%016qx, RBX: 0x%016qx, RCX: 0x%016qx, RDX: 0x%016qx\n" - "RSP: 0x%016qx, RBP: 0x%016qx, RSI: 0x%016qx, RDI: 0x%016qx\n" - "R8: 0x%016qx, R9: 0x%016qx, R10: 0x%016qx, R11: 0x%016qx\n" - "R12: 0x%016qx, R13: 0x%016qx, R14: 0x%016qx, R15: 0x%016qx\n" - "RFL: 0x%016qx, RIP: 0x%016qx\n", - cpu_number(), current_thread(), ss64p->isf.trapno, ss64p->isf.err, - get_cr0(), get_cr2(), get_cr3(), get_cr4(), - ss64p->rax, ss64p->rbx, ss64p->rcx, ss64p->rdx, - ss64p->isf.rsp, ss64p->rbp, ss64p->rsi, ss64p->rdi, - ss64p->r8, ss64p->r9, ss64p->r10, ss64p->r11, - ss64p->r12, ss64p->r13, ss64p->r14, ss64p->r15, - ss64p->isf.rflags, ss64p->isf.rip); - } else { - x86_saved_state32_t *ss32p = saved_state32(esp); - panic("Machine Check (CPU:%d, thread:%p, trapno:0x%x, err:0x%x)," - "registers:\n" - "CR0: 0x%08x, CR2: 0x%08x, CR3: 0x%08x, CR4: 0x%08x\n" - "EAX: 0x%08x, EBX: 0x%08x, ECX: 0x%08x, EDX: 0x%08x\n" - "ESP: 0x%08x, EBP: 0x%08x, ESI: 0x%08x, EDI: 0x%08x\n" - "EFL: 0x%08x, EIP: 0x%08x\n", - cpu_number(), current_thread(), ss32p->trapno, ss32p->err, - get_cr0(), get_cr2(), get_cr3(), get_cr4(), - ss32p->eax, ss32p->ebx, ss32p->ecx, ss32p->edx, - ss32p->uesp, ss32p->ebp, ss32p->esi, ss32p->edi, - ss32p->efl, ss32p->eip); - } -} - /* * Trap from user mode. */ @@ -839,18 +886,18 @@ void user_trap( x86_saved_state_t *saved_state) { - int exc; - int code; - int err; - unsigned int subcode; - int type; - user_addr_t vaddr; - vm_prot_t prot; - thread_t thread = current_thread(); - ast_t *myast; - boolean_t intr; - kern_return_t kret; - user_addr_t rip; + int exc; + int err; + mach_exception_code_t code; + mach_exception_subcode_t subcode; + int type; + user_addr_t vaddr; + vm_prot_t prot; + thread_t thread = current_thread(); + ast_t *myast; + kern_return_t kret; + user_addr_t rip; + unsigned long dr6 = 0; /* 32 bit for i386, 64 bit for x86_64 */ assert((is_saved_state32(saved_state) && !thread_is_64bit(thread)) || (is_saved_state64(saved_state) && thread_is_64bit(thread))); @@ -860,23 +907,42 @@ user_trap( regs = saved_state64(saved_state); + /* Record cpu where state was captured */ + regs->isf.cpu = cpu_number(); + type = regs->isf.trapno; - err = regs->isf.err & 0xffff; + err = (int)regs->isf.err & 0xffff; vaddr = (user_addr_t)regs->cr2; rip = (user_addr_t)regs->isf.rip; } else { - x86_saved_state32_t *regs; + x86_saved_state32_t *regs; regs = saved_state32(saved_state); + /* Record cpu where state was captured */ + regs->cpu = cpu_number(); + type = regs->trapno; err = regs->err & 0xffff; vaddr = (user_addr_t)regs->cr2; rip = (user_addr_t)regs->eip; } - KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_EXCP_UTRAP_x86, type)) | DBG_FUNC_NONE, - (int)(vaddr>>32), (int)vaddr, (int)(rip>>32), (int)rip, 0); + if ((type == T_DEBUG) && thread->machine.ids) { + unsigned long clear = 0; + /* Stash and clear this processor's DR6 value, in the event + * this was a debug register match + */ + __asm__ volatile ("mov %%db6, %0" : "=r" (dr6)); + __asm__ volatile ("mov %0, %%db6" : : "r" (clear)); + } + + pal_sti(); + + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + (MACHDBG_CODE(DBG_MACH_EXCP_UTRAP_x86, type)) | DBG_FUNC_NONE, + (unsigned)(vaddr>>32), (unsigned)vaddr, + (unsigned)(rip>>32), (unsigned)rip, 0); code = 0; subcode = 0; @@ -886,21 +952,34 @@ user_trap( kprintf("user_trap(0x%08x) type=%d vaddr=0x%016llx\n", saved_state, type, vaddr); #endif - myast = ast_pending(); - if (perfASTHook) { + + perfASTCallback astfn = perfASTHook; + if (__improbable(astfn != NULL)) { + myast = ast_pending(); if (*myast & AST_CHUD_ALL) { - perfASTHook(type, saved_state, 0, 0); + astfn(AST_CHUD_ALL, myast); } - } else { - *myast &= ~AST_CHUD_ALL; } /* Is there a hook? */ - if (perfTrapHook) { - if (perfTrapHook(type, saved_state, 0, 0) == KERN_SUCCESS) + perfCallback fn = perfTrapHook; + if (__improbable(fn != NULL)) { + if (fn(type, saved_state, 0, 0) == KERN_SUCCESS) return; /* If it succeeds, we are done... */ } +#if CONFIG_DTRACE + /* + * DTrace does not consume all user traps, only INT_3's for now. + * Avoid needlessly calling tempDTraceTrapHook here, and let the + * INT_3 case handle them. + */ +#endif + + DEBUG_KPRINT_SYSCALL_MASK(1, + "user_trap: type=0x%x(%s) err=0x%x cr2=%p rip=%p\n", + type, trap_type[type], err, (void *)(long) vaddr, (void *)(long) rip); + switch (type) { case T_DIVIDE_ERROR: @@ -911,12 +990,11 @@ user_trap( case T_DEBUG: { pcb_t pcb; - unsigned int clear = 0; /* - * get dr6 and set it in the thread's pcb before - * returning to userland + * Update the PCB with this processor's DR6 value + * in the event this was a debug register match. */ - pcb = thread->machine.pcb; + pcb = THREAD_TO_PCB(thread); if (pcb->ids) { /* * We can get and set the status register @@ -925,22 +1003,22 @@ user_trap( * used on x86_64 */ if (thread_is_64bit(thread)) { - uint32_t dr6; x86_debug_state64_t *ids = pcb->ids; - dr6 = (uint32_t)ids->dr6; - __asm__ volatile ("movl %%db6, %0" : "=r" (dr6)); ids->dr6 = dr6; } else { /* 32 bit thread */ x86_debug_state32_t *ids = pcb->ids; - __asm__ volatile ("movl %%db6, %0" : "=r" (ids->dr6)); + ids->dr6 = (uint32_t) dr6; } - __asm__ volatile ("movl %0, %%db6" : : "r" (clear)); } exc = EXC_BREAKPOINT; code = EXC_I386_SGL; break; } case T_INT3: +#if CONFIG_DTRACE + if (dtrace_user_probe(saved_state) == KERN_SUCCESS) + return; /* If it succeeds, we are done... */ +#endif exc = EXC_BREAKPOINT; code = EXC_I386_BPT; break; @@ -961,15 +1039,14 @@ user_trap( break; case T_NO_FPU: - case 32: /* XXX */ fpnoextflt(); return; case T_FPU_FAULT: - fpextovrflt(); + fpextovrflt(); /* Propagates exception directly, doesn't return */ return; - case 10: /* invalid TSS == iret with NT flag set */ + case T_INVALID_TSS: /* invalid TSS == iret with NT flag set */ exc = EXC_BAD_INSTRUCTION; code = EXC_I386_INVTSSFLT; subcode = err; @@ -988,68 +1065,83 @@ user_trap( break; case T_GENERAL_PROTECTION: - exc = EXC_BAD_INSTRUCTION; + /* + * There's a wide range of circumstances which generate this + * class of exception. From user-space, many involve bad + * addresses (such as a non-canonical 64-bit address). + * So we map this to EXC_BAD_ACCESS (and thereby SIGSEGV). + * The trouble is cr2 doesn't contain the faulting address; + * we'd need to decode the faulting instruction to really + * determine this. We'll leave that to debuggers. + * However, attempted execution of privileged instructions + * (e.g. cli) also generate GP faults and so we map these to + * to EXC_BAD_ACCESS (and thence SIGSEGV) also - rather than + * EXC_BAD_INSTRUCTION which is more accurate. We just can't + * win! + */ + exc = EXC_BAD_ACCESS; code = EXC_I386_GPFLT; subcode = err; break; case T_PAGE_FAULT: - prot = VM_PROT_READ; + { + prot = VM_PROT_READ; if (err & T_PF_WRITE) prot |= VM_PROT_WRITE; -#if PAE - if (err & T_PF_EXECUTE) + if (__improbable(err & T_PF_EXECUTE)) prot |= VM_PROT_EXECUTE; -#endif - kret = vm_fault(thread->map, vm_map_trunc_page(vaddr), - prot, FALSE, - THREAD_ABORTSAFE, NULL, 0); + kret = vm_fault(thread->map, + vaddr, + prot, FALSE, + THREAD_ABORTSAFE, NULL, 0); + + if (__probable((kret == KERN_SUCCESS) || (kret == KERN_ABORTED))) { + thread_exception_return(); + /*NOTREACHED*/ + } user_page_fault_continue(kret); - - /* NOTREACHED */ + } /* NOTREACHED */ break; case T_SSE_FLOAT_ERROR: - fpSSEexterrflt(); + fpSSEexterrflt(); /* Propagates exception directly, doesn't return */ return; case T_FLOATING_POINT_ERROR: - fpexterrflt(); + fpexterrflt(); /* Propagates exception directly, doesn't return */ return; + case T_DTRACE_RET: +#if CONFIG_DTRACE + if (dtrace_user_probe(saved_state) == KERN_SUCCESS) + return; /* If it succeeds, we are done... */ +#endif + /* + * If we get an INT 0x7f when we do not expect to, + * treat it as an illegal instruction + */ + exc = EXC_BAD_INSTRUCTION; + code = EXC_I386_INVOP; + break; + default: -#if MACH_KGDB - Debugger("Unanticipated user trap"); - return; -#endif /* MACH_KGDB */ -#if MACH_KDB - if (kdb_trap(type, err, saved_state)) - return; -#endif /* MACH_KDB */ - panic("user trap"); + panic("Unexpected user trap, type %d", type); return; } - intr = ml_set_interrupts_enabled(FALSE); - myast = ast_pending(); - while (*myast & AST_ALL) { - ast_taken(AST_ALL, intr); - ml_set_interrupts_enabled(FALSE); - myast = ast_pending(); - } - ml_set_interrupts_enabled(intr); - + /* Note: Codepaths that directly return from user_trap() have pending + * ASTs processed in locore + */ i386_exception(exc, code, subcode); - /*NOTREACHED*/ + /* NOTREACHED */ } /* * Handle AST traps for i386. - * Check for delayed floating-point exception from - * AT-bus machines. */ extern void log_thread_action (thread_t, char *); @@ -1083,11 +1175,13 @@ i386_astintr(int preemption) void i386_exception( int exc, - int code, - int subcode) + mach_exception_code_t code, + mach_exception_subcode_t subcode) { - exception_data_type_t codes[EXCEPTION_CODE_MAX]; + mach_exception_data_type_t codes[EXCEPTION_CODE_MAX]; + DEBUG_KPRINT_SYSCALL_MACH("i386_exception: exc=%d code=0x%llx subcode=0x%llx\n", + exc, code, subcode); codes[0] = code; /* new exception interface */ codes[1] = subcode; exception_triage(exc, codes, 2); @@ -1095,124 +1189,59 @@ i386_exception( } -void -kernel_preempt_check(void) -{ - ast_t *myast; - boolean_t intr; - - /* - * disable interrupts to both prevent pre-emption - * and to keep the ast state from changing via - * an interrupt handler making something runnable - */ - intr = ml_set_interrupts_enabled(FALSE); - - myast = ast_pending(); - - if ((*myast & AST_URGENT) && intr == TRUE && get_interrupt_level() == 0) { - /* - * can handle interrupts and preemptions - * at this point - */ - ml_set_interrupts_enabled(intr); - - /* - * now cause the PRE-EMPTION trap - */ - __asm__ volatile (" int $0xff"); - } else { - /* - * if interrupts were already disabled or - * we're in an interrupt context, we can't - * preempt... of course if AST_URGENT - * isn't set we also don't want to - */ - ml_set_interrupts_enabled(intr); - } -} - -#if MACH_KDB - -extern void db_i386_state(x86_saved_state32_t *regs); - -#include - -void -db_i386_state( - x86_saved_state32_t *regs) -{ - db_printf("eip %8x\n", regs->eip); - db_printf("trap %8x\n", regs->trapno); - db_printf("err %8x\n", regs->err); - db_printf("efl %8x\n", regs->efl); - db_printf("ebp %8x\n", regs->ebp); - db_printf("esp %8x\n", regs->cr2); - db_printf("uesp %8x\n", regs->uesp); - db_printf("cs %8x\n", regs->cs & 0xff); - db_printf("ds %8x\n", regs->ds & 0xff); - db_printf("es %8x\n", regs->es & 0xff); - db_printf("fs %8x\n", regs->fs & 0xff); - db_printf("gs %8x\n", regs->gs & 0xff); - db_printf("ss %8x\n", regs->ss & 0xff); - db_printf("eax %8x\n", regs->eax); - db_printf("ebx %8x\n", regs->ebx); - db_printf("ecx %8x\n", regs->ecx); - db_printf("edx %8x\n", regs->edx); - db_printf("esi %8x\n", regs->esi); - db_printf("edi %8x\n", regs->edi); -} - -#endif /* MACH_KDB */ - -/* Synchronize a thread's i386_kernel_state (if any) with the given - * i386_saved_state_t obtained from the trap/IPI handler; called in +/* Synchronize a thread's x86_kernel_state (if any) with the given + * x86_saved_state_t obtained from the trap/IPI handler; called in * kernel_trap() prior to entering the debugger, and when receiving - * an "MP_KDP" IPI. + * an "MP_KDP" IPI. Called with null saved_state if an incoming IPI + * was detected from the kernel while spinning with interrupts masked. */ void -sync_iss_to_iks(x86_saved_state32_t *saved_state) +sync_iss_to_iks(x86_saved_state_t *saved_state) { - struct x86_kernel_state32 *iks; + struct x86_kernel_state *iks; vm_offset_t kstack; boolean_t record_active_regs = FALSE; - if ((kstack = current_thread()->kernel_stack) != 0) { - x86_saved_state32_t *regs; + /* The PAL may have a special way to sync registers */ + if (saved_state && saved_state->flavor == THREAD_STATE_NONE) + pal_get_kern_regs( saved_state ); - regs = saved_state; + if ((kstack = current_thread()->kernel_stack) != 0) { + x86_saved_state64_t *regs = saved_state64(saved_state); iks = STACK_IKS(kstack); - /* - * Did we take the trap/interrupt in kernel mode? - */ - if (regs == USER_REGS32(current_thread())) + /* Did we take the trap/interrupt in kernel mode? */ + if (saved_state == NULL || /* NULL => polling in kernel */ + regs == USER_REGS64(current_thread())) record_active_regs = TRUE; else { - iks->k_ebx = regs->ebx; - iks->k_esp = (int)regs; - iks->k_ebp = regs->ebp; - iks->k_edi = regs->edi; - iks->k_esi = regs->esi; - iks->k_eip = regs->eip; + iks->k_rbx = regs->rbx; + iks->k_rsp = regs->isf.rsp; + iks->k_rbp = regs->rbp; + iks->k_r12 = regs->r12; + iks->k_r13 = regs->r13; + iks->k_r14 = regs->r14; + iks->k_r15 = regs->r15; + iks->k_rip = regs->isf.rip; } } if (record_active_regs == TRUE) { - /* - * Show the trap handler path - */ - __asm__ volatile("movl %%ebx, %0" : "=m" (iks->k_ebx)); - __asm__ volatile("movl %%esp, %0" : "=m" (iks->k_esp)); - __asm__ volatile("movl %%ebp, %0" : "=m" (iks->k_ebp)); - __asm__ volatile("movl %%edi, %0" : "=m" (iks->k_edi)); - __asm__ volatile("movl %%esi, %0" : "=m" (iks->k_esi)); - /* - * "Current" instruction pointer - */ - __asm__ volatile("movl $1f, %0\n1:" : "=m" (iks->k_eip)); + /* Show the trap handler path */ + __asm__ volatile("movq %%rbx, %0" : "=m" (iks->k_rbx)); + __asm__ volatile("movq %%rsp, %0" : "=m" (iks->k_rsp)); + __asm__ volatile("movq %%rbp, %0" : "=m" (iks->k_rbp)); + __asm__ volatile("movq %%r12, %0" : "=m" (iks->k_r12)); + __asm__ volatile("movq %%r13, %0" : "=m" (iks->k_r13)); + __asm__ volatile("movq %%r14, %0" : "=m" (iks->k_r14)); + __asm__ volatile("movq %%r15, %0" : "=m" (iks->k_r15)); + /* "Current" instruction pointer */ + __asm__ volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:" + : "=m" (iks->k_rip) + : + : "rax"); } } @@ -1223,26 +1252,45 @@ sync_iss_to_iks(x86_saved_state32_t *saved_state) * or user space. */ void -sync_iss_to_iks_unconditionally(__unused x86_saved_state32_t *saved_state) { - struct x86_kernel_state32 *iks; +sync_iss_to_iks_unconditionally(__unused x86_saved_state_t *saved_state) { + struct x86_kernel_state *iks; vm_offset_t kstack; - boolean_t record_active_regs = FALSE; if ((kstack = current_thread()->kernel_stack) != 0) { - iks = STACK_IKS(kstack); - /* - * Show the trap handler path - */ - __asm__ volatile("movl %%ebx, %0" : "=m" (iks->k_ebx)); - __asm__ volatile("movl %%esp, %0" : "=m" (iks->k_esp)); - __asm__ volatile("movl %%ebp, %0" : "=m" (iks->k_ebp)); - __asm__ volatile("movl %%edi, %0" : "=m" (iks->k_edi)); - __asm__ volatile("movl %%esi, %0" : "=m" (iks->k_esi)); - /* - * "Current" instruction pointer - */ - __asm__ volatile("movl $1f, %0\n1:" : "=m" (iks->k_eip)); + /* Display the trap handler path */ + __asm__ volatile("movq %%rbx, %0" : "=m" (iks->k_rbx)); + __asm__ volatile("movq %%rsp, %0" : "=m" (iks->k_rsp)); + __asm__ volatile("movq %%rbp, %0" : "=m" (iks->k_rbp)); + __asm__ volatile("movq %%r12, %0" : "=m" (iks->k_r12)); + __asm__ volatile("movq %%r13, %0" : "=m" (iks->k_r13)); + __asm__ volatile("movq %%r14, %0" : "=m" (iks->k_r14)); + __asm__ volatile("movq %%r15, %0" : "=m" (iks->k_r15)); + /* "Current" instruction pointer */ + __asm__ volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:" : "=m" (iks->k_rip)::"rax"); + } +} +#if DEBUG +extern void thread_exception_return_internal(void) __dead2; + +void thread_exception_return(void) { + thread_t thread = current_thread(); + ml_set_interrupts_enabled(FALSE); + if (thread_is_64bit(thread) != task_has_64BitAddr(thread->task)) { + panic("Task/thread bitness mismatch %p %p, task: %d, thread: %d", thread, thread->task, thread_is_64bit(thread), task_has_64BitAddr(thread->task)); } + + if (thread_is_64bit(thread)) { + if ((gdt_desc_p(USER64_CS)->access & ACC_PL_U) == 0) { + panic("64-GDT mismatch %p, descriptor: %p", thread, gdt_desc_p(USER64_CS)); + } + } else { + if ((gdt_desc_p(USER_CS)->access & ACC_PL_U) == 0) { + panic("32-GDT mismatch %p, descriptor: %p", thread, gdt_desc_p(USER_CS)); + + } + } + thread_exception_return_internal(); } +#endif