xref: /xnu-10063.121.3/osfmk/i386/trap.c (revision 2c2f96dc2b9a4408a43d3150ae9c105355ca3daa)
1 /*
2  * Copyright (c) 2000-2020 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 /*
29  * @OSF_COPYRIGHT@
30  */
31 /*
32  * Mach Operating System
33  * Copyright (c) 1991,1990,1989,1988 Carnegie Mellon University
34  * All Rights Reserved.
35  *
36  * Permission to use, copy, modify and distribute this software and its
37  * documentation is hereby granted, provided that both the copyright
38  * notice and this permission notice appear in all copies of the
39  * software, derivative works or modified versions, and any portions
40  * thereof, and that both notices appear in supporting documentation.
41  *
42  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
43  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
44  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
45  *
46  * Carnegie Mellon requests users of this software to return to
47  *
48  *  Software Distribution Coordinator  or  [email protected]
49  *  School of Computer Science
50  *  Carnegie Mellon University
51  *  Pittsburgh PA 15213-3890
52  *
53  * any improvements or extensions that they make and grant Carnegie Mellon
54  * the rights to redistribute these changes.
55  */
56 /*
57  */
58 
59 /*
60  * Hardware trap/fault handler.
61  */
62 
63 #include <mach_kdp.h>
64 #include <mach_ldebug.h>
65 
66 #include <types.h>
67 #include <i386/eflags.h>
68 #include <i386/trap_internal.h>
69 #include <i386/pmap.h>
70 #include <i386/fpu.h>
71 #include <i386/panic_notify.h>
72 #include <i386/lapic.h>
73 
74 #include <mach/exception.h>
75 #include <mach/kern_return.h>
76 #include <mach/vm_param.h>
77 #include <mach/i386/thread_status.h>
78 
79 #include <vm/vm_kern.h>
80 #include <vm/vm_fault.h>
81 
82 #include <kern/kern_types.h>
83 #include <kern/processor.h>
84 #include <kern/thread.h>
85 #include <kern/task.h>
86 #include <kern/restartable.h>
87 #include <kern/sched.h>
88 #include <kern/sched_prim.h>
89 #include <kern/exception.h>
90 #include <kern/spl.h>
91 #include <kern/misc_protos.h>
92 #include <kern/debug.h>
93 #if CONFIG_TELEMETRY
94 #include <kern/telemetry.h>
95 #endif
96 #include <kern/zalloc_internal.h>
97 #include <sys/kdebug.h>
98 #include <kperf/kperf.h>
99 #include <prng/random.h>
100 #include <prng/entropy.h>
101 
102 #include <string.h>
103 
104 #include <i386/postcode.h>
105 #include <i386/mp_desc.h>
106 #include <i386/proc_reg.h>
107 #include <i386/machine_routines.h>
108 #if CONFIG_MCA
109 #include <i386/machine_check.h>
110 #endif
111 #include <mach/i386/syscall_sw.h>
112 
113 #include <libkern/OSDebug.h>
114 #include <i386/cpu_threads.h>
115 #include <machine/pal_routines.h>
116 #include <i386/lbr.h>
117 
118 extern void throttle_lowpri_io(int);
119 extern void kprint_state(x86_saved_state64_t *saved_state);
120 #if DEVELOPMENT || DEBUG
121 int insnstream_force_cacheline_mismatch = 0;
122 extern int panic_on_cacheline_mismatch;
123 extern char panic_on_trap_procname[];
124 extern uint32_t panic_on_trap_mask;
125 #endif
126 
127 extern int insn_copyin_count;
128 
129 /*
130  * Forward declarations
131  */
132 static void panic_trap(x86_saved_state64_t *saved_state, uint32_t pl, kern_return_t fault_result) __dead2;
133 static void set_recovery_ip(x86_saved_state64_t *saved_state, vm_offset_t ip);
134 #if DEVELOPMENT || DEBUG
135 static __attribute__((noinline)) void copy_instruction_stream(thread_t thread, uint64_t rip, int trap_code, bool inspect_cacheline);
136 #else
137 static __attribute__((noinline)) void copy_instruction_stream(thread_t thread, uint64_t rip, int trap_code);
138 #endif
139 
140 #if CONFIG_DTRACE
141 /* See <rdar://problem/4613924> */
142 perfCallback tempDTraceTrapHook = NULL; /* Pointer to DTrace fbt trap hook routine */
143 
144 extern boolean_t dtrace_tally_fault(user_addr_t);
145 extern boolean_t dtrace_handle_trap(int, x86_saved_state_t *);
146 #endif
147 
148 #ifdef MACH_BSD
149 extern char *   proc_name_address(void *p);
150 #endif /* MACH_BSD */
151 
152 extern boolean_t pmap_smep_enabled;
153 extern boolean_t pmap_smap_enabled;
154 
155 __attribute__((noreturn))
156 void
thread_syscall_return(kern_return_t ret)157 thread_syscall_return(
158 	kern_return_t ret)
159 {
160 	thread_t        thr_act = current_thread();
161 	boolean_t       is_mach;
162 	int             code;
163 
164 	pal_register_cache_state(thr_act, DIRTY);
165 
166 	if (thread_is_64bit_addr(thr_act)) {
167 		x86_saved_state64_t     *regs;
168 
169 		regs = USER_REGS64(thr_act);
170 
171 		code = (int) (regs->rax & SYSCALL_NUMBER_MASK);
172 		is_mach = (regs->rax & SYSCALL_CLASS_MASK)
173 		    == (SYSCALL_CLASS_MACH << SYSCALL_CLASS_SHIFT);
174 		if (kdebug_enable && is_mach) {
175 			/* Mach trap */
176 			KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
177 			    MACHDBG_CODE(DBG_MACH_EXCP_SC, code) | DBG_FUNC_END,
178 			    ret, 0, 0, 0, 0);
179 		}
180 		regs->rax = ret;
181 #if DEBUG
182 		if (is_mach) {
183 			DEBUG_KPRINT_SYSCALL_MACH(
184 				"thread_syscall_return: 64-bit mach ret=%u\n",
185 				ret);
186 		} else {
187 			DEBUG_KPRINT_SYSCALL_UNIX(
188 				"thread_syscall_return: 64-bit unix ret=%u\n",
189 				ret);
190 		}
191 #endif
192 	} else {
193 		x86_saved_state32_t     *regs;
194 
195 		regs = USER_REGS32(thr_act);
196 
197 		code = ((int) regs->eax);
198 		is_mach = (code < 0);
199 		if (kdebug_enable && is_mach) {
200 			/* Mach trap */
201 			KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
202 			    MACHDBG_CODE(DBG_MACH_EXCP_SC, -code) | DBG_FUNC_END,
203 			    ret, 0, 0, 0, 0);
204 		}
205 		regs->eax = ret;
206 #if DEBUG
207 		if (is_mach) {
208 			DEBUG_KPRINT_SYSCALL_MACH(
209 				"thread_syscall_return: 32-bit mach ret=%u\n",
210 				ret);
211 		} else {
212 			DEBUG_KPRINT_SYSCALL_UNIX(
213 				"thread_syscall_return: 32-bit unix ret=%u\n",
214 				ret);
215 		}
216 #endif
217 	}
218 
219 #if DEBUG || DEVELOPMENT
220 	kern_allocation_name_t
221 	prior __assert_only = thread_get_kernel_state(thr_act)->allocation_name;
222 	assertf(prior == NULL, "thread_set_allocation_name(\"%s\") not cleared", kern_allocation_get_name(prior));
223 #endif /* DEBUG || DEVELOPMENT */
224 
225 	throttle_lowpri_io(1);
226 
227 	thread_exception_return();
228 	/*NOTREACHED*/
229 }
230 
231 /*
232  * Fault recovery in copyin/copyout routines.
233  */
234 struct recovery {
235 	uintptr_t       fault_addr;
236 	uintptr_t       recover_addr;
237 };
238 
239 extern struct recovery  recover_table[];
240 extern struct recovery  recover_table_end[];
241 
242 const char *    trap_type[] = {TRAP_NAMES};
243 unsigned        TRAP_TYPES = sizeof(trap_type) / sizeof(trap_type[0]);
244 
245 extern void     PE_incoming_interrupt(int interrupt);
246 
247 #if defined(__x86_64__) && DEBUG
248 void
kprint_state(x86_saved_state64_t * saved_state)249 kprint_state(x86_saved_state64_t        *saved_state)
250 {
251 	kprintf("current_cpu_datap() 0x%lx\n", (uintptr_t)current_cpu_datap());
252 	kprintf("Current GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_GS_BASE));
253 	kprintf("Kernel  GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_KERNEL_GS_BASE));
254 	kprintf("state at 0x%lx:\n", (uintptr_t) saved_state);
255 
256 	kprintf("      rdi    0x%llx\n", saved_state->rdi);
257 	kprintf("      rsi    0x%llx\n", saved_state->rsi);
258 	kprintf("      rdx    0x%llx\n", saved_state->rdx);
259 	kprintf("      r10    0x%llx\n", saved_state->r10);
260 	kprintf("      r8     0x%llx\n", saved_state->r8);
261 	kprintf("      r9     0x%llx\n", saved_state->r9);
262 
263 	kprintf("      cr2    0x%llx\n", saved_state->cr2);
264 	kprintf("real  cr2    0x%lx\n", get_cr2());
265 	kprintf("      r15    0x%llx\n", saved_state->r15);
266 	kprintf("      r14    0x%llx\n", saved_state->r14);
267 	kprintf("      r13    0x%llx\n", saved_state->r13);
268 	kprintf("      r12    0x%llx\n", saved_state->r12);
269 	kprintf("      r11    0x%llx\n", saved_state->r11);
270 	kprintf("      rbp    0x%llx\n", saved_state->rbp);
271 	kprintf("      rbx    0x%llx\n", saved_state->rbx);
272 	kprintf("      rcx    0x%llx\n", saved_state->rcx);
273 	kprintf("      rax    0x%llx\n", saved_state->rax);
274 
275 	kprintf("      gs     0x%x\n", saved_state->gs);
276 	kprintf("      fs     0x%x\n", saved_state->fs);
277 
278 	kprintf("  isf.trapno 0x%x\n", saved_state->isf.trapno);
279 	kprintf("  isf._pad   0x%x\n", saved_state->isf._pad);
280 	kprintf("  isf.trapfn 0x%llx\n", saved_state->isf.trapfn);
281 	kprintf("  isf.err    0x%llx\n", saved_state->isf.err);
282 	kprintf("  isf.rip    0x%llx\n", saved_state->isf.rip);
283 	kprintf("  isf.cs     0x%llx\n", saved_state->isf.cs);
284 	kprintf("  isf.rflags 0x%llx\n", saved_state->isf.rflags);
285 	kprintf("  isf.rsp    0x%llx\n", saved_state->isf.rsp);
286 	kprintf("  isf.ss     0x%llx\n", saved_state->isf.ss);
287 }
288 #endif
289 
290 
291 /*
292  * Non-zero indicates latency assert is enabled and capped at valued
293  * absolute time units.
294  */
295 
296 uint64_t interrupt_latency_cap = 0;
297 boolean_t ilat_assert = FALSE;
298 
299 void
interrupt_latency_tracker_setup(void)300 interrupt_latency_tracker_setup(void)
301 {
302 	uint32_t ilat_cap_us;
303 	if (PE_parse_boot_argn("interrupt_latency_cap_us", &ilat_cap_us, sizeof(ilat_cap_us))) {
304 		interrupt_latency_cap = ilat_cap_us * NSEC_PER_USEC;
305 		nanoseconds_to_absolutetime(interrupt_latency_cap, &interrupt_latency_cap);
306 	} else {
307 		interrupt_latency_cap = LockTimeOut;
308 	}
309 	PE_parse_boot_argn("-interrupt_latency_assert_enable", &ilat_assert, sizeof(ilat_assert));
310 }
311 
312 void
interrupt_reset_latency_stats(void)313 interrupt_reset_latency_stats(void)
314 {
315 	uint32_t i;
316 	for (i = 0; i < real_ncpus; i++) {
317 		cpu_data_ptr[i]->cpu_max_observed_int_latency =
318 		    cpu_data_ptr[i]->cpu_max_observed_int_latency_vector = 0;
319 	}
320 }
321 
322 void
interrupt_populate_latency_stats(char * buf,unsigned bufsize)323 interrupt_populate_latency_stats(char *buf, unsigned bufsize)
324 {
325 	uint32_t i, tcpu = ~0;
326 	uint64_t cur_max = 0;
327 
328 	for (i = 0; i < real_ncpus; i++) {
329 		if (cur_max < cpu_data_ptr[i]->cpu_max_observed_int_latency) {
330 			cur_max = cpu_data_ptr[i]->cpu_max_observed_int_latency;
331 			tcpu = i;
332 		}
333 	}
334 
335 	if (tcpu < real_ncpus) {
336 		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);
337 	}
338 }
339 
340 uint32_t interrupt_timer_coalescing_enabled = 1;
341 uint64_t interrupt_coalesced_timers;
342 
343 /*
344  * Handle interrupts:
345  *  - local APIC interrupts (IPIs, timers, etc) are handled by the kernel,
346  *  - device interrupts go to the platform expert.
347  */
348 void
interrupt(x86_saved_state_t * state)349 interrupt(x86_saved_state_t *state)
350 {
351 	uint64_t        rip;
352 	uint64_t        rsp;
353 	int             interrupt_num;
354 	boolean_t       user_mode = FALSE;
355 	int             ipl;
356 	int             cnum = cpu_number();
357 	cpu_data_t      *cdp = cpu_data_ptr[cnum];
358 	int             itype = DBG_INTR_TYPE_UNKNOWN;
359 	int             handled;
360 
361 
362 	x86_saved_state64_t     *state64 = saved_state64(state);
363 	rip = state64->isf.rip;
364 	rsp = state64->isf.rsp;
365 	interrupt_num = state64->isf.trapno;
366 	if (state64->isf.cs & 0x03) {
367 		user_mode = TRUE;
368 	}
369 
370 #if DEVELOPMENT || DEBUG
371 	uint64_t frameptr = is_saved_state64(state) ? state64->rbp : saved_state32(state)->ebp;
372 	uint32_t traptrace_index = traptrace_start(interrupt_num, rip, mach_absolute_time(), frameptr);
373 #endif
374 
375 	if (cpu_data_ptr[cnum]->lcpu.package->num_idle == topoParms.nLThreadsPerPackage) {
376 		cpu_data_ptr[cnum]->cpu_hwIntpexits[interrupt_num]++;
377 	}
378 
379 	if (interrupt_num == (LAPIC_DEFAULT_INTERRUPT_BASE + LAPIC_INTERPROCESSOR_INTERRUPT)) {
380 		itype = DBG_INTR_TYPE_IPI;
381 	} else if (interrupt_num == (LAPIC_DEFAULT_INTERRUPT_BASE + LAPIC_TIMER_INTERRUPT)) {
382 		itype = DBG_INTR_TYPE_TIMER;
383 	} else {
384 		itype = DBG_INTR_TYPE_OTHER;
385 	}
386 
387 	KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
388 	    MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_START,
389 	    interrupt_num,
390 	    (user_mode ? rip : VM_KERNEL_UNSLIDE(rip)),
391 	    user_mode, itype, 0);
392 
393 	SCHED_STATS_INC(interrupt_count);
394 
395 #if CONFIG_TELEMETRY
396 	if (telemetry_needs_record) {
397 		telemetry_mark_curthread(user_mode, FALSE);
398 	}
399 #endif
400 
401 	ipl = get_preemption_level();
402 
403 	/*
404 	 * Handle local APIC interrupts
405 	 * else call platform expert for devices.
406 	 */
407 	handled = lapic_interrupt(interrupt_num, state);
408 
409 	if (!handled) {
410 		if (interrupt_num == (LAPIC_DEFAULT_INTERRUPT_BASE + LAPIC_CMCI_INTERRUPT)) {
411 			/*
412 			 * CMCI can be signalled on any logical processor, and the kexts
413 			 * that implement handling CMCI use IOKit to register handlers for
414 			 * the CMCI vector, so if we see a CMCI, do not encode a CPU
415 			 * number in bits 8:31 (since the vector is the same regardless of
416 			 * the handling CPU).
417 			 */
418 			PE_incoming_interrupt(interrupt_num);
419 		} else if (cnum <= lapic_max_interrupt_cpunum) {
420 			PE_incoming_interrupt((cnum << 8) | interrupt_num);
421 		}
422 	}
423 
424 	if (__improbable(get_preemption_level() != ipl)) {
425 		panic("Preemption level altered by interrupt vector 0x%x: initial 0x%x, final: 0x%x", interrupt_num, ipl, get_preemption_level());
426 	}
427 
428 
429 	if (__improbable(cdp->cpu_nested_istack)) {
430 		cdp->cpu_nested_istack_events++;
431 	} else {
432 		uint64_t ctime = mach_absolute_time();
433 		uint64_t int_latency = ctime - cdp->cpu_int_event_time;
434 		uint64_t esdeadline, ehdeadline;
435 		/* Attempt to process deferred timers in the context of
436 		 * this interrupt, unless interrupt time has already exceeded
437 		 * TCOAL_ILAT_THRESHOLD.
438 		 */
439 #define TCOAL_ILAT_THRESHOLD (30000ULL)
440 
441 		if ((int_latency < TCOAL_ILAT_THRESHOLD) &&
442 		    interrupt_timer_coalescing_enabled) {
443 			esdeadline = cdp->rtclock_timer.queue.earliest_soft_deadline;
444 			ehdeadline = cdp->rtclock_timer.deadline;
445 			if ((ctime >= esdeadline) && (ctime < ehdeadline)) {
446 				interrupt_coalesced_timers++;
447 				TCOAL_DEBUG(0x88880000 | DBG_FUNC_START, ctime, esdeadline, ehdeadline, interrupt_coalesced_timers, 0);
448 				rtclock_intr(state);
449 				TCOAL_DEBUG(0x88880000 | DBG_FUNC_END, ctime, esdeadline, interrupt_coalesced_timers, 0, 0);
450 			} else {
451 				TCOAL_DEBUG(0x77770000, ctime, cdp->rtclock_timer.queue.earliest_soft_deadline, cdp->rtclock_timer.deadline, interrupt_coalesced_timers, 0);
452 			}
453 		}
454 
455 		if (__improbable(ilat_assert && (int_latency > interrupt_latency_cap) && !machine_timeout_suspended())) {
456 			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);
457 		}
458 
459 		if (__improbable(int_latency > cdp->cpu_max_observed_int_latency)) {
460 			cdp->cpu_max_observed_int_latency = int_latency;
461 			cdp->cpu_max_observed_int_latency_vector = interrupt_num;
462 		}
463 	}
464 
465 	/*
466 	 * Having serviced the interrupt first, look at the interrupted stack depth.
467 	 */
468 	if (!user_mode) {
469 		uint64_t depth = cdp->cpu_kernel_stack
470 		    + sizeof(struct thread_kernel_state)
471 		    + sizeof(struct i386_exception_link *)
472 		    - rsp;
473 		if (__improbable(depth > kernel_stack_depth_max)) {
474 			kernel_stack_depth_max = (vm_offset_t)depth;
475 			KERNEL_DEBUG_CONSTANT(
476 				MACHDBG_CODE(DBG_MACH_SCHED, MACH_STACK_DEPTH),
477 				(long) depth, (long) VM_KERNEL_UNSLIDE(rip), 0, 0, 0);
478 		}
479 	}
480 
481 	if (cnum == master_cpu) {
482 		entropy_collect();
483 	}
484 
485 #if KPERF
486 	kperf_interrupt();
487 #endif /* KPERF */
488 
489 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_END,
490 	    interrupt_num);
491 
492 	assert(ml_get_interrupts_enabled() == FALSE);
493 
494 #if DEVELOPMENT || DEBUG
495 	if (traptrace_index != TRAPTRACE_INVALID_INDEX) {
496 		traptrace_end(traptrace_index, mach_absolute_time());
497 	}
498 #endif
499 }
500 
501 static inline void
reset_dr7(void)502 reset_dr7(void)
503 {
504 	long dr7 = 0x400; /* magic dr7 reset value; 32 bit on i386, 64 bit on x86_64 */
505 	__asm__ volatile ("mov %0,%%dr7" : : "r" (dr7));
506 }
507 #if MACH_KDP
508 unsigned kdp_has_active_watchpoints = 0;
509 #define NO_WATCHPOINTS (!kdp_has_active_watchpoints)
510 #else
511 #define NO_WATCHPOINTS 1
512 #endif
513 
514 static uint32_t bound_chk_violations_event;
515 
516 static void
xnu_soft_trap_handle_breakpoint(__unused void * tstate,uint16_t comment)517 xnu_soft_trap_handle_breakpoint(
518 	__unused void     *tstate,
519 	uint16_t          comment)
520 {
521 	if (comment == CLANG_SOFT_TRAP_BOUND_CHK) {
522 		os_atomic_inc(&bound_chk_violations_event, relaxed);
523 	}
524 }
525 
526 KERNEL_BRK_DESCRIPTOR_DEFINE(clang_desc,
527     .type                = KERNEL_BRK_TYPE_CLANG,
528     .base                = CLANG_X86_TRAP_START,
529     .max                 = CLANG_X86_TRAP_END,
530     .options             = KERNEL_BRK_UNRECOVERABLE,
531     .handle_breakpoint   = NULL);
532 
533 KERNEL_BRK_DESCRIPTOR_DEFINE(xnu_soft_traps_desc,
534     .type                = KERNEL_BRK_TYPE_TELEMETRY,
535     .base                = XNU_SOFT_TRAP_START,
536     .max                 = XNU_SOFT_TRAP_END,
537     .options             = KERNEL_BRK_RECOVERABLE | KERNEL_BRK_CORE_ANALYTICS,
538     .handle_breakpoint   = xnu_soft_trap_handle_breakpoint);
539 
540 KERNEL_BRK_DESCRIPTOR_DEFINE(libcxx_desc,
541     .type                = KERNEL_BRK_TYPE_LIBCXX,
542     .base                = LIBCXX_TRAP_START,
543     .max                 = LIBCXX_TRAP_END,
544     .options             = KERNEL_BRK_UNRECOVERABLE,
545     .handle_breakpoint   = NULL);
546 
547 KERNEL_BRK_DESCRIPTOR_DEFINE(xnu_hard_traps_desc,
548     .type                = KERNEL_BRK_TYPE_XNU,
549     .base                = XNU_HARD_TRAP_START,
550     .max                 = XNU_HARD_TRAP_END,
551     .options             = KERNEL_BRK_UNRECOVERABLE,
552     .handle_breakpoint   = NULL);
553 
554 static bool
handle_kernel_breakpoint(x86_saved_state64_t * state)555 handle_kernel_breakpoint(x86_saved_state64_t *state)
556 {
557 	uint16_t comment;
558 	const struct kernel_brk_descriptor *desc;
559 	uint8_t inst_buf[8];
560 	uint32_t prefix16 = 0x80B90F67; /* Encoding prefix for ud1 <16-bit code>(%eax), %eax */
561 	uint32_t prefix8 = 0x40B90F67; /* Encoding prefix for ud1 <8-bit code>(%eax), %eax */
562 	bool found_prefix8 = false;
563 
564 	vm_size_t sz = ml_nofault_copy(state->isf.rip, (vm_offset_t)inst_buf, sizeof(inst_buf));
565 	if (sz != sizeof(inst_buf)) {
566 		return false;
567 	}
568 
569 	if (bcmp(inst_buf, &prefix16, sizeof(prefix16)) == 0) {
570 		/* The two bytes following the prefix is our code */
571 		comment = inst_buf[5] << 8 | inst_buf[4];
572 	} else if (bcmp(inst_buf, &prefix8, sizeof(prefix8)) == 0) {
573 		/* The one byte following the prefix is our code */
574 		found_prefix8 = true;
575 		comment = inst_buf[4];
576 	} else {
577 		return false;
578 	}
579 
580 	desc = find_brk_descriptor_by_comment(comment);
581 
582 	if (!desc) {
583 		return false;
584 	}
585 
586 	if (desc->options & KERNEL_BRK_TELEMETRY_OPTIONS) {
587 		telemetry_kernel_brk(desc->type, desc->options, (void *)state, comment);
588 	}
589 
590 	if (desc->handle_breakpoint) {
591 		desc->handle_breakpoint(state, comment); /* May trigger panic */
592 	}
593 
594 	/* Still alive? Check if we should recover. */
595 	if (desc->options & KERNEL_BRK_RECOVERABLE) {
596 		/* ud1 can be five or eight-byte long depending on the prefix */
597 		set_recovery_ip(state, state->isf.rip + (found_prefix8 ? 5 : 8));
598 		return true;
599 	}
600 
601 	return false;
602 }
603 
604 /*
605  * Trap from kernel mode.  Only page-fault errors are recoverable,
606  * and then only in special circumstances.  All other errors are
607  * fatal.  Return value indicates if trap was handled.
608  */
609 
610 void
kernel_trap(x86_saved_state_t * state,uintptr_t * lo_spp)611 kernel_trap(
612 	x86_saved_state_t       *state,
613 	uintptr_t *lo_spp)
614 {
615 	x86_saved_state64_t     *saved_state;
616 	int                     code;
617 	user_addr_t             vaddr;
618 	int                     type;
619 	vm_map_t                map = 0;        /* protected by T_PAGE_FAULT */
620 	kern_return_t           result = KERN_FAILURE;
621 	kern_return_t           fault_result = KERN_SUCCESS;
622 	thread_t                thread;
623 	boolean_t               intr;
624 	vm_prot_t               prot;
625 	struct recovery         *rp;
626 	vm_offset_t             kern_ip;
627 	int                     is_user;
628 	int                     trap_pl = get_preemption_level();
629 
630 	thread = current_thread();
631 
632 	if (__improbable(is_saved_state32(state))) {
633 		panic("kernel_trap(%p) with 32-bit state", state);
634 	}
635 	saved_state = saved_state64(state);
636 
637 	/* Record cpu where state was captured */
638 	saved_state->isf.cpu = cpu_number();
639 
640 	vaddr = (user_addr_t)saved_state->cr2;
641 	type  = saved_state->isf.trapno;
642 	code  = (int)(saved_state->isf.err & 0xffff);
643 	intr  = (saved_state->isf.rflags & EFL_IF) != 0;        /* state of ints at trap */
644 	kern_ip = (vm_offset_t)saved_state->isf.rip;
645 
646 	is_user = (vaddr < VM_MAX_USER_PAGE_ADDRESS);
647 
648 #if DEVELOPMENT || DEBUG
649 	uint32_t traptrace_index = traptrace_start(type, kern_ip, mach_absolute_time(), saved_state->rbp);
650 #endif
651 
652 #if CONFIG_DTRACE
653 	/*
654 	 * Is there a DTrace hook?
655 	 */
656 	if (__improbable(tempDTraceTrapHook != NULL)) {
657 		if (tempDTraceTrapHook(type, state, lo_spp, 0) == KERN_SUCCESS) {
658 			/*
659 			 * If it succeeds, we are done...
660 			 */
661 			goto common_return;
662 		}
663 	}
664 
665 	/* Handle traps originated from probe context. */
666 	if (thread != THREAD_NULL && thread->t_dtrace_inprobe) {
667 		if (dtrace_handle_trap(type, state)) {
668 			goto common_return;
669 		}
670 	}
671 
672 #endif /* CONFIG_DTRACE */
673 
674 	/*
675 	 * we come here with interrupts off as we don't want to recurse
676 	 * on preemption below.  but we do want to re-enable interrupts
677 	 * as soon we possibly can to hold latency down
678 	 */
679 	if (__improbable(T_PREEMPT == type)) {
680 		ast_taken_kernel();
681 
682 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
683 		    (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86, type)) | DBG_FUNC_NONE,
684 		    0, 0, 0, VM_KERNEL_UNSLIDE(kern_ip), 0);
685 
686 		goto common_return;
687 	}
688 
689 	user_addr_t     kd_vaddr = is_user ? vaddr : VM_KERNEL_UNSLIDE(vaddr);
690 	KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
691 	    (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86, type)) | DBG_FUNC_NONE,
692 	    (unsigned)(kd_vaddr >> 32), (unsigned)kd_vaddr, is_user,
693 	    VM_KERNEL_UNSLIDE(kern_ip), 0);
694 
695 
696 	if (T_PAGE_FAULT == type) {
697 		/*
698 		 * assume we're faulting in the kernel map
699 		 */
700 		map = kernel_map;
701 
702 		if (__probable((thread != THREAD_NULL) && (thread->map != kernel_map) &&
703 		    (vaddr < VM_MAX_USER_PAGE_ADDRESS))) {
704 			/* fault occurred in userspace */
705 			map = thread->map;
706 
707 			/* Intercept a potential Supervisor Mode Execute
708 			 * Protection fault. These criteria identify
709 			 * both NX faults and SMEP faults, but both
710 			 * are fatal. We avoid checking PTEs (racy).
711 			 * (The VM could just redrive a SMEP fault, hence
712 			 * the intercept).
713 			 */
714 			if (__improbable((code == (T_PF_PROT | T_PF_EXECUTE)) &&
715 			    (pmap_smep_enabled) && (saved_state->isf.rip == vaddr))) {
716 				goto debugger_entry;
717 			}
718 
719 			/*
720 			 * Additionally check for SMAP faults...
721 			 * which are characterized by page-present and
722 			 * the AC bit unset (i.e. not from copyin/out path).
723 			 */
724 			if (__improbable(code & T_PF_PROT &&
725 			    pmap_smap_enabled &&
726 			    (saved_state->isf.rflags & EFL_AC) == 0)) {
727 				goto debugger_entry;
728 			}
729 
730 			/*
731 			 * If we're not sharing cr3 with the user
732 			 * and we faulted in copyio,
733 			 * then switch cr3 here and dismiss the fault.
734 			 */
735 			if (no_shared_cr3 &&
736 			    (thread->machine.specFlags & CopyIOActive) &&
737 			    map->pmap->pm_cr3 != get_cr3_base()) {
738 				pmap_assert(current_cpu_datap()->cpu_pmap_pcid_enabled == FALSE);
739 				set_cr3_raw(map->pmap->pm_cr3);
740 				return;
741 			}
742 			if (__improbable(vaddr < PAGE_SIZE) &&
743 			    ((thread->machine.specFlags & CopyIOActive) == 0)) {
744 				goto debugger_entry;
745 			}
746 		}
747 	}
748 
749 	(void) ml_set_interrupts_enabled(intr);
750 
751 	switch (type) {
752 	case T_NO_FPU:
753 		fpnoextflt();
754 		goto common_return;
755 
756 	case T_FPU_FAULT:
757 		fpextovrflt();
758 		goto common_return;
759 
760 	case T_FLOATING_POINT_ERROR:
761 		fpexterrflt();
762 		goto common_return;
763 
764 	case T_SSE_FLOAT_ERROR:
765 		fpSSEexterrflt();
766 		goto common_return;
767 
768 	case T_INVALID_OPCODE:
769 		if (handle_kernel_breakpoint(saved_state)) {
770 			goto common_return;
771 		}
772 		fpUDflt(kern_ip);
773 		goto debugger_entry;
774 
775 	case T_DEBUG:
776 		/*
777 		 * Re-enable LBR tracing for core/panic files if necessary. i386_lbr_enable confirms LBR should be re-enabled.
778 		 */
779 		i386_lbr_enable();
780 		if ((saved_state->isf.rflags & EFL_TF) == 0 && NO_WATCHPOINTS) {
781 			/* We've somehow encountered a debug
782 			 * register match that does not belong
783 			 * to the kernel debugger.
784 			 * This isn't supposed to happen.
785 			 */
786 			reset_dr7();
787 			goto common_return;
788 		}
789 		goto debugger_entry;
790 	case T_INT3:
791 		goto debugger_entry;
792 	case T_PAGE_FAULT:
793 
794 #if CONFIG_DTRACE
795 		if (thread != THREAD_NULL && thread->t_dtrace_inprobe) { /* Executing under dtrace_probe? */
796 			if (dtrace_tally_fault(vaddr)) { /* Should a fault under dtrace be ignored? */
797 				/*
798 				 * DTrace has "anticipated" the possibility of this fault, and has
799 				 * established the suitable recovery state. Drop down now into the
800 				 * recovery handling code in "case T_GENERAL_PROTECTION:".
801 				 */
802 				goto FALL_THROUGH;
803 			}
804 		}
805 #endif /* CONFIG_DTRACE */
806 
807 		prot = VM_PROT_READ;
808 
809 		if (code & T_PF_WRITE) {
810 			prot |= VM_PROT_WRITE;
811 		}
812 		if (code & T_PF_EXECUTE) {
813 			prot |= VM_PROT_EXECUTE;
814 		}
815 
816 		fault_result = result = vm_fault(map,
817 		    vaddr,
818 		    prot,
819 		    FALSE, VM_KERN_MEMORY_NONE,
820 		    THREAD_UNINT, NULL, 0);
821 
822 		if (result == KERN_SUCCESS) {
823 			goto common_return;
824 		}
825 		/*
826 		 * fall through
827 		 */
828 #if CONFIG_DTRACE
829 FALL_THROUGH:
830 #endif /* CONFIG_DTRACE */
831 
832 	case T_GENERAL_PROTECTION:
833 		/*
834 		 * If there is a failure recovery address
835 		 * for this fault, go there.
836 		 */
837 		for (rp = recover_table; rp < recover_table_end; rp++) {
838 			if (kern_ip == rp->fault_addr) {
839 				set_recovery_ip(saved_state, rp->recover_addr);
840 				goto common_return;
841 			}
842 		}
843 
844 		/*
845 		 * Unanticipated page-fault errors in kernel
846 		 * should not happen.
847 		 *
848 		 * fall through...
849 		 */
850 		OS_FALLTHROUGH;
851 	default:
852 		/*
853 		 * Exception 15 is reserved but some chips may generate it
854 		 * spuriously. Seen at startup on AMD Athlon-64.
855 		 */
856 		if (type == 15) {
857 			kprintf("kernel_trap() ignoring spurious trap 15\n");
858 			goto common_return;
859 		}
860 debugger_entry:
861 		/* Ensure that the i386_kernel_state at the base of the
862 		 * current thread's stack (if any) is synchronized with the
863 		 * context at the moment of the trap, to facilitate
864 		 * access through the debugger.
865 		 */
866 		sync_iss_to_iks(state);
867 #if  MACH_KDP
868 		if (kdp_i386_trap(type, saved_state, result, (vm_offset_t)vaddr)) {
869 			goto common_return;
870 		}
871 #endif
872 	}
873 	if (type == T_PAGE_FAULT) {
874 		panic_fault_address = vaddr;
875 	}
876 	pal_cli();
877 	panic_trap(saved_state, trap_pl, fault_result);
878 	/*
879 	 * NO RETURN
880 	 */
881 
882 common_return:
883 #if DEVELOPMENT || DEBUG
884 	if (traptrace_index != TRAPTRACE_INVALID_INDEX) {
885 		traptrace_end(traptrace_index, mach_absolute_time());
886 	}
887 #endif
888 	return;
889 }
890 
891 static void
set_recovery_ip(x86_saved_state64_t * saved_state,vm_offset_t ip)892 set_recovery_ip(x86_saved_state64_t  *saved_state, vm_offset_t ip)
893 {
894 	saved_state->isf.rip = ip;
895 }
896 
897 static void
panic_trap(x86_saved_state64_t * regs,uint32_t pl,kern_return_t fault_result)898 panic_trap(x86_saved_state64_t *regs, uint32_t pl, kern_return_t fault_result)
899 {
900 	const char      *trapname = "Unknown";
901 	pal_cr_t        cr0, cr2, cr3, cr4;
902 	boolean_t       potential_smep_fault = FALSE, potential_kernel_NX_fault = FALSE;
903 	boolean_t       potential_smap_fault = FALSE;
904 
905 	pal_get_control_registers( &cr0, &cr2, &cr3, &cr4 );
906 	assert(ml_get_interrupts_enabled() == FALSE);
907 	current_cpu_datap()->cpu_fatal_trap_state = regs;
908 	/*
909 	 * Issue an I/O port read if one has been requested - this is an
910 	 * event logic analyzers can use as a trigger point.
911 	 */
912 	panic_notify();
913 
914 	kprintf("CPU %d panic trap number 0x%x, rip 0x%016llx\n",
915 	    cpu_number(), regs->isf.trapno, regs->isf.rip);
916 	kprintf("cr0 0x%016llx cr2 0x%016llx cr3 0x%016llx cr4 0x%016llx\n",
917 	    cr0, cr2, cr3, cr4);
918 
919 	if (regs->isf.trapno < TRAP_TYPES) {
920 		trapname = trap_type[regs->isf.trapno];
921 	}
922 
923 	if ((regs->isf.trapno == T_PAGE_FAULT) && (regs->isf.err == (T_PF_PROT | T_PF_EXECUTE)) && (regs->isf.rip == regs->cr2)) {
924 		if (pmap_smep_enabled && (regs->isf.rip < VM_MAX_USER_PAGE_ADDRESS)) {
925 			potential_smep_fault = TRUE;
926 		} else if (regs->isf.rip >= VM_MIN_KERNEL_AND_KEXT_ADDRESS) {
927 			potential_kernel_NX_fault = TRUE;
928 		}
929 	} else if (pmap_smap_enabled &&
930 	    regs->isf.trapno == T_PAGE_FAULT &&
931 	    regs->isf.err & T_PF_PROT &&
932 	    regs->cr2 < VM_MAX_USER_PAGE_ADDRESS &&
933 	    regs->isf.rip >= VM_MIN_KERNEL_AND_KEXT_ADDRESS) {
934 		potential_smap_fault = TRUE;
935 	}
936 
937 #undef panic
938 	panic("Kernel trap at 0x%016llx, type %d=%s, registers:\n"
939 	    "CR0: 0x%016llx, CR2: 0x%016llx, CR3: 0x%016llx, CR4: 0x%016llx\n"
940 	    "RAX: 0x%016llx, RBX: 0x%016llx, RCX: 0x%016llx, RDX: 0x%016llx\n"
941 	    "RSP: 0x%016llx, RBP: 0x%016llx, RSI: 0x%016llx, RDI: 0x%016llx\n"
942 	    "R8:  0x%016llx, R9:  0x%016llx, R10: 0x%016llx, R11: 0x%016llx\n"
943 	    "R12: 0x%016llx, R13: 0x%016llx, R14: 0x%016llx, R15: 0x%016llx\n"
944 	    "RFL: 0x%016llx, RIP: 0x%016llx, CS:  0x%016llx, SS:  0x%016llx\n"
945 	    "Fault CR2: 0x%016llx, Error code: 0x%016llx, Fault CPU: 0x%x%s%s%s%s, PL: %d, VF: %d\n",
946 	    regs->isf.rip, regs->isf.trapno, trapname,
947 	    cr0, cr2, cr3, cr4,
948 	    regs->rax, regs->rbx, regs->rcx, regs->rdx,
949 	    regs->isf.rsp, regs->rbp, regs->rsi, regs->rdi,
950 	    regs->r8, regs->r9, regs->r10, regs->r11,
951 	    regs->r12, regs->r13, regs->r14, regs->r15,
952 	    regs->isf.rflags, regs->isf.rip, regs->isf.cs & 0xFFFF,
953 	    regs->isf.ss & 0xFFFF, regs->cr2, regs->isf.err, regs->isf.cpu,
954 	    virtualized ? " VMM" : "",
955 	    potential_kernel_NX_fault ? " Kernel NX fault" : "",
956 	    potential_smep_fault ? " SMEP/User NX fault" : "",
957 	    potential_smap_fault ? " SMAP fault" : "",
958 	    pl,
959 	    fault_result);
960 }
961 
962 #if CONFIG_DTRACE
963 extern kern_return_t dtrace_user_probe(x86_saved_state_t *);
964 #endif
965 
966 #if DEBUG
967 uint32_t fsigs[2];
968 uint32_t fsigns, fsigcs;
969 #endif
970 
971 /*
972  *	Trap from user mode.
973  */
974 void
user_trap(x86_saved_state_t * saved_state)975 user_trap(
976 	x86_saved_state_t *saved_state)
977 {
978 	int                     exc;
979 	int                     err;
980 	mach_exception_code_t   code;
981 	mach_exception_subcode_t subcode;
982 	int                     type;
983 	user_addr_t             vaddr;
984 	vm_prot_t               prot;
985 	thread_t                thread = current_thread();
986 	kern_return_t           kret;
987 	user_addr_t             rip;
988 	unsigned long           dr6 = 0; /* 32 bit for i386, 64 bit for x86_64 */
989 	int                     current_cpu = cpu_number();
990 #if DEVELOPMENT || DEBUG
991 	bool                    inspect_cacheline = false;
992 	uint32_t                traptrace_index;
993 #endif
994 	assert((is_saved_state32(saved_state) && !thread_is_64bit_addr(thread)) ||
995 	    (is_saved_state64(saved_state) && thread_is_64bit_addr(thread)));
996 
997 	if (is_saved_state64(saved_state)) {
998 		x86_saved_state64_t     *regs;
999 
1000 		regs = saved_state64(saved_state);
1001 
1002 		/* Record cpu where state was captured */
1003 		regs->isf.cpu = current_cpu;
1004 
1005 		type = regs->isf.trapno;
1006 		err  = (int)regs->isf.err & 0xffff;
1007 		vaddr = (user_addr_t)regs->cr2;
1008 		rip   = (user_addr_t)regs->isf.rip;
1009 #if DEVELOPMENT || DEBUG
1010 		traptrace_index = traptrace_start(type, rip, mach_absolute_time(), regs->rbp);
1011 #endif
1012 	} else {
1013 		x86_saved_state32_t     *regs;
1014 
1015 		regs = saved_state32(saved_state);
1016 
1017 		/* Record cpu where state was captured */
1018 		regs->cpu = current_cpu;
1019 
1020 		type  = regs->trapno;
1021 		err   = regs->err & 0xffff;
1022 		vaddr = (user_addr_t)regs->cr2;
1023 		rip   = (user_addr_t)regs->eip;
1024 #if DEVELOPMENT || DEBUG
1025 		traptrace_index = traptrace_start(type, rip, mach_absolute_time(), regs->ebp);
1026 #endif
1027 	}
1028 
1029 #if DEVELOPMENT || DEBUG
1030 	/*
1031 	 * Copy the cacheline of code into the thread's instruction stream save area
1032 	 * before enabling interrupts (the assumption is that we have not otherwise faulted or
1033 	 * trapped since the original cache line stores).  If the saved code is not valid,
1034 	 * we'll catch it below when we process the copyin() for unhandled faults.
1035 	 */
1036 	if (thread->machine.insn_copy_optout == false &&
1037 	    (type == T_PAGE_FAULT || type == T_INVALID_OPCODE || type == T_GENERAL_PROTECTION)) {
1038 #define CACHELINE_SIZE 64
1039 		THREAD_TO_PCB(thread)->insn_cacheline[CACHELINE_SIZE] = (uint8_t)(rip & (CACHELINE_SIZE - 1));
1040 		bcopy(&cpu_shadowp(current_cpu)->cpu_rtimes[0],
1041 		    &THREAD_TO_PCB(thread)->insn_cacheline[0],
1042 		    sizeof(THREAD_TO_PCB(thread)->insn_cacheline) - 1);
1043 		inspect_cacheline = true;
1044 	}
1045 #endif
1046 
1047 	if (type == T_DEBUG) {
1048 		if (thread->machine.ids) {
1049 			unsigned long clear = 0;
1050 			/* Stash and clear this processor's DR6 value, in the event
1051 			 * this was a debug register match
1052 			 */
1053 			__asm__ volatile ("mov %%db6, %0" : "=r" (dr6));
1054 			__asm__ volatile ("mov %0, %%db6" : : "r" (clear));
1055 		}
1056 		/* [Re]Enable LBRs *BEFORE* enabling interrupts to ensure we hit the right CPU */
1057 		i386_lbr_enable();
1058 	}
1059 
1060 	if (type == T_PAGE_FAULT) {
1061 		thread_reset_pcs_will_fault(thread);
1062 	}
1063 
1064 	pal_sti();
1065 
1066 	KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
1067 	    (MACHDBG_CODE(DBG_MACH_EXCP_UTRAP_x86, type)) | DBG_FUNC_NONE,
1068 	    (unsigned)(vaddr >> 32), (unsigned)vaddr,
1069 	    (unsigned)(rip >> 32), (unsigned)rip, 0);
1070 
1071 	code = 0;
1072 	subcode = 0;
1073 	exc = 0;
1074 
1075 #if CONFIG_DTRACE
1076 	/*
1077 	 * DTrace does not consume all user traps, only INT_3's for now.
1078 	 * Avoid needlessly calling tempDTraceTrapHook here, and let the
1079 	 * INT_3 case handle them.
1080 	 */
1081 #endif
1082 
1083 	DEBUG_KPRINT_SYSCALL_MASK(1,
1084 	    "user_trap: type=0x%x(%s) err=0x%x cr2=%p rip=%p\n",
1085 	    type, trap_type[type], err, (void *)(long) vaddr, (void *)(long) rip);
1086 
1087 	switch (type) {
1088 	case T_DIVIDE_ERROR:
1089 		exc = EXC_ARITHMETIC;
1090 		code = EXC_I386_DIV;
1091 		break;
1092 
1093 	case T_DEBUG:
1094 	{
1095 		pcb_t   pcb;
1096 		/*
1097 		 * Update the PCB with this processor's DR6 value
1098 		 * in the event this was a debug register match.
1099 		 */
1100 		pcb = THREAD_TO_PCB(thread);
1101 		if (pcb->ids) {
1102 			/*
1103 			 * We can get and set the status register
1104 			 * in 32-bit mode even on a 64-bit thread
1105 			 * because the high order bits are not
1106 			 * used on x86_64
1107 			 */
1108 			if (thread_is_64bit_addr(thread)) {
1109 				x86_debug_state64_t *ids = pcb->ids;
1110 				ids->dr6 = dr6;
1111 			} else {         /* 32 bit thread */
1112 				x86_debug_state32_t *ids = pcb->ids;
1113 				ids->dr6 = (uint32_t) dr6;
1114 			}
1115 		}
1116 		exc = EXC_BREAKPOINT;
1117 		code = EXC_I386_SGL;
1118 		break;
1119 	}
1120 	case T_INT3:
1121 #if CONFIG_DTRACE
1122 		if (dtrace_user_probe(saved_state) == KERN_SUCCESS) {
1123 			return; /* If it succeeds, we are done... */
1124 		}
1125 #endif
1126 		exc = EXC_BREAKPOINT;
1127 		code = EXC_I386_BPT;
1128 		break;
1129 
1130 	case T_OVERFLOW:
1131 		exc = EXC_ARITHMETIC;
1132 		code = EXC_I386_INTO;
1133 		break;
1134 
1135 	case T_OUT_OF_BOUNDS:
1136 		exc = EXC_SOFTWARE;
1137 		code = EXC_I386_BOUND;
1138 		break;
1139 
1140 	case T_INVALID_OPCODE:
1141 		if (fpUDflt(rip) == 1) {
1142 			exc = EXC_BAD_INSTRUCTION;
1143 			code = EXC_I386_INVOP;
1144 		}
1145 		break;
1146 
1147 	case T_NO_FPU:
1148 		fpnoextflt();
1149 		break;
1150 
1151 	case T_FPU_FAULT:
1152 		fpextovrflt();
1153 		/*
1154 		 * Raise exception.
1155 		 */
1156 		exc = EXC_BAD_ACCESS;
1157 		code = VM_PROT_READ | VM_PROT_EXECUTE;
1158 		subcode = 0;
1159 		break;
1160 
1161 	case T_INVALID_TSS:     /* invalid TSS == iret with NT flag set */
1162 		exc = EXC_BAD_INSTRUCTION;
1163 		code = EXC_I386_INVTSSFLT;
1164 		subcode = err;
1165 		break;
1166 
1167 	case T_SEGMENT_NOT_PRESENT:
1168 		exc = EXC_BAD_INSTRUCTION;
1169 		code = EXC_I386_SEGNPFLT;
1170 		subcode = err;
1171 		break;
1172 
1173 	case T_STACK_FAULT:
1174 		exc = EXC_BAD_INSTRUCTION;
1175 		code = EXC_I386_STKFLT;
1176 		subcode = err;
1177 		break;
1178 
1179 	case T_GENERAL_PROTECTION:
1180 		/*
1181 		 * There's a wide range of circumstances which generate this
1182 		 * class of exception. From user-space, many involve bad
1183 		 * addresses (such as a non-canonical 64-bit address).
1184 		 * So we map this to EXC_BAD_ACCESS (and thereby SIGSEGV).
1185 		 * The trouble is cr2 doesn't contain the faulting address;
1186 		 * we'd need to decode the faulting instruction to really
1187 		 * determine this. We'll leave that to debuggers.
1188 		 * However, attempted execution of privileged instructions
1189 		 * (e.g. cli) also generate GP faults and so we map these to
1190 		 * to EXC_BAD_ACCESS (and thence SIGSEGV) also - rather than
1191 		 * EXC_BAD_INSTRUCTION which is more accurate. We just can't
1192 		 * win!
1193 		 */
1194 		exc = EXC_BAD_ACCESS;
1195 		code = EXC_I386_GPFLT;
1196 		subcode = err;
1197 		break;
1198 
1199 	case T_PAGE_FAULT:
1200 	{
1201 		prot = VM_PROT_READ;
1202 
1203 		if (err & T_PF_WRITE) {
1204 			prot |= VM_PROT_WRITE;
1205 		}
1206 		if (__improbable(err & T_PF_EXECUTE)) {
1207 			prot |= VM_PROT_EXECUTE;
1208 		}
1209 #if DEVELOPMENT || DEBUG
1210 		bool do_simd_hash = thread_fpsimd_hash_enabled();
1211 		uint32_t fsig = 0;
1212 		fsig = do_simd_hash ? thread_fpsimd_hash(thread) : 0;
1213 #if DEBUG
1214 		fsigs[0] = fsig;
1215 #endif
1216 #endif
1217 		kret = vm_fault(thread->map,
1218 		    vaddr,
1219 		    prot, FALSE, VM_KERN_MEMORY_NONE,
1220 		    THREAD_ABORTSAFE, NULL, 0);
1221 #if DEVELOPMENT || DEBUG
1222 		if (do_simd_hash && fsig) {
1223 			uint32_t fsig2 = thread_fpsimd_hash(thread);
1224 #if DEBUG
1225 			fsigcs++;
1226 			fsigs[1] = fsig2;
1227 #endif
1228 			if (fsig != fsig2) {
1229 				panic("FP/SIMD state hash mismatch across fault thread: %p 0x%x->0x%x", thread, fsig, fsig2);
1230 			}
1231 		} else {
1232 #if DEBUG
1233 			fsigns++;
1234 #endif
1235 		}
1236 #endif
1237 		if (__probable((kret == KERN_SUCCESS) || (kret == KERN_ABORTED))) {
1238 			break;
1239 		} else if (__improbable(kret == KERN_FAILURE)) {
1240 			/*
1241 			 * For a user trap, vm_fault() should never return KERN_FAILURE.
1242 			 * If it does, we're leaking preemption disables somewhere in the kernel.
1243 			 */
1244 			panic("vm_fault() KERN_FAILURE from user fault on thread %p", thread);
1245 		}
1246 
1247 		/* PAL debug hook (empty on x86) */
1248 		pal_dbg_page_fault(thread, vaddr, kret);
1249 		exc = EXC_BAD_ACCESS;
1250 		code = kret;
1251 		subcode = vaddr;
1252 	}
1253 	break;
1254 
1255 	case T_SSE_FLOAT_ERROR:
1256 		fpSSEexterrflt();
1257 		exc = EXC_ARITHMETIC;
1258 		code = EXC_I386_SSEEXTERR;
1259 		subcode = ((struct x86_fx_thread_state *)thread->machine.ifps)->fx_MXCSR;
1260 		break;
1261 
1262 
1263 	case T_FLOATING_POINT_ERROR:
1264 		fpexterrflt();
1265 		exc = EXC_ARITHMETIC;
1266 		code = EXC_I386_EXTERR;
1267 		subcode = ((struct x86_fx_thread_state *)thread->machine.ifps)->fx_status;
1268 		break;
1269 
1270 	case T_DTRACE_RET:
1271 #if CONFIG_DTRACE
1272 		if (dtrace_user_probe(saved_state) == KERN_SUCCESS) {
1273 			return; /* If it succeeds, we are done... */
1274 		}
1275 #endif
1276 		/*
1277 		 * If we get an INT 0x7f when we do not expect to,
1278 		 * treat it as an illegal instruction
1279 		 */
1280 		exc = EXC_BAD_INSTRUCTION;
1281 		code = EXC_I386_INVOP;
1282 		break;
1283 
1284 	default:
1285 		panic("Unexpected user trap, type %d", type);
1286 	}
1287 
1288 	if (type == T_PAGE_FAULT) {
1289 		thread_reset_pcs_done_faulting(thread);
1290 	}
1291 
1292 	if (exc != 0) {
1293 		uint16_t cs;
1294 		boolean_t intrs;
1295 
1296 		if (is_saved_state64(saved_state)) {
1297 			cs = saved_state64(saved_state)->isf.cs;
1298 		} else {
1299 			cs = saved_state32(saved_state)->cs;
1300 		}
1301 
1302 		if (last_branch_enabled_modes == LBR_ENABLED_USERMODE) {
1303 			intrs = ml_set_interrupts_enabled(FALSE);
1304 			/*
1305 			 * This is a bit racy (it's possible for this thread to migrate to another CPU, then
1306 			 * migrate back, but that seems rather rare in practice), but good enough to ensure
1307 			 * the LBRs are saved before proceeding with exception/signal dispatch.
1308 			 */
1309 			if (current_cpu == cpu_number()) {
1310 				i386_lbr_synch(thread);
1311 			}
1312 			ml_set_interrupts_enabled(intrs);
1313 		}
1314 
1315 		/*
1316 		 * Do not try to copyin from the instruction stream if the page fault was due
1317 		 * to an access to rip and was unhandled.
1318 		 * Do not deal with cases when %cs != USER[64]_CS
1319 		 * And of course there's no need to copy the instruction stream if the boot-arg
1320 		 * was set to 0.
1321 		 */
1322 		if (thread->machine.insn_copy_optout == false && insn_copyin_count > 0 &&
1323 		    (cs == USER64_CS || cs == USER_CS) && (type != T_PAGE_FAULT || vaddr != rip)) {
1324 #if DEVELOPMENT || DEBUG
1325 			copy_instruction_stream(thread, rip, type, inspect_cacheline);
1326 #else
1327 			copy_instruction_stream(thread, rip, type);
1328 #endif
1329 		}
1330 
1331 #if DEVELOPMENT || DEBUG
1332 		if (traptrace_index != TRAPTRACE_INVALID_INDEX) {
1333 			traptrace_end(traptrace_index, mach_absolute_time());
1334 		}
1335 #endif
1336 		/*
1337 		 * Note: Codepaths that directly return from user_trap() have pending
1338 		 * ASTs processed in locore
1339 		 */
1340 		i386_exception(exc, code, subcode);
1341 		/* NOTREACHED */
1342 	} else {
1343 #if DEVELOPMENT || DEBUG
1344 		if (traptrace_index != TRAPTRACE_INVALID_INDEX) {
1345 			traptrace_end(traptrace_index, mach_absolute_time());
1346 		}
1347 #endif
1348 	}
1349 }
1350 
1351 /*
1352  * Copyin up to x86_INSTRUCTION_STATE_MAX_INSN_BYTES bytes from the page that includes `rip`,
1353  * ensuring that we stay on the same page, clipping the start or end, as needed.
1354  * Add the clipped amount back at the start or end, depending on where it fits.
1355  * Consult the variable populated by the boot-arg `insn_capcnt'
1356  */
1357 static __attribute__((noinline)) void
copy_instruction_stream(thread_t thread,uint64_t rip,int __unused trap_code,bool inspect_cacheline)1358 copy_instruction_stream(thread_t thread, uint64_t rip, int __unused trap_code
1359 #if DEVELOPMENT || DEBUG
1360     , bool inspect_cacheline
1361 #endif
1362     )
1363 {
1364 #if x86_INSTRUCTION_STATE_MAX_INSN_BYTES > 4096
1365 #error x86_INSTRUCTION_STATE_MAX_INSN_BYTES cannot exceed a page in size.
1366 #endif
1367 	pcb_t pcb = THREAD_TO_PCB(thread);
1368 	vm_map_offset_t pagemask = ~vm_map_page_mask(current_map());
1369 	vm_map_offset_t rip_page = rip & pagemask;
1370 	vm_map_offset_t start_addr;
1371 	vm_map_offset_t insn_offset;
1372 	vm_map_offset_t end_addr = rip + (insn_copyin_count / 2);
1373 	void *stack_buffer;
1374 	int copyin_err = 0;
1375 #if defined(MACH_BSD) && (DEVELOPMENT || DEBUG)
1376 	void *procname;
1377 #endif
1378 
1379 #if DEVELOPMENT || DEBUG
1380 	assert(insn_copyin_count <= x86_INSTRUCTION_STATE_MAX_INSN_BYTES);
1381 #else
1382 	if (insn_copyin_count > x86_INSTRUCTION_STATE_MAX_INSN_BYTES ||
1383 	    insn_copyin_count < 64 /* CACHELINE_SIZE */) {
1384 		return;
1385 	}
1386 #endif
1387 
1388 #pragma clang diagnostic push
1389 #pragma clang diagnostic ignored "-Walloca"
1390 	stack_buffer = __builtin_alloca(insn_copyin_count);
1391 #pragma clang diagnostic pop
1392 
1393 	if (rip >= (insn_copyin_count / 2)) {
1394 		start_addr = rip - (insn_copyin_count / 2);
1395 	} else {
1396 		start_addr = 0;
1397 	}
1398 
1399 	if (start_addr < rip_page) {
1400 		insn_offset = (insn_copyin_count / 2) - (rip_page - start_addr);
1401 		end_addr += (rip_page - start_addr);
1402 		start_addr = rip_page;
1403 	} else if (end_addr >= (rip_page + (~pagemask + 1))) {
1404 		start_addr -= (end_addr - (rip_page + (~pagemask + 1))); /* Adjust start address backward */
1405 		/* Adjust instruction offset due to start address change */
1406 		insn_offset = (insn_copyin_count / 2) + (end_addr - (rip_page + (~pagemask + 1)));
1407 		end_addr = rip_page + (~pagemask + 1);  /* clip to the start of the next page (non-inclusive */
1408 	} else {
1409 		insn_offset = insn_copyin_count / 2;
1410 	}
1411 
1412 	disable_preemption();   /* Prevent copyin from faulting in the instruction stream */
1413 	if (
1414 #if DEVELOPMENT || DEBUG
1415 		(insnstream_force_cacheline_mismatch < 2) &&
1416 #endif
1417 		((end_addr > start_addr) && (copyin_err = copyin(start_addr, stack_buffer, end_addr - start_addr)) == 0)) {
1418 		enable_preemption();
1419 
1420 		if (pcb->insn_state == 0) {
1421 			pcb->insn_state = kalloc_data(sizeof(x86_instruction_state_t), Z_WAITOK);
1422 		}
1423 
1424 		if (pcb->insn_state != 0) {
1425 			bcopy(stack_buffer, pcb->insn_state->insn_bytes, end_addr - start_addr);
1426 			bzero(&pcb->insn_state->insn_bytes[end_addr - start_addr],
1427 			    insn_copyin_count - (end_addr - start_addr));
1428 
1429 			pcb->insn_state->insn_stream_valid_bytes = (int)(end_addr - start_addr);
1430 			pcb->insn_state->insn_offset = (int)insn_offset;
1431 
1432 #if DEVELOPMENT || DEBUG
1433 			/* Now try to validate the cacheline we read at early-fault time matches the code
1434 			 * copied in. Before we do that, we have to make sure the buffer contains a valid
1435 			 * cacheline by looking for the 2 sentinel values written in the event the cacheline
1436 			 * could not be copied.
1437 			 */
1438 #define CACHELINE_DATA_NOT_PRESENT 0xdeadc0debeefcafeULL
1439 #define CACHELINE_MASK (CACHELINE_SIZE - 1)
1440 
1441 			if (inspect_cacheline &&
1442 			    (*(uint64_t *)(uintptr_t)&pcb->insn_cacheline[0] != CACHELINE_DATA_NOT_PRESENT &&
1443 			    *(uint64_t *)(uintptr_t)&pcb->insn_cacheline[8] != CACHELINE_DATA_NOT_PRESENT)) {
1444 				/*
1445 				 * The position of the cacheline in the instruction buffer is at offset
1446 				 * insn_offset - (rip & CACHELINE_MASK)
1447 				 */
1448 				if (__improbable((rip & CACHELINE_MASK) > insn_offset)) {
1449 					printf("thread %p code cacheline @ %p clipped wrt copied-in code (offset %d)\n",
1450 					    thread, (void *)(rip & ~CACHELINE_MASK), (int)(rip & CACHELINE_MASK));
1451 				} else if (bcmp(&pcb->insn_state->insn_bytes[insn_offset - (rip & CACHELINE_MASK)],
1452 				    &pcb->insn_cacheline[0], CACHELINE_SIZE) != 0
1453 				    || insnstream_force_cacheline_mismatch
1454 				    ) {
1455 #if x86_INSTRUCTION_STATE_CACHELINE_SIZE != CACHELINE_SIZE
1456 #error cacheline size mismatch
1457 #endif
1458 					bcopy(&pcb->insn_cacheline[0], &pcb->insn_state->insn_cacheline[0],
1459 					    x86_INSTRUCTION_STATE_CACHELINE_SIZE);
1460 					/* Mark the instruction stream as being out-of-synch */
1461 					pcb->insn_state->out_of_synch = 1;
1462 
1463 					printf("thread %p code cacheline @ %p mismatches with copied-in code [trap 0x%x]\n",
1464 					    thread, (void *)(rip & ~CACHELINE_MASK), trap_code);
1465 					for (int i = 0; i < 8; i++) {
1466 						printf("\t[%d] cl=0x%08llx vs. ci=0x%08llx\n", i, *(uint64_t *)(uintptr_t)&pcb->insn_cacheline[i * 8],
1467 						    *(uint64_t *)(uintptr_t)&pcb->insn_state->insn_bytes[(i * 8) + insn_offset - (rip & CACHELINE_MASK)]);
1468 					}
1469 					if (panic_on_cacheline_mismatch) {
1470 						panic("Cacheline mismatch while processing unhandled exception.");
1471 					}
1472 				} else {
1473 					pcb->insn_state->out_of_synch = 0;
1474 				}
1475 			} else if (inspect_cacheline) {
1476 				printf("thread %p could not capture code cacheline at fault IP %p [offset %d]\n",
1477 				    (void *)thread, (void *)rip, (int)(insn_offset - (rip & CACHELINE_MASK)));
1478 				pcb->insn_state->out_of_synch = 0;
1479 			}
1480 #else
1481 			pcb->insn_state->out_of_synch = 0;
1482 #endif /* DEVELOPMENT || DEBUG */
1483 
1484 #if defined(MACH_BSD) && (DEVELOPMENT || DEBUG)
1485 			if (panic_on_trap_procname[0] != 0) {
1486 				task_t task = get_threadtask(thread);
1487 				char procnamebuf[65] = {0};
1488 
1489 				if (get_bsdtask_info(task) != NULL) {
1490 					procname = proc_name_address(get_bsdtask_info(task));
1491 					strlcpy(procnamebuf, procname, sizeof(procnamebuf));
1492 
1493 					if (strcasecmp(panic_on_trap_procname, procnamebuf) == 0 &&
1494 					    ((1U << trap_code) & panic_on_trap_mask) != 0) {
1495 						panic("Panic requested on trap type 0x%x for process `%s'", trap_code,
1496 						    panic_on_trap_procname);
1497 						/*NORETURN*/
1498 					}
1499 				}
1500 			}
1501 #endif /* MACH_BSD && (DEVELOPMENT || DEBUG) */
1502 		}
1503 	} else {
1504 		enable_preemption();
1505 
1506 		pcb->insn_state_copyin_failure_errorcode = copyin_err;
1507 #if DEVELOPMENT || DEBUG
1508 		if (inspect_cacheline && pcb->insn_state == 0) {
1509 			pcb->insn_state = kalloc_data(sizeof(x86_instruction_state_t), Z_WAITOK);
1510 		}
1511 		if (pcb->insn_state != 0) {
1512 			pcb->insn_state->insn_stream_valid_bytes = 0;
1513 			pcb->insn_state->insn_offset = 0;
1514 
1515 			if (inspect_cacheline &&
1516 			    (*(uint64_t *)(uintptr_t)&pcb->insn_cacheline[0] != CACHELINE_DATA_NOT_PRESENT &&
1517 			    *(uint64_t *)(uintptr_t)&pcb->insn_cacheline[8] != CACHELINE_DATA_NOT_PRESENT)) {
1518 				/*
1519 				 * We can still copy the cacheline into the instruction state structure
1520 				 * if it contains valid data
1521 				 */
1522 				pcb->insn_state->out_of_synch = 1;
1523 				bcopy(&pcb->insn_cacheline[0], &pcb->insn_state->insn_cacheline[0],
1524 				    x86_INSTRUCTION_STATE_CACHELINE_SIZE);
1525 			}
1526 		}
1527 #endif /* DEVELOPMENT || DEBUG */
1528 	}
1529 }
1530 
1531 /*
1532  * Handle exceptions for i386.
1533  *
1534  * If we are an AT bus machine, we must turn off the AST for a
1535  * delayed floating-point exception.
1536  *
1537  * If we are providing floating-point emulation, we may have
1538  * to retrieve the real register values from the floating point
1539  * emulator.
1540  */
1541 void
i386_exception(int exc,mach_exception_code_t code,mach_exception_subcode_t subcode)1542 i386_exception(
1543 	int     exc,
1544 	mach_exception_code_t code,
1545 	mach_exception_subcode_t subcode)
1546 {
1547 	mach_exception_data_type_t   codes[EXCEPTION_CODE_MAX];
1548 
1549 	DEBUG_KPRINT_SYSCALL_MACH("i386_exception: exc=%d code=0x%llx subcode=0x%llx\n",
1550 	    exc, code, subcode);
1551 	codes[0] = code;                /* new exception interface */
1552 	codes[1] = subcode;
1553 	exception_triage(exc, codes, 2);
1554 	/*NOTREACHED*/
1555 }
1556 
1557 
1558 /* Synchronize a thread's x86_kernel_state (if any) with the given
1559  * x86_saved_state_t obtained from the trap/IPI handler; called in
1560  * kernel_trap() prior to entering the debugger, and when receiving
1561  * an "MP_KDP" IPI. Called with null saved_state if an incoming IPI
1562  * was detected from the kernel while spinning with interrupts masked.
1563  */
1564 
1565 void
sync_iss_to_iks(x86_saved_state_t * saved_state)1566 sync_iss_to_iks(x86_saved_state_t *saved_state)
1567 {
1568 	struct x86_kernel_state *iks = NULL;
1569 	vm_offset_t kstack;
1570 	boolean_t record_active_regs = FALSE;
1571 
1572 	/* The PAL may have a special way to sync registers */
1573 	if (saved_state && saved_state->flavor == THREAD_STATE_NONE) {
1574 		pal_get_kern_regs( saved_state );
1575 	}
1576 
1577 	if (current_thread() != NULL &&
1578 	    (kstack = current_thread()->kernel_stack) != 0) {
1579 		x86_saved_state64_t     *regs = saved_state64(saved_state);
1580 
1581 		iks = STACK_IKS(kstack);
1582 
1583 		/* Did we take the trap/interrupt in kernel mode? */
1584 		if (saved_state == NULL || /* NULL => polling in kernel */
1585 		    regs == USER_REGS64(current_thread())) {
1586 			record_active_regs = TRUE;
1587 		} else {
1588 			iks->k_rbx = regs->rbx;
1589 			iks->k_rsp = regs->isf.rsp;
1590 			iks->k_rbp = regs->rbp;
1591 			iks->k_r12 = regs->r12;
1592 			iks->k_r13 = regs->r13;
1593 			iks->k_r14 = regs->r14;
1594 			iks->k_r15 = regs->r15;
1595 			iks->k_rip = regs->isf.rip;
1596 		}
1597 	}
1598 
1599 	if (record_active_regs == TRUE) {
1600 		/* Show the trap handler path */
1601 		__asm__ volatile ("movq %%rbx, %0" : "=m" (iks->k_rbx));
1602 		__asm__ volatile ("movq %%rsp, %0" : "=m" (iks->k_rsp));
1603 		__asm__ volatile ("movq %%rbp, %0" : "=m" (iks->k_rbp));
1604 		__asm__ volatile ("movq %%r12, %0" : "=m" (iks->k_r12));
1605 		__asm__ volatile ("movq %%r13, %0" : "=m" (iks->k_r13));
1606 		__asm__ volatile ("movq %%r14, %0" : "=m" (iks->k_r14));
1607 		__asm__ volatile ("movq %%r15, %0" : "=m" (iks->k_r15));
1608 		/* "Current" instruction pointer */
1609 		__asm__ volatile ("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
1610                                   : "=m" (iks->k_rip)
1611                                   :
1612                                   : "rax");
1613 	}
1614 }
1615 
1616 /*
1617  * This is used by the NMI interrupt handler (from mp.c) to
1618  * uncondtionally sync the trap handler context to the IKS
1619  * irrespective of whether the NMI was fielded in kernel
1620  * or user space.
1621  */
1622 void
sync_iss_to_iks_unconditionally(__unused x86_saved_state_t * saved_state)1623 sync_iss_to_iks_unconditionally(__unused x86_saved_state_t *saved_state)
1624 {
1625 	struct x86_kernel_state *iks;
1626 	vm_offset_t kstack;
1627 
1628 	if ((kstack = current_thread()->kernel_stack) != 0) {
1629 		iks = STACK_IKS(kstack);
1630 		/* Display the trap handler path */
1631 		__asm__ volatile ("movq %%rbx, %0" : "=m" (iks->k_rbx));
1632 		__asm__ volatile ("movq %%rsp, %0" : "=m" (iks->k_rsp));
1633 		__asm__ volatile ("movq %%rbp, %0" : "=m" (iks->k_rbp));
1634 		__asm__ volatile ("movq %%r12, %0" : "=m" (iks->k_r12));
1635 		__asm__ volatile ("movq %%r13, %0" : "=m" (iks->k_r13));
1636 		__asm__ volatile ("movq %%r14, %0" : "=m" (iks->k_r14));
1637 		__asm__ volatile ("movq %%r15, %0" : "=m" (iks->k_r15));
1638 		/* "Current" instruction pointer */
1639 		__asm__ volatile ("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:" : "=m" (iks->k_rip)::"rax");
1640 	}
1641 }
1642 
1643 #if DEBUG
1644 #define TERI 1
1645 #endif
1646 
1647 #if TERI
1648 extern void     thread_exception_return_internal(void) __dead2;
1649 
1650 void
thread_exception_return(void)1651 thread_exception_return(void)
1652 {
1653 	thread_t thread = current_thread();
1654 	task_t   task   = current_task();
1655 
1656 	ml_set_interrupts_enabled(FALSE);
1657 	if (thread_is_64bit_addr(thread) != task_has_64Bit_addr(task)) {
1658 		panic("Task/thread bitness mismatch %p %p, task: %d, thread: %d",
1659 		    thread, task, thread_is_64bit_addr(thread), task_has_64Bit_addr(task));
1660 	}
1661 
1662 	if (thread_is_64bit_addr(thread)) {
1663 		if ((gdt_desc_p(USER64_CS)->access & ACC_PL_U) == 0) {
1664 			panic("64-GDT mismatch %p, descriptor: %p", thread, gdt_desc_p(USER64_CS));
1665 		}
1666 	} else {
1667 		if ((gdt_desc_p(USER_CS)->access & ACC_PL_U) == 0) {
1668 			panic("32-GDT mismatch %p, descriptor: %p", thread, gdt_desc_p(USER_CS));
1669 		}
1670 	}
1671 	assert(get_preemption_level() == 0);
1672 	thread_exception_return_internal();
1673 }
1674 #endif
1675 
1676 #if DEVELOPMENT || DEBUG
1677 static int trap_handled;
1678 
1679 static void
handle_recoverable_kernel_trap(__unused void * tstate,uint16_t comment)1680 handle_recoverable_kernel_trap(
1681 	__unused void     *tstate,
1682 	uint16_t          comment)
1683 {
1684 	assert(comment == TEST_RECOVERABLE_SOFT_TRAP);
1685 
1686 	printf("Recoverable trap handled.\n");
1687 	trap_handled = 1;
1688 }
1689 
1690 KERNEL_BRK_DESCRIPTOR_DEFINE(test_desc,
1691     .type                = KERNEL_BRK_TYPE_TEST,
1692     .base                = TEST_RECOVERABLE_SOFT_TRAP,
1693     .max                 = TEST_RECOVERABLE_SOFT_TRAP,
1694     .options             = KERNEL_BRK_RECOVERABLE,
1695     .handle_breakpoint   = handle_recoverable_kernel_trap);
1696 
1697 static int
recoverable_kernel_trap_test(__unused int64_t in,int64_t * out)1698 recoverable_kernel_trap_test(__unused int64_t in, int64_t *out)
1699 {
1700 	ml_recoverable_trap(TEST_RECOVERABLE_SOFT_TRAP);
1701 
1702 	*out = trap_handled;
1703 	return 0;
1704 }
1705 
1706 SYSCTL_TEST_REGISTER(recoverable_kernel_trap, recoverable_kernel_trap_test);
1707 #endif
1708