/*
- * Copyright (c) 2000-2008 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
* Hardware trap/fault handler.
*/
-#include <mach_kdb.h>
-#include <mach_kgdb.h>
#include <mach_kdp.h>
#include <mach_ldebug.h>
#include <sys/kdebug.h>
-#if MACH_KGDB
-#include <kgdb/kgdb_defs.h>
-#endif /* MACH_KGDB */
-
-#if MACH_KDB
-#include <debug.h>
-#include <ddb/db_watch.h>
-#include <ddb/db_run.h>
-#include <ddb/db_break.h>
-#include <ddb/db_trap.h>
-#endif /* MACH_KDB */
-
#include <string.h>
#include <i386/postcode.h>
#include <mach/i386/syscall_sw.h>
#include <libkern/OSDebug.h>
+#include <i386/cpu_threads.h>
+#include <machine/pal_routines.h>
-extern void throttle_lowpri_io(boolean_t);
-
+extern void throttle_lowpri_io(int);
+extern void kprint_state(x86_saved_state64_t *saved_state);
/*
* Forward declarations
#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 set_recovery_ip(x86_saved_state64_t *saved_state, vm_offset_t ip);
#endif
volatile perfCallback perfTrapHook = NULL; /* Pointer to CHUD trap hook routine */
-volatile perfCallback perfASTHook = NULL; /* Pointer to CHUD AST hook routine */
#if CONFIG_DTRACE
/* See <rdar://problem/4613924> */
extern boolean_t dtrace_tally_fault(user_addr_t);
#endif
+extern boolean_t pmap_smep_enabled;
+
void
thread_syscall_return(
kern_return_t ret)
boolean_t is_mach;
int code;
+ pal_register_cache_state(thr_act, DIRTY);
if (thread_is_64bit(thr_act)) {
x86_saved_state64_t *regs;
== (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);
}
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);
}
}
-#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);
-
-
- 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*/
- }
- }
-}
-
-#endif /* MACH_KDB */
-
-void
+static inline void
user_page_fault_continue(
kern_return_t kr)
{
thread_t thread = current_thread();
- ast_t *myast;
- boolean_t intr;
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(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 ((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 */
- 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, saved_state32(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*/
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
-static void
-print_state(x86_saved_state64_t *saved_state)
+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(" isf.rsp 0x%llx\n", saved_state->isf.rsp);
kprintf(" isf.ss 0x%llx\n", saved_state->isf.ss);
}
+#endif
+
+
/*
- * K64 debug - fatal handler for debug code in the trap vectors.
+ * Non-zero indicates latency assert is enabled and capped at valued
+ * absolute time units.
*/
-extern void
-panic_idt64(x86_saved_state_t *rsp);
+
+uint64_t interrupt_latency_cap = 0;
+boolean_t ilat_assert = FALSE;
+
void
-panic_idt64(x86_saved_state_t *rsp)
-{
- print_state(saved_state64(rsp));
- panic("panic_idt64");
+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));
}
-#endif
-extern void PE_incoming_interrupt(int interrupt);
+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);
+}
/*
* Handle interrupts:
uint64_t rsp;
int interrupt_num;
boolean_t user_mode = FALSE;
+ int ipl;
+ int cnum = cpu_number();
+ int itype = 0;
-
- if (is_saved_state64(state) == TRUE) {
+ if (is_saved_state64(state) == TRUE) {
x86_saved_state64_t *state64;
state64 = saved_state64(state);
interrupt_num = state32->trapno;
}
- KERNEL_DEBUG_CONSTANT(
+ 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, (long) 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();
+
/*
* Handle local APIC interrupts
* else call platform expert for devices.
- */
+ */
if (!lapic_interrupt(interrupt_num, state))
PE_incoming_interrupt(interrupt_num);
- KERNEL_DEBUG_CONSTANT(
+ 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_IST(KDEBUG_TRACE,
MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_END,
- 0, 0, 0, 0, 0);
+ interrupt_num, 0, 0, 0, 0);
+
+ if (cpu_data_ptr[cnum]->cpu_nested_istack) {
+ cpu_data_ptr[cnum]->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);
+ }
+ 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;
+ }
+ }
/*
* Having serviced the interrupt first, look at the interrupted stack depth.
*/
if (!user_mode) {
- uint64_t depth = current_cpu_datap()->cpu_kernel_stack
+ uint64_t depth = cpu_data_ptr[cnum]->cpu_kernel_stack
+ sizeof(struct x86_kernel_state)
+ sizeof(struct i386_exception_link *)
- rsp;
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);
}
}
}
void
kernel_trap(
- x86_saved_state_t *state)
+ x86_saved_state_t *state,
+ uintptr_t *lo_spp)
{
#ifdef __i386__
x86_saved_state32_t *saved_state;
int fault_in_copy_window = -1;
#endif
int is_user = 0;
-#if MACH_KDB
- pt_entry_t *pte;
-#endif /* MACH_KDB */
thread = current_thread();
#ifdef __i386__
- if (is_saved_state64(state))
- panic("kernel_trap(%p) with 64-bit state", state);
+ 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();
+
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 (is_saved_state32(state))
+ if (__improbable(is_saved_state32(state)))
panic("kernel_trap(%p) with 32-bit state", state);
saved_state = saved_state64(state);
+
+ /* Record cpu where state was captured */
+ saved_state->isf.cpu = cpu_number();
+
vaddr = (user_addr_t)saved_state->cr2;
type = saved_state->isf.trapno;
code = (int)(saved_state->isf.err & 0xffff);
myast = ast_pending();
- perfCallback fn = perfASTHook;
- if (fn) {
+ perfASTCallback astfn = perfASTHook;
+ if (__improbable(astfn != NULL)) {
if (*myast & AST_CHUD_ALL)
- fn(type, NULL, 0, 0);
+ astfn(AST_CHUD_ALL, myast);
} else
*myast &= ~AST_CHUD_ALL;
/*
* Is there a hook?
*/
- fn = perfTrapHook;
- if (fn) {
+ perfCallback fn = perfTrapHook;
+ if (__improbable(fn != NULL)) {
if (fn(type, NULL, 0, 0) == KERN_SUCCESS) {
/*
* If it succeeds, we are done...
}
#if CONFIG_DTRACE
- if (tempDTraceTrapHook) {
- if (tempDTraceTrapHook(type, state, 0, 0) == KERN_SUCCESS) {
+ if (__improbable(tempDTraceTrapHook != NULL)) {
+ if (tempDTraceTrapHook(type, state, lo_spp, 0) == KERN_SUCCESS) {
/*
* If it succeeds, we are done...
*/
* on preemption below. but we do want to re-enable interrupts
* as soon we possibly can to hold latency down
*/
- if (T_PREEMPT == type) {
+ if (__improbable(T_PREEMPT == type)) {
ast_taken(AST_PREEMPTION, FALSE);
- 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;
}
*/
map = kernel_map;
- if (thread != THREAD_NULL && thread->map != kernel_map) {
+ if (__probable(thread != THREAD_NULL && thread->map != kernel_map)) {
#if NCOPY_WINDOWS > 0
vm_offset_t copy_window_base;
vm_offset_t kvaddr;
* 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) {
is_user = -1;
}
#else
- if (vaddr < VM_MAX_USER_PAGE_ADDRESS) {
+ 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;
+ }
+
/*
* If we're not sharing cr3 with the user
* and we faulted in copyio,
*/
if (no_shared_cr3 &&
(thread->machine.specFlags&CopyIOActive) &&
- map->pmap->pm_cr3 != get_cr3()) {
- set_cr3(map->pmap->pm_cr3);
+ 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;
}
}
#endif
}
}
- KERNEL_DEBUG_CONSTANT(
+ 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)(vaddr >> 32), (unsigned)vaddr, is_user, kern_ip, 0);
+ (unsigned)(kd_vaddr >> 32), (unsigned)kd_vaddr, is_user,
+ VM_KERNEL_UNSLIDE(kern_ip), 0);
(void) ml_set_interrupts_enabled(intr);
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.
- */
-
- 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 */
#if CONFIG_DTRACE
- if (thread->options & TH_OPT_DTRACE) { /* Executing under dtrace_probe? */
+ 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
}
}
#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),
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%qx, kpdp = 0x%qx\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 */
#endif /* CONFIG_DTRACE */
case T_GENERAL_PROTECTION:
-#if defined(__x86_64__) && DEBUG
- print_state(saved_state);
-#endif
/*
* If there is a failure recovery address
* for this fault, go there.
/*
* Check thread recovery address also.
*/
- if (thread->recover) {
+ if (thread != THREAD_NULL && thread->recover) {
set_recovery_ip(saved_state, thread->recover);
thread->recover = 0;
return;
*
* fall through...
*/
-
default:
/*
* Exception 15 is reserved but some chips may generate it
* 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 (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
}
#endif
}
-
+ pal_cli();
panic_trap(saved_state);
/*
* NO RETURN
panic_trap(x86_saved_state32_t *regs)
{
const char *trapname = "Unknown";
- uint32_t cr0 = get_cr0();
- uint32_t cr2 = get_cr2();
- uint32_t cr3 = get_cr3();
- uint32_t cr4 = get_cr4();
+ 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.
"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",
+ "Error code: 0x%08x%s\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, regs->ds, regs->err);
+ regs->efl,regs->eip,regs->cs & 0xFFFF, regs->ds & 0xFFFF, regs->err,
+ virtualized ? " VMM" : "");
/*
* This next statement is not executed,
* but it's needed to stop the compiler using tail call optimization
cr0 = 0;
}
#else
+
+
static void
panic_trap(x86_saved_state64_t *regs)
{
const char *trapname = "Unknown";
- uint64_t cr0 = get_cr0();
- uint64_t cr2 = get_cr2();
- uint64_t cr3 = get_cr3();
- uint64_t cr4 = get_cr4();
+ pal_cr_t cr0, cr2, cr3, cr4;
+ boolean_t potential_smep_fault = FALSE, potential_kernel_NX_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.
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;
+ }
+ }
+
#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"
"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"
- "Error code: 0x%016llx\n",
+ "Fault CR2: 0x%016llx, Error code: 0x%016llx, Fault CPU: 0x%x%s%s%s\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, regs->isf.ss,
- regs->isf.err);
+ 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" : "");
/*
* This next statement is not executed,
* but it's needed to stop the compiler using tail call optimization
}
#endif
-extern void kprintf_break_lock(void);
-
-#ifdef __i386__
-static void
-panic_32(__unused int code, __unused int pc, __unused const char *msg, boolean_t do_mca_dump, boolean_t do_bt)
-{
- struct i386_tss *my_ktss = current_ktss();
-
- /* Set postcode (DEBUG only) */
- postcode(pc);
-
- /*
- * 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 (do_mca_dump) {
-#if CONFIG_MCA
- /*
- * Dump the contents of the machine check MSRs (if any).
- */
- mca_dump();
-#endif
- }
-
-#if MACH_KDP
- /*
- * Print backtrace leading to first fault:
- */
- if (do_bt)
- panic_i386_backtrace((void *) my_ktss->ebp, 10, NULL, FALSE, NULL);
-#endif
-
- panic("%s at 0x%08x, 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",
- msg,
- my_ktss->eip, current_thread(), code,
- (uint32_t)get_cr0(), (uint32_t)get_cr2(), (uint32_t)get_cr3(), (uint32_t)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 double-fault
- * is taken in kernel space.
- * Kernel stack overflow is one route here.
- */
-void
-panic_double_fault32(int code)
-{
- panic_32(code, PANIC_DOUBLE_FAULT, "Double fault", FALSE, TRUE);
-}
-
-/*
- * Called from locore on a special reserved stack after a machine-check
- */
-void
-panic_machine_check32(int code)
-{
- panic_32(code, PANIC_MACHINE_CHECK, "Machine-check", TRUE, FALSE);
-}
-#endif /* __i386__ */
-
-static void
-panic_64(x86_saved_state_t *sp, __unused int pc, __unused const char *msg, boolean_t do_mca_dump)
-{
- /* Set postcode (DEBUG only) */
- postcode(pc);
-
- /*
- * 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 (do_mca_dump) {
-#if CONFIG_MCA
- /*
- * Dump the contents of the machine check MSRs (if any).
- */
- mca_dump();
-#endif
- }
-
-#ifdef __i386__
- /*
- * Dump the interrupt stack frame at last kernel entry.
- */
- if (is_saved_state64(sp)) {
- x86_saved_state64_t *ss64p = saved_state64(sp);
- panic("%s 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",
- msg,
- current_thread(), ss64p->isf.trapno, ss64p->isf.err,
- (uint32_t)get_cr0(), (uint32_t)get_cr2(), (uint32_t)get_cr3(), (uint32_t)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(sp);
- panic("%s at 0x%08x, 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",
- msg,
- ss32p->eip, current_thread(), ss32p->trapno, ss32p->err,
- (uint32_t)get_cr0(), (uint32_t)get_cr2(), (uint32_t)get_cr3(), (uint32_t)get_cr4(),
- ss32p->eax, ss32p->ebx, ss32p->ecx, ss32p->edx,
- ss32p->uesp, ss32p->ebp, ss32p->esi, ss32p->edi,
- ss32p->efl, ss32p->eip);
- }
-#else
- x86_saved_state64_t *regs = saved_state64(sp);
- panic("%s thread:%p at 0x%016llx, registers:\n"
- "CR0: 0x%016lx, CR2: 0x%016lx, CR3: 0x%016lx, CR4: 0x%016lx\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"
- "Error code: 0x%016llx\n",
- msg,
- current_thread(), regs->isf.rip,
- get_cr0(), get_cr2(), get_cr3(), get_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, regs->isf.ss,
- regs->isf.err);
-#endif
-}
-
-void
-panic_double_fault64(x86_saved_state_t *sp)
-{
- panic_64(sp, PANIC_DOUBLE_FAULT, "Double fault", FALSE);
-
-}
-void
-
-panic_machine_check64(x86_saved_state_t *sp)
-{
- panic_64(sp, PANIC_MACHINE_CHECK, "Machine Check", TRUE);
-
-}
-
#if CONFIG_DTRACE
extern kern_return_t dtrace_user_probe(x86_saved_state_t *);
#endif
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)));
regs = saved_state64(saved_state);
+ /* Record cpu where state was captured */
+ regs->isf.cpu = cpu_number();
+
type = regs->isf.trapno;
err = (int)regs->isf.err & 0xffff;
vaddr = (user_addr_t)regs->cr2;
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(
+ 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);
kprintf("user_trap(0x%08x) type=%d vaddr=0x%016llx\n",
saved_state, type, vaddr);
#endif
- myast = ast_pending();
- perfCallback fn = perfASTHook;
- if (fn) {
+
+ perfASTCallback astfn = perfASTHook;
+ if (__improbable(astfn != NULL)) {
+ myast = ast_pending();
if (*myast & AST_CHUD_ALL) {
- fn(type, saved_state, 0, 0);
+ astfn(AST_CHUD_ALL, myast);
}
- } else {
- *myast &= ~AST_CHUD_ALL;
}
/* Is there a hook? */
- fn = perfTrapHook;
- if (fn) {
+ perfCallback fn = perfTrapHook;
+ if (__improbable(fn != NULL)) {
if (fn(type, saved_state, 0, 0) == KERN_SUCCESS)
return; /* If it succeeds, we are done... */
}
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:
case T_DEBUG:
{
pcb_t pcb;
- long clear = 0; /* 32 bit for i386, 64 bit for x86_64 */
/*
- * 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
* because the high order bits are not
* used on x86_64
*/
- unsigned long dr6_temp; /* 32 bit for i386, 64 bit for x86_64 */
- __asm__ volatile ("mov %%db6, %0" : "=r" (dr6_temp)); /* Register constraint by necessity */
if (thread_is_64bit(thread)) {
x86_debug_state64_t *ids = pcb->ids;
- ids->dr6 = dr6_temp;
+ ids->dr6 = dr6;
} else { /* 32 bit thread */
x86_debug_state32_t *ids = pcb->ids;
- ids->dr6 = (uint32_t) dr6_temp;
+ ids->dr6 = (uint32_t) dr6;
}
- __asm__ volatile ("mov %0, %%db6" : : "r" (clear));
}
exc = EXC_BREAKPOINT;
code = EXC_I386_SGL;
break;
case T_PAGE_FAULT:
+ {
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);
- user_page_fault_continue(kret);
-
+ if (__probable((kret == KERN_SUCCESS) || (kret == KERN_ABORTED))) {
+ thread_exception_return();
/* NOTREACHED */
+ }
+
+ user_page_fault_continue(kret);
+ } /* NOTREACHED */
break;
case T_SSE_FLOAT_ERROR:
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;
}
/*
* Handle AST traps for i386.
- * Check for delayed floating-point exception from
- * AT-bus machines.
*/
extern void log_thread_action (thread_t, char *);
}
-
-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 <ddb/db_output.h>
-
-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
* kernel_trap() prior to entering the debugger, and when receiving
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 )
+ pal_get_kern_regs( saved_state );
+
if ((kstack = current_thread()->kernel_stack) != 0) {
#ifdef __i386__
x86_saved_state32_t *regs = saved_state32(saved_state);
iks = STACK_IKS(kstack);
-
- /* Did we take the trap/interrupt in kernel mode? */
+ /* Did we take the trap/interrupt in kernel mode? */
#ifdef __i386__
if (regs == USER_REGS32(current_thread()))
record_active_regs = TRUE;
projects / apple / xnu.git / blobdiff