X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/7ddcb079202367355dddccdfa4318e57d50318be..d9a64523371fa019c4575bb400cbbc3a50ac9903:/osfmk/i386/trap.c diff --git a/osfmk/i386/trap.c b/osfmk/i386/trap.c index a07b1b8cb..ef03460db 100644 --- a/osfmk/i386/trap.c +++ b/osfmk/i386/trap.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2010 Apple Inc. All rights reserved. + * Copyright (c) 2000-2018 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -60,8 +60,6 @@ * Hardware trap/fault handler. */ -#include -#include #include #include @@ -91,20 +89,12 @@ #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 @@ -117,7 +107,7 @@ #include #include - +#include #include extern void throttle_lowpri_io(int); @@ -127,16 +117,8 @@ extern void kprint_state(x86_saved_state64_t *saved_state); * Forward declarations */ static void user_page_fault_continue(kern_return_t kret); -#ifdef __i386__ -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); -extern void panic_64(x86_saved_state_t *, int, const char *, boolean_t); -#else -static void panic_trap(x86_saved_state64_t *saved_state); +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); -#endif - -volatile perfCallback perfTrapHook = NULL; /* Pointer to CHUD trap hook routine */ #if CONFIG_DTRACE /* See */ @@ -145,7 +127,10 @@ perfCallback tempDTraceTrapHook = NULL; /* Pointer to DTrace fbt trap hook routi 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) @@ -156,7 +141,7 @@ thread_syscall_return( pal_register_cache_state(thr_act, DIRTY); - if (thread_is_64bit(thr_act)) { + if (thread_is_64bit_addr(thr_act)) { x86_saved_state64_t *regs; regs = USER_REGS64(thr_act); @@ -166,7 +151,7 @@ thread_syscall_return( == (SYSCALL_CLASS_MACH << SYSCALL_CLASS_SHIFT); if (kdebug_enable && is_mach) { /* Mach trap */ - KERNEL_DEBUG_CONSTANT( + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, MACHDBG_CODE(DBG_MACH_EXCP_SC,code)|DBG_FUNC_END, ret, 0, 0, 0, 0); } @@ -190,7 +175,7 @@ thread_syscall_return( is_mach = (code < 0); if (kdebug_enable && is_mach) { /* Mach trap */ - KERNEL_DEBUG_CONSTANT( + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, MACHDBG_CODE(DBG_MACH_EXCP_SC,-code)|DBG_FUNC_END, ret, 0, 0, 0, 0); } @@ -206,50 +191,19 @@ thread_syscall_return( ret); #endif } - throttle_lowpri_io(TRUE); - - 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); - - pal_register_cache_state(thr_act, DIRTY); - if (is_saved_state64(iss)) { - x86_saved_state64_t *regs; - - regs = saved_state64(iss); +#if DEBUG || DEVELOPMENT + kern_allocation_name_t + prior __assert_only = thread_get_kernel_state(thr_act)->allocation_name; + assertf(prior == NULL, "thread_set_allocation_name(\"%s\") not cleared", kern_allocation_get_name(prior)); +#endif /* DEBUG || DEVELOPMENT */ - if (kdb_trap(regs->isf.trapno, (int)regs->isf.err, (void *)regs)) { - thread_exception_return(); - /*NOTREACHED*/ - } + throttle_lowpri_io(1); - } 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 */ static inline void user_page_fault_continue( @@ -258,61 +212,20 @@ user_page_fault_continue( thread_t thread = current_thread(); user_addr_t vaddr; -#if MACH_KDB - x86_saved_state_t *regs = USER_STATE(thread); - int err; - int trapno; - - assert((is_saved_state32(regs) && !thread_is_64bit(thread)) || - (is_saved_state64(regs) && thread_is_64bit(thread))); -#endif - - if (thread_is_64bit(thread)) { - x86_saved_state64_t *uregs; + if (thread_is_64bit_addr(thread)) { + x86_saved_state64_t *uregs; uregs = USER_REGS64(thread); -#if MACH_KDB - trapno = uregs->isf.trapno; - err = (int)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 (__probable((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, - saved_state32(regs))) - kdb_trap(T_WATCHPOINT, 0, saved_state32(regs)); -#endif /* MACH_KDB */ - thread_exception_return(); - /*NOTREACHED*/ - } - -#if MACH_KDB - if (debug_all_traps_with_kdb && - kdb_trap(trapno, err, saved_state32(regs))) { - thread_exception_return(); - /*NOTREACHED*/ - } -#endif /* MACH_KDB */ /* PAL debug hook */ pal_dbg_page_fault( thread, vaddr, kr ); @@ -352,9 +265,6 @@ kprint_state(x86_saved_state64_t *saved_state) 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(" v_arg6 0x%llx\n", saved_state->v_arg6); - kprintf(" v_arg7 0x%llx\n", saved_state->v_arg7); - kprintf(" v_arg8 0x%llx\n", saved_state->v_arg8); kprintf(" cr2 0x%llx\n", saved_state->cr2); kprintf("real cr2 0x%lx\n", get_cr2()); @@ -427,6 +337,9 @@ void interrupt_populate_latency_stats(char *buf, unsigned bufsize) { 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, @@ -441,63 +354,89 @@ interrupt(x86_saved_state_t *state) boolean_t user_mode = FALSE; int ipl; int cnum = cpu_number(); - - 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; - } - - KERNEL_DEBUG_CONSTANT( + cpu_data_t *cdp = cpu_data_ptr[cnum]; + int itype = DBG_INTR_TYPE_UNKNOWN; + + x86_saved_state64_t *state64 = saved_state64(state); + rip = state64->isf.rip; + rsp = state64->isf.rsp; + interrupt_num = state64->isf.trapno; + if(state64->isf.cs & 0x03) + user_mode = TRUE; + + 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 = DBG_INTR_TYPE_IPI; + else if (interrupt_num == (LAPIC_DEFAULT_INTERRUPT_BASE + LAPIC_TIMER_INTERRUPT)) + itype = DBG_INTR_TYPE_TIMER; + else + itype = DBG_INTR_TYPE_OTHER; + + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_START, - interrupt_num, rip, user_mode, 0, 0); + interrupt_num, + (user_mode ? rip : VM_KERNEL_UNSLIDE(rip)), + user_mode, itype, 0); SCHED_STATS_INTERRUPT(current_processor()); - ipl = get_preemption_level(); +#if CONFIG_TELEMETRY + if (telemetry_needs_record) { + telemetry_mark_curthread(user_mode, FALSE); + } +#endif + ipl = get_preemption_level(); + /* * Handle local APIC interrupts * else call platform expert for devices. */ - if (!lapic_interrupt(interrupt_num, state)) + 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()); } - KERNEL_DEBUG_CONSTANT( - MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_END, - interrupt_num, 0, 0, 0, 0); - if (cpu_data_ptr[cnum]->cpu_nested_istack) { - cpu_data_ptr[cnum]->cpu_nested_istack_events++; + if (__improbable(cdp->cpu_nested_istack)) { + cdp->cpu_nested_istack_events++; } else { - uint64_t int_latency = mach_absolute_time() - cpu_data_ptr[cnum]->cpu_int_event_time; - if (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, cpu_data_ptr[cnum]->cpu_prior_signals, cpu_data_ptr[cnum]->cpu_signals); + 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 (int_latency > cpu_data_ptr[cnum]->cpu_max_observed_int_latency) { - cpu_data_ptr[cnum]->cpu_max_observed_int_latency = int_latency; - cpu_data_ptr[cnum]->cpu_max_observed_int_latency_vector = interrupt_num; + + 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; } } @@ -505,17 +444,29 @@ interrupt(x86_saved_state_t *state) * Having serviced the interrupt first, look at the interrupted stack depth. */ if (!user_mode) { - uint64_t depth = cpu_data_ptr[cnum]->cpu_kernel_stack - + sizeof(struct x86_kernel_state) + uint64_t depth = cdp->cpu_kernel_stack + + sizeof(struct thread_kernel_state) + sizeof(struct i386_exception_link *) - rsp; - if (depth > kernel_stack_depth_max) { + 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) rip, 0, 0, 0); + (long) depth, (long) VM_KERNEL_UNSLIDE(rip), 0, 0, 0); } } + + if (cnum == master_cpu) + ml_entropy_collect(); + +#if KPERF + kperf_interrupt(); +#endif /* KPERF */ + + KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_END, + interrupt_num); + + assert(ml_get_interrupts_enabled() == FALSE); } static inline void @@ -541,18 +492,14 @@ kernel_trap( x86_saved_state_t *state, uintptr_t *lo_spp) { -#ifdef __i386__ - x86_saved_state32_t *saved_state; -#else x86_saved_state64_t *saved_state; -#endif int code; user_addr_t vaddr; int type; 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; @@ -560,29 +507,11 @@ kernel_trap( #if NCOPY_WINDOWS > 0 int fault_in_copy_window = -1; #endif - int is_user = 0; -#if MACH_KDB - pt_entry_t *pte; -#endif /* MACH_KDB */ - - thread = current_thread(); + int is_user; + int trap_pl = get_preemption_level(); -#ifdef __i386__ - if (__improbable(is_saved_state64(state))) { - panic_64(state, 0, "Kernel trap with 64-bit state", FALSE); - } - - saved_state = saved_state32(state); - - /* Record cpu where state was captured (trampolines don't set this) */ - saved_state->cpu = cpu_number(); + thread = current_thread(); - 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 */ - kern_ip = (vm_offset_t)saved_state->eip; -#else if (__improbable(is_saved_state32(state))) panic("kernel_trap(%p) with 32-bit state", state); saved_state = saved_state64(state); @@ -595,31 +524,13 @@ kernel_trap( 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; -#endif - myast = ast_pending(); - - perfASTCallback astfn = perfASTHook; - if (__improbable(astfn != NULL)) { - if (*myast & AST_CHUD_ALL) - astfn(AST_CHUD_ALL, myast); - } else - *myast &= ~AST_CHUD_ALL; - - /* - * Is there a hook? - */ - perfCallback fn = perfTrapHook; - if (__improbable(fn != NULL)) { - if (fn(type, NULL, 0, 0) == KERN_SUCCESS) { - /* - * If it succeeds, we are done... - */ - return; - } - } + is_user = (vaddr < VM_MAX_USER_PAGE_ADDRESS); #if CONFIG_DTRACE + /* + * Is there a DTrace hook? + */ if (__improbable(tempDTraceTrapHook != NULL)) { if (tempDTraceTrapHook(type, state, lo_spp, 0) == KERN_SUCCESS) { /* @@ -636,13 +547,21 @@ kernel_trap( * as soon we possibly can to hold latency down */ if (__improbable(T_PREEMPT == type)) { - ast_taken(AST_PREEMPTION, FALSE); + ast_taken_kernel(); - KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86, type)) | DBG_FUNC_NONE, - 0, 0, 0, kern_ip, 0); + 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 @@ -677,13 +596,34 @@ 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; - is_user = -1; + + /* 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 @@ -697,16 +637,15 @@ kernel_trap( 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, - (unsigned)(vaddr >> 32), (unsigned)vaddr, is_user, kern_ip, 0); - - (void) ml_set_interrupts_enabled(intr); switch (type) { @@ -724,14 +663,15 @@ kernel_trap( return; case T_SSE_FLOAT_ERROR: - fpSSEexterrflt(); + fpSSEexterrflt(); return; - case T_DEBUG: -#ifdef __i386__ - if ((saved_state->efl & EFL_TF) == 0 && NO_WATCHPOINTS) -#else + + case T_INVALID_OPCODE: + fpUDflt(kern_ip); + goto debugger_entry; + + case T_DEBUG: if ((saved_state->isf.rflags & EFL_TF) == 0 && NO_WATCHPOINTS) -#endif { /* We've somehow encountered a debug * register match that does not belong @@ -742,30 +682,10 @@ kernel_trap( return; } goto debugger_entry; -#ifdef __x86_64__ case T_INT3: goto debugger_entry; -#endif case T_PAGE_FAULT: -#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 */ - #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? */ @@ -778,35 +698,20 @@ kernel_trap( } } #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 - result = vm_fault(map, - vm_map_trunc_page(vaddr), + fault_result = result = vm_fault(map, + vaddr, prot, - FALSE, + FALSE, VM_KERN_MEMORY_NONE, 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) { @@ -867,99 +772,31 @@ debugger_entry: * access through the debugger. */ sync_iss_to_iks(state); -#if MACH_KDB -restart_debugger: -#endif /* MACH_KDB */ #if MACH_KDP - if (current_debugger != KDB_CUR_DB) { - if (kdp_i386_trap(type, saved_state, result, (vm_offset_t)vaddr)) - return; - } else { -#endif /* MACH_KDP */ -#if MACH_KDB - 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 MACH_KDP - } + if (kdp_i386_trap(type, saved_state, result, (vm_offset_t)vaddr)) + return; #endif } - __asm__ volatile("cli":::"cc"); - panic_trap(saved_state); + pal_cli(); + panic_trap(saved_state, trap_pl, fault_result); /* * NO RETURN */ } - -#ifdef __i386__ -static void -set_recovery_ip(x86_saved_state32_t *saved_state, vm_offset_t ip) -{ - saved_state->eip = ip; -} -#else static void set_recovery_ip(x86_saved_state64_t *saved_state, vm_offset_t ip) { saved_state->isf.rip = ip; } -#endif - - -#ifdef __i386__ -static void -panic_trap(x86_saved_state32_t *regs) -{ - const char *trapname = "Unknown"; - pal_cr_t cr0, cr2, cr3, cr4; - - pal_get_control_registers( &cr0, &cr2, &cr3, &cr4 ); - - /* - * 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", - cr0, cr2, cr3, cr4); - - if (regs->trapno < TRAP_TYPES) - trapname = trap_type[regs->trapno]; -#undef panic - panic("Kernel trap at 0x%08x, 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" - "Error code: 0x%08x\n", - regs->eip, 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 & 0xFFFF, regs->ds & 0xFFFF, regs->err); - /* - * This next statement is not executed, - * but it's needed to stop the compiler using tail call optimization - * for the panic call - which confuses the subsequent backtrace. - */ - cr0 = 0; -} -#else - static void -panic_trap(x86_saved_state64_t *regs) +panic_trap(x86_saved_state64_t *regs, uint32_t pl, kern_return_t fault_result) { 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); @@ -970,14 +807,28 @@ panic_trap(x86_saved_state64_t *regs) */ panic_io_port_read(); - kprintf("panic trap number 0x%x, rip 0x%016llx\n", - regs->isf.trapno, regs->isf.rip); + kprintf("CPU %d panic trap number 0x%x, rip 0x%016llx\n", + cpu_number(), 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->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" @@ -986,7 +837,7 @@ panic_trap(x86_saved_state64_t *regs) "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" - "CR2: 0x%016llx, Error code: 0x%016llx, Faulting CPU: 0x%x%s\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, @@ -995,7 +846,12 @@ panic_trap(x86_saved_state64_t *regs) 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 @@ -1003,12 +859,16 @@ panic_trap(x86_saved_state64_t *regs) */ cr0 = 0; } -#endif #if CONFIG_DTRACE extern kern_return_t dtrace_user_probe(x86_saved_state_t *); #endif +#if DEBUG +uint32_t fsigs[2]; +uint32_t fsigns, fsigcs; +#endif + /* * Trap from user mode. */ @@ -1024,13 +884,12 @@ user_trap( 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))); + assert((is_saved_state32(saved_state) && !thread_is_64bit_addr(thread)) || + (is_saved_state64(saved_state) && thread_is_64bit_addr(thread))); if (is_saved_state64(saved_state)) { x86_saved_state64_t *regs; @@ -1069,7 +928,7 @@ user_trap( pal_sti(); - KERNEL_DEBUG_CONSTANT( + 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); @@ -1078,31 +937,14 @@ user_trap( subcode = 0; exc = 0; -#if DEBUG_TRACE - kprintf("user_trap(0x%08x) type=%d vaddr=0x%016llx\n", - saved_state, type, vaddr); -#endif - - perfASTCallback astfn = perfASTHook; - if (__improbable(astfn != NULL)) { - myast = ast_pending(); - if (*myast & AST_CHUD_ALL) { - astfn(AST_CHUD_ALL, myast); - } - } - - /* Is there a hook? */ - 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); @@ -1129,7 +971,7 @@ user_trap( * because the high order bits are not * used on x86_64 */ - if (thread_is_64bit(thread)) { + if (thread_is_64bit_addr(thread)) { x86_debug_state64_t *ids = pcb->ids; ids->dr6 = dr6; } else { /* 32 bit thread */ @@ -1161,6 +1003,9 @@ user_trap( break; case T_INVALID_OPCODE: +#if !defined(RC_HIDE_XNU_J137) + fpUDflt(rip); /* May return from exception directly */ +#endif exc = EXC_BAD_INSTRUCTION; code = EXC_I386_INVOP; break; @@ -1212,21 +1057,47 @@ user_trap( 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 (__improbable(err & T_PF_EXECUTE)) prot |= VM_PROT_EXECUTE; +#if DEVELOPMENT || DEBUG + uint32_t fsig = 0; + fsig = thread_fpsimd_hash(thread); +#if DEBUG + fsigs[0] = fsig; +#endif +#endif + kret = vm_fault(thread->map, + vaddr, + prot, FALSE, VM_KERN_MEMORY_NONE, + THREAD_ABORTSAFE, NULL, 0); +#if DEVELOPMENT || DEBUG + if (fsig) { + uint32_t fsig2 = thread_fpsimd_hash(thread); +#if DEBUG + fsigcs++; + fsigs[1] = fsig2; +#endif + if (fsig != fsig2) { + panic("FP/SIMD state hash mismatch across fault thread: %p 0x%x->0x%x", thread, fsig, fsig2); + } + } else { +#if DEBUG + fsigns++; +#endif + } #endif - kret = vm_fault(thread->map, vm_map_trunc_page(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: @@ -1252,14 +1123,6 @@ user_trap( break; default: -#if MACH_KGDB - Debugger("Unanticipated user trap"); - return; -#endif /* MACH_KGDB */ -#if MACH_KDB - if (kdb_trap(type, err, saved_state32(saved_state))) - return; -#endif /* MACH_KDB */ panic("Unexpected user trap, type %d", type); return; } @@ -1270,29 +1133,6 @@ user_trap( /* NOTREACHED */ } - -/* - * Handle AST traps for i386. - */ - -extern void log_thread_action (thread_t, char *); - -void -i386_astintr(int preemption) -{ - ast_t mask = AST_ALL; - spl_t s; - - if (preemption) - mask = AST_PREEMPTION; - - s = splsched(); - - ast_taken(mask, s); - - splx(s); -} - /* * Handle exceptions for i386. * @@ -1320,79 +1160,33 @@ i386_exception( } -#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_state_t *saved_state) { - struct x86_kernel_state *iks; + struct x86_kernel_state *iks = NULL; vm_offset_t kstack; boolean_t record_active_regs = FALSE; /* The PAL may have a special way to sync registers */ - if( saved_state->flavor == THREAD_STATE_NONE ) + if (saved_state && saved_state->flavor == THREAD_STATE_NONE) pal_get_kern_regs( saved_state ); - if ((kstack = current_thread()->kernel_stack) != 0) { -#ifdef __i386__ - x86_saved_state32_t *regs = saved_state32(saved_state); -#else + if (current_thread() != NULL && + (kstack = current_thread()->kernel_stack) != 0) { x86_saved_state64_t *regs = saved_state64(saved_state); -#endif iks = STACK_IKS(kstack); /* Did we take the trap/interrupt in kernel mode? */ -#ifdef __i386__ - if (regs == USER_REGS32(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; - } -#else - if (regs == USER_REGS64(current_thread())) + if (saved_state == NULL || /* NULL => polling in kernel */ + regs == USER_REGS64(current_thread())) record_active_regs = TRUE; else { iks->k_rbx = regs->rbx; @@ -1404,20 +1198,9 @@ sync_iss_to_iks(x86_saved_state_t *saved_state) iks->k_r15 = regs->r15; iks->k_rip = regs->isf.rip; } -#endif } if (record_active_regs == TRUE) { -#ifdef __i386__ - /* 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)); -#else /* Show the trap handler path */ __asm__ volatile("movq %%rbx, %0" : "=m" (iks->k_rbx)); __asm__ volatile("movq %%rsp, %0" : "=m" (iks->k_rsp)); @@ -1431,7 +1214,6 @@ sync_iss_to_iks(x86_saved_state_t *saved_state) : "=m" (iks->k_rip) : : "rax"); -#endif } } @@ -1448,16 +1230,6 @@ sync_iss_to_iks_unconditionally(__unused x86_saved_state_t *saved_state) { if ((kstack = current_thread()->kernel_stack) != 0) { iks = STACK_IKS(kstack); -#ifdef __i386__ - /* Display 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)); -#else /* Display the trap handler path */ __asm__ volatile("movq %%rbx, %0" : "=m" (iks->k_rbx)); __asm__ volatile("movq %%rsp, %0" : "=m" (iks->k_rsp)); @@ -1468,6 +1240,33 @@ sync_iss_to_iks_unconditionally(__unused x86_saved_state_t *saved_state) { __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 +#define TERI 1 #endif + +#if TERI +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_addr(thread) != task_has_64Bit_addr(thread->task)) { + panic("Task/thread bitness mismatch %p %p, task: %d, thread: %d", thread, thread->task, thread_is_64bit_addr(thread), task_has_64Bit_addr(thread->task)); } + + if (thread_is_64bit_addr(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