xref: /xnu-12377.41.6/osfmk/i386/trap.c (revision bbb1b6f9e71b8cdde6e5cd6f4841f207dee3d828)
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 #include <vm/vm_map_xnu.h>
82 
83 #include <kern/kern_types.h>
84 #include <kern/processor.h>
85 #include <kern/thread.h>
86 #include <kern/task.h>
87 #include <kern/restartable.h>
88 #include <kern/sched.h>
89 #include <kern/sched_prim.h>
90 #include <kern/exception.h>
91 #include <kern/spl.h>
92 #include <kern/misc_protos.h>
93 #include <kern/debug.h>
94 #if CONFIG_TELEMETRY
95 #include <kern/telemetry.h>
96 #include <kern/trap_telemetry.h>
97 #endif
98 #include <kern/zalloc_internal.h>
99 #include <sys/kdebug.h>
100 #include <kperf/kperf.h>
101 #include <prng/random.h>
102 #include <prng/entropy.h>
103 
104 #include <string.h>
105 
106 #include <i386/postcode.h>
107 #include <i386/mp_desc.h>
108 #include <i386/proc_reg.h>
109 #include <i386/machine_routines.h>
110 #if CONFIG_MCA
111 #include <i386/machine_check.h>
112 #endif
113 #include <mach/i386/syscall_sw.h>
114 
115 #include <libkern/OSDebug.h>
116 #include <i386/cpu_threads.h>
117 #include <machine/pal_routines.h>
118 #include <i386/lbr.h>
119 
120 extern void throttle_lowpri_io(int);
121 extern void kprint_state(x86_saved_state64_t *saved_state);
122 #if DEVELOPMENT || DEBUG
123 int insnstream_force_cacheline_mismatch = 0;
124 extern int panic_on_cacheline_mismatch;
125 extern char panic_on_trap_procname[];
126 extern uint32_t panic_on_trap_mask;
127 #endif
128 
129 extern int insn_copyin_count;
130 
131 /*
132  * Forward declarations
133  */
134 static void panic_trap(x86_saved_state64_t *saved_state, uint16_t comment, const char *trapreason, uint32_t pl, kern_return_t fault_result) __dead2;
135 static void set_recovery_ip(x86_saved_state64_t *saved_state, vm_offset_t ip);
136 #if DEVELOPMENT || DEBUG
137 static __attribute__((noinline)) void copy_instruction_stream(thread_t thread, uint64_t rip, int trap_code, bool inspect_cacheline);
138 #else
139 static __attribute__((noinline)) void copy_instruction_stream(thread_t thread, uint64_t rip, int trap_code);
140 #endif
141 
142 #if CONFIG_DTRACE
143 /* See <rdar://problem/4613924> */
144 perfCallback tempDTraceTrapHook = NULL; /* Pointer to DTrace fbt trap hook routine */
145 
146 extern boolean_t dtrace_tally_fault(user_addr_t);
147 extern boolean_t dtrace_handle_trap(int, x86_saved_state_t *);
148 #endif
149 
150 #ifdef MACH_BSD
151 extern char *   proc_name_address(void *p);
152 #endif /* MACH_BSD */
153 
154 extern boolean_t pmap_smep_enabled;
155 extern boolean_t pmap_smap_enabled;
156 
157 __attribute__((noreturn))
158 void
thread_syscall_return(kern_return_t ret)159 thread_syscall_return(
160 	kern_return_t ret)
161 {
162 	thread_t        thr_act = current_thread();
163 	boolean_t       is_mach;
164 	int             code;
165 
166 	pal_register_cache_state(thr_act, DIRTY);
167 
168 	if (thread_is_64bit_addr(thr_act)) {
169 		x86_saved_state64_t     *regs;
170 
171 		regs = USER_REGS64(thr_act);
172 
173 		code = (int) (regs->rax & SYSCALL_NUMBER_MASK);
174 		is_mach = (regs->rax & SYSCALL_CLASS_MASK)
175 		    == (SYSCALL_CLASS_MACH << SYSCALL_CLASS_SHIFT);
176 		if (kdebug_enable && is_mach) {
177 			/* Mach trap */
178 			KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
179 			    MACHDBG_CODE(DBG_MACH_EXCP_SC, code) | DBG_FUNC_END,
180 			    ret, 0, 0, 0, 0);
181 		}
182 		regs->rax = ret;
183 #if DEBUG
184 		if (is_mach) {
185 			DEBUG_KPRINT_SYSCALL_MACH(
186 				"thread_syscall_return: 64-bit mach ret=%u\n",
187 				ret);
188 		} else {
189 			DEBUG_KPRINT_SYSCALL_UNIX(
190 				"thread_syscall_return: 64-bit unix ret=%u\n",
191 				ret);
192 		}
193 #endif
194 	} else {
195 		x86_saved_state32_t     *regs;
196 
197 		regs = USER_REGS32(thr_act);
198 
199 		code = ((int) regs->eax);
200 		is_mach = (code < 0);
201 		if (kdebug_enable && is_mach) {
202 			/* Mach trap */
203 			KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
204 			    MACHDBG_CODE(DBG_MACH_EXCP_SC, -code) | DBG_FUNC_END,
205 			    ret, 0, 0, 0, 0);
206 		}
207 		regs->eax = ret;
208 #if DEBUG
209 		if (is_mach) {
210 			DEBUG_KPRINT_SYSCALL_MACH(
211 				"thread_syscall_return: 32-bit mach ret=%u\n",
212 				ret);
213 		} else {
214 			DEBUG_KPRINT_SYSCALL_UNIX(
215 				"thread_syscall_return: 32-bit unix ret=%u\n",
216 				ret);
217 		}
218 #endif
219 	}
220 
221 #if DEBUG || DEVELOPMENT
222 	kern_allocation_name_t
223 	prior __assert_only = thread_get_kernel_state(thr_act)->allocation_name;
224 	assertf(prior == NULL, "thread_set_allocation_name(\"%s\") not cleared", kern_allocation_get_name(prior));
225 #endif /* DEBUG || DEVELOPMENT */
226 
227 	throttle_lowpri_io(1);
228 
229 	thread_exception_return();
230 	/*NOTREACHED*/
231 }
232 
233 /*
234  * Fault recovery in copyin/copyout routines.
235  */
236 struct recovery {
237 	uintptr_t       fault_addr;
238 	uintptr_t       recover_addr;
239 };
240 
241 extern struct recovery  recover_table[];
242 extern struct recovery  recover_table_end[];
243 
244 const char *    trap_type[] = {TRAP_NAMES};
245 unsigned        TRAP_TYPES = sizeof(trap_type) / sizeof(trap_type[0]);
246 
247 extern void     PE_incoming_interrupt(int interrupt);
248 
249 #if defined(__x86_64__) && DEBUG
250 void
kprint_state(x86_saved_state64_t * saved_state)251 kprint_state(x86_saved_state64_t        *saved_state)
252 {
253 	kprintf("current_cpu_datap() 0x%lx\n", (uintptr_t)current_cpu_datap());
254 	kprintf("Current GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_GS_BASE));
255 	kprintf("Kernel  GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_KERNEL_GS_BASE));
256 	kprintf("state at 0x%lx:\n", (uintptr_t) saved_state);
257 
258 	kprintf("      rdi    0x%llx\n", saved_state->rdi);
259 	kprintf("      rsi    0x%llx\n", saved_state->rsi);
260 	kprintf("      rdx    0x%llx\n", saved_state->rdx);
261 	kprintf("      r10    0x%llx\n", saved_state->r10);
262 	kprintf("      r8     0x%llx\n", saved_state->r8);
263 	kprintf("      r9     0x%llx\n", saved_state->r9);
264 
265 	kprintf("      cr2    0x%llx\n", saved_state->cr2);
266 	kprintf("real  cr2    0x%lx\n", get_cr2());
267 	kprintf("      r15    0x%llx\n", saved_state->r15);
268 	kprintf("      r14    0x%llx\n", saved_state->r14);
269 	kprintf("      r13    0x%llx\n", saved_state->r13);
270 	kprintf("      r12    0x%llx\n", saved_state->r12);
271 	kprintf("      r11    0x%llx\n", saved_state->r11);
272 	kprintf("      rbp    0x%llx\n", saved_state->rbp);
273 	kprintf("      rbx    0x%llx\n", saved_state->rbx);
274 	kprintf("      rcx    0x%llx\n", saved_state->rcx);
275 	kprintf("      rax    0x%llx\n", saved_state->rax);
276 
277 	kprintf("      gs     0x%x\n", saved_state->gs);
278 	kprintf("      fs     0x%x\n", saved_state->fs);
279 
280 	kprintf("  isf.trapno 0x%x\n", saved_state->isf.trapno);
281 	kprintf("  isf._pad   0x%x\n", saved_state->isf._pad);
282 	kprintf("  isf.trapfn 0x%llx\n", saved_state->isf.trapfn);
283 	kprintf("  isf.err    0x%llx\n", saved_state->isf.err);
284 	kprintf("  isf.rip    0x%llx\n", saved_state->isf.rip);
285 	kprintf("  isf.cs     0x%llx\n", saved_state->isf.cs);
286 	kprintf("  isf.rflags 0x%llx\n", saved_state->isf.rflags);
287 	kprintf("  isf.rsp    0x%llx\n", saved_state->isf.rsp);
288 	kprintf("  isf.ss     0x%llx\n", saved_state->isf.ss);
289 }
290 #endif
291 
292 
293 /*
294  * Non-zero indicates latency assert is enabled and capped at valued
295  * absolute time units.
296  */
297 
298 uint64_t interrupt_latency_cap = 0;
299 boolean_t ilat_assert = FALSE;
300 
301 void
interrupt_latency_tracker_setup(void)302 interrupt_latency_tracker_setup(void)
303 {
304 	uint32_t ilat_cap_us;
305 	if (PE_parse_boot_argn("interrupt_latency_cap_us", &ilat_cap_us, sizeof(ilat_cap_us))) {
306 		interrupt_latency_cap = ilat_cap_us * NSEC_PER_USEC;
307 		nanoseconds_to_absolutetime(interrupt_latency_cap, &interrupt_latency_cap);
308 	} else {
309 		interrupt_latency_cap = LockTimeOut;
310 	}
311 	PE_parse_boot_argn("-interrupt_latency_assert_enable", &ilat_assert, sizeof(ilat_assert));
312 }
313 
314 void
interrupt_reset_latency_stats(void)315 interrupt_reset_latency_stats(void)
316 {
317 	uint32_t i;
318 	for (i = 0; i < real_ncpus; i++) {
319 		cpu_data_ptr[i]->cpu_max_observed_int_latency =
320 		    cpu_data_ptr[i]->cpu_max_observed_int_latency_vector = 0;
321 	}
322 }
323 
324 void
interrupt_populate_latency_stats(char * buf,unsigned bufsize)325 interrupt_populate_latency_stats(char *buf, unsigned bufsize)
326 {
327 	uint32_t i, tcpu = ~0;
328 	uint64_t cur_max = 0;
329 
330 	for (i = 0; i < real_ncpus; i++) {
331 		if (cur_max < cpu_data_ptr[i]->cpu_max_observed_int_latency) {
332 			cur_max = cpu_data_ptr[i]->cpu_max_observed_int_latency;
333 			tcpu = i;
334 		}
335 	}
336 
337 	if (tcpu < real_ncpus) {
338 		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);
339 	}
340 }
341 
342 uint32_t interrupt_timer_coalescing_enabled = 1;
343 uint64_t interrupt_coalesced_timers;
344 
345 /*
346  * Handle interrupts:
347  *  - local APIC interrupts (IPIs, timers, etc) are handled by the kernel,
348  *  - device interrupts go to the platform expert.
349  */
350 void
interrupt(x86_saved_state_t * state)351 interrupt(x86_saved_state_t *state)
352 {
353 	uint64_t        rip;
354 	uint64_t        rsp;
355 	int             interrupt_num;
356 	boolean_t       user_mode = FALSE;
357 	int             ipl;
358 	int             cnum = cpu_number();
359 	cpu_data_t      *cdp = cpu_data_ptr[cnum];
360 	int             itype = DBG_INTR_TYPE_UNKNOWN;
361 	int             handled;
362 
363 
364 	x86_saved_state64_t     *state64 = saved_state64(state);
365 	rip = state64->isf.rip;
366 	rsp = state64->isf.rsp;
367 	interrupt_num = state64->isf.trapno;
368 	if (state64->isf.cs & 0x03) {
369 		user_mode = TRUE;
370 	}
371 
372 #if DEVELOPMENT || DEBUG
373 	uint64_t frameptr = is_saved_state64(state) ? state64->rbp : saved_state32(state)->ebp;
374 	uint32_t traptrace_index = traptrace_start(interrupt_num, rip, mach_absolute_time(), frameptr);
375 #endif
376 
377 	if (cpu_data_ptr[cnum]->lcpu.package->num_idle == topoParms.nLThreadsPerPackage) {
378 		cpu_data_ptr[cnum]->cpu_hwIntpexits[interrupt_num]++;
379 	}
380 
381 	if (interrupt_num == (LAPIC_DEFAULT_INTERRUPT_BASE + LAPIC_INTERPROCESSOR_INTERRUPT)) {
382 		itype = DBG_INTR_TYPE_IPI;
383 	} else if (interrupt_num == (LAPIC_DEFAULT_INTERRUPT_BASE + LAPIC_TIMER_INTERRUPT)) {
384 		itype = DBG_INTR_TYPE_TIMER;
385 	} else {
386 		itype = DBG_INTR_TYPE_OTHER;
387 	}
388 
389 	KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
390 	    MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_START,
391 	    interrupt_num,
392 	    (user_mode ? rip : VM_KERNEL_UNSLIDE(rip)),
393 	    user_mode, itype, 0);
394 
395 	SCHED_STATS_INC(interrupt_count);
396 
397 	ipl = get_preemption_level();
398 
399 	/*
400 	 * Handle local APIC interrupts
401 	 * else call platform expert for devices.
402 	 */
403 	handled = lapic_interrupt(interrupt_num, state);
404 
405 	if (!handled) {
406 		if (interrupt_num == (LAPIC_DEFAULT_INTERRUPT_BASE + LAPIC_CMCI_INTERRUPT)) {
407 			/*
408 			 * CMCI can be signalled on any logical processor, and the kexts
409 			 * that implement handling CMCI use IOKit to register handlers for
410 			 * the CMCI vector, so if we see a CMCI, do not encode a CPU
411 			 * number in bits 8:31 (since the vector is the same regardless of
412 			 * the handling CPU).
413 			 */
414 			PE_incoming_interrupt(interrupt_num);
415 		} else if (cnum <= lapic_max_interrupt_cpunum) {
416 			PE_incoming_interrupt((cnum << 8) | interrupt_num);
417 		}
418 	}
419 
420 	if (__improbable(get_preemption_level() != ipl)) {
421 		panic("Preemption level altered by interrupt vector 0x%x: initial 0x%x, final: 0x%x", interrupt_num, ipl, get_preemption_level());
422 	}
423 
424 
425 	if (__improbable(cdp->cpu_nested_istack)) {
426 		cdp->cpu_nested_istack_events++;
427 	} else {
428 		uint64_t ctime = mach_absolute_time();
429 		uint64_t int_latency = ctime - cdp->cpu_int_event_time;
430 		uint64_t esdeadline, ehdeadline;
431 		/* Attempt to process deferred timers in the context of
432 		 * this interrupt, unless interrupt time has already exceeded
433 		 * TCOAL_ILAT_THRESHOLD.
434 		 */
435 #define TCOAL_ILAT_THRESHOLD (30000ULL)
436 
437 		if ((int_latency < TCOAL_ILAT_THRESHOLD) &&
438 		    interrupt_timer_coalescing_enabled) {
439 			esdeadline = cdp->rtclock_timer.queue.earliest_soft_deadline;
440 			ehdeadline = cdp->rtclock_timer.deadline;
441 			if ((ctime >= esdeadline) && (ctime < ehdeadline)) {
442 				interrupt_coalesced_timers++;
443 				TCOAL_DEBUG(0x88880000 | DBG_FUNC_START, ctime, esdeadline, ehdeadline, interrupt_coalesced_timers, 0);
444 				rtclock_intr(state);
445 				TCOAL_DEBUG(0x88880000 | DBG_FUNC_END, ctime, esdeadline, interrupt_coalesced_timers, 0, 0);
446 			} else {
447 				TCOAL_DEBUG(0x77770000, ctime, cdp->rtclock_timer.queue.earliest_soft_deadline, cdp->rtclock_timer.deadline, interrupt_coalesced_timers, 0);
448 			}
449 		}
450 
451 		if (__improbable(ilat_assert && (int_latency > interrupt_latency_cap) && !machine_timeout_suspended())) {
452 			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);
453 		}
454 
455 		if (__improbable(int_latency > cdp->cpu_max_observed_int_latency)) {
456 			cdp->cpu_max_observed_int_latency = int_latency;
457 			cdp->cpu_max_observed_int_latency_vector = interrupt_num;
458 		}
459 	}
460 
461 	/*
462 	 * Having serviced the interrupt first, look at the interrupted stack depth.
463 	 */
464 	if (!user_mode) {
465 		uint64_t depth = cdp->cpu_kernel_stack
466 		    + sizeof(struct thread_kernel_state)
467 		    + sizeof(struct i386_exception_link *)
468 		    - rsp;
469 		if (__improbable(depth > kernel_stack_depth_max)) {
470 			kernel_stack_depth_max = (vm_offset_t)depth;
471 			KERNEL_DEBUG_CONSTANT(
472 				MACHDBG_CODE(DBG_MACH_SCHED, MACH_STACK_DEPTH),
473 				(long) depth, (long) VM_KERNEL_UNSLIDE(rip), 0, 0, 0);
474 		}
475 	}
476 
477 	if (cnum == master_cpu) {
478 		entropy_collect();
479 	}
480 
481 #if KPERF
482 	kperf_interrupt();
483 #endif /* KPERF */
484 
485 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_END,
486 	    interrupt_num);
487 
488 	assert(ml_get_interrupts_enabled() == FALSE);
489 
490 #if DEVELOPMENT || DEBUG
491 	if (traptrace_index != TRAPTRACE_INVALID_INDEX) {
492 		traptrace_end(traptrace_index, mach_absolute_time());
493 	}
494 #endif
495 }
496 
497 static inline void
reset_dr7(void)498 reset_dr7(void)
499 {
500 	long dr7 = 0x400; /* magic dr7 reset value; 32 bit on i386, 64 bit on x86_64 */
501 	__asm__ volatile ("mov %0,%%dr7" : : "r" (dr7));
502 }
503 #if MACH_KDP
504 unsigned kdp_has_active_watchpoints = 0;
505 #define NO_WATCHPOINTS (!kdp_has_active_watchpoints)
506 #else
507 #define NO_WATCHPOINTS 1
508 #endif
509 
510 static uint32_t bound_chk_violations_event;
511 
512 static const char *
xnu_soft_trap_handle_breakpoint(void * tstate,uint16_t comment)513 xnu_soft_trap_handle_breakpoint(void *tstate, uint16_t comment)
514 {
515 #pragma unused(tstate)
516 	if (comment == CLANG_SOFT_TRAP_BOUND_CHK) {
517 		os_atomic_inc(&bound_chk_violations_event, relaxed);
518 	}
519 
520 	return NULL;
521 }
522 
523 static const char *
xnu_hard_trap_handle_breakpoint(void * tstate,uint16_t comment)524 xnu_hard_trap_handle_breakpoint(void *tstate, uint16_t comment)
525 {
526 	kernel_panic_reason_t pr = PERCPU_GET(panic_reason);
527 	x86_saved_state64_t *state = tstate;
528 
529 	switch (comment) {
530 	case XNU_HARD_TRAP_SAFE_UNLINK:
531 		snprintf(pr->buf, sizeof(pr->buf),
532 		    "panic: corrupt list around element %p",
533 		    (void *)state->rax);
534 		return pr->buf;
535 
536 	case XNU_HARD_TRAP_STRING_CHK:
537 		return "panic: string operation caused an overflow";
538 
539 	case XNU_HARD_TRAP_ASSERT_FAILURE:
540 		/*
541 		 * Read the implicit assert arguments, see:
542 		 * ML_TRAP_REGISTER_1: rax
543 		 * ML_TRAP_REGISTER_2: r10
544 		 * ML_TRAP_REGISTER_3: r11
545 		 */
546 		panic_assert_format(pr->buf, sizeof(pr->buf),
547 		    (struct mach_assert_hdr *)state->rax,
548 		    state->r10, state->r11);
549 		return pr->buf;
550 
551 	default:
552 		return NULL;
553 	}
554 }
555 
556 KERNEL_BRK_DESCRIPTOR_DEFINE(clang_desc,
557     .type                = TRAP_TELEMETRY_TYPE_KERNEL_BRK_CLANG,
558     .base                = CLANG_X86_TRAP_START,
559     .max                 = CLANG_X86_TRAP_END,
560     .options             = BRK_TELEMETRY_OPTIONS_FATAL_DEFAULT,
561     .handle_breakpoint   = NULL);
562 
563 KERNEL_BRK_DESCRIPTOR_DEFINE(xnu_soft_traps_desc,
564     .type                = TRAP_TELEMETRY_TYPE_KERNEL_BRK_TELEMETRY,
565     .base                = XNU_SOFT_TRAP_START,
566     .max                 = XNU_SOFT_TRAP_END,
567     .options             = BRK_TELEMETRY_OPTIONS_RECOVERABLE_DEFAULT(
568 	    /* enable_telemetry */ true),
569     .handle_breakpoint   = xnu_soft_trap_handle_breakpoint);
570 
571 KERNEL_BRK_DESCRIPTOR_DEFINE(libcxx_desc,
572     .type                = TRAP_TELEMETRY_TYPE_KERNEL_BRK_LIBCXX,
573     .base                = LIBCXX_TRAP_START,
574     .max                 = LIBCXX_TRAP_END,
575     .options             = BRK_TELEMETRY_OPTIONS_FATAL_DEFAULT,
576     .handle_breakpoint   = NULL);
577 
578 KERNEL_BRK_DESCRIPTOR_DEFINE(xnu_hard_traps_desc,
579     .type                = TRAP_TELEMETRY_TYPE_KERNEL_BRK_XNU,
580     .base                = XNU_HARD_TRAP_START,
581     .max                 = XNU_HARD_TRAP_END,
582     .options             = BRK_TELEMETRY_OPTIONS_FATAL_DEFAULT,
583     .handle_breakpoint   = xnu_hard_trap_handle_breakpoint);
584 
585 static bool
handle_kernel_breakpoint(x86_saved_state64_t * state,const char ** reason,uint16_t * out_comment)586 handle_kernel_breakpoint(
587 	x86_saved_state64_t    *state,
588 	const char            **reason,
589 	uint16_t               *out_comment)
590 {
591 	uint16_t comment;
592 	const struct kernel_brk_descriptor *desc;
593 	uint8_t inst_buf[8];
594 	uint32_t prefix16 = 0x80B90F67; /* Encoding prefix for ud1 <16-bit code>(%eax), %eax */
595 	uint32_t prefix8 = 0x40B90F67; /* Encoding prefix for ud1 <8-bit code>(%eax), %eax */
596 	bool found_prefix8 = false;
597 
598 	vm_size_t sz = ml_nofault_copy(state->isf.rip, (vm_offset_t)inst_buf, sizeof(inst_buf));
599 	if (sz != sizeof(inst_buf)) {
600 		return false;
601 	}
602 
603 	if (bcmp(inst_buf, &prefix16, sizeof(prefix16)) == 0) {
604 		/* The two bytes following the prefix is our code */
605 		comment = inst_buf[5] << 8 | inst_buf[4];
606 	} else if (bcmp(inst_buf, &prefix8, sizeof(prefix8)) == 0) {
607 		/* The one byte following the prefix is our code */
608 		found_prefix8 = true;
609 		comment = inst_buf[4];
610 	} else {
611 		return false;
612 	}
613 
614 	if (out_comment) {
615 		*out_comment = comment;
616 	}
617 	desc = find_kernel_brk_descriptor_by_comment(comment);
618 
619 	if (!desc) {
620 		return false;
621 	}
622 
623 	if (desc->options.enable_trap_telemetry) {
624 		trap_telemetry_report_exception(
625 			/* trap_type   */ desc->type,
626 			/* trap_code   */ comment,
627 			/* options     */ desc->options.telemetry_options,
628 			/* saved_state */ (void *)state);
629 	}
630 
631 	if (desc->handle_breakpoint) {
632 		*reason = desc->handle_breakpoint(state, comment);
633 	}
634 
635 	/* Still alive? Check if we should recover. */
636 	if (desc->options.recoverable) {
637 		/* ud1 can be five or eight-byte long depending on the prefix */
638 		set_recovery_ip(state, state->isf.rip + (found_prefix8 ? 5 : 8));
639 		return true;
640 	}
641 
642 	return false;
643 }
644 
645 // Find a recovery entry for an instruction address if one is present.
646 static struct recovery const*
find_recovery_entry(vm_offset_t kern_ip)647 find_recovery_entry(vm_offset_t kern_ip)
648 {
649 	for (struct recovery const* rp = recover_table; rp < recover_table_end; rp++) {
650 		if (kern_ip == rp->fault_addr) {
651 			return rp;
652 		}
653 	}
654 	return NULL;
655 }
656 
657 /*
658  * Trap from kernel mode.  Only page-fault errors are recoverable,
659  * and then only in special circumstances.  All other errors are
660  * fatal.  Return value indicates if trap was handled.
661  */
662 
663 void
kernel_trap(x86_saved_state_t * state,uintptr_t * lo_spp)664 kernel_trap(
665 	x86_saved_state_t       *state,
666 	uintptr_t *lo_spp)
667 {
668 	const char             *reason = NULL;
669 	uint16_t                trapcomment = 0;
670 
671 	x86_saved_state64_t     *saved_state;
672 	int                     code;
673 	user_addr_t             vaddr;
674 	int                     type;
675 	vm_map_t                map = 0;        /* protected by T_PAGE_FAULT */
676 	kern_return_t           result = KERN_FAILURE;
677 	kern_return_t           fault_result = KERN_SUCCESS;
678 	thread_t                thread;
679 	boolean_t               intr;
680 	vm_prot_t               prot;
681 	struct recovery const   *rp = NULL;
682 	vm_offset_t             kern_ip;
683 	int                     is_user;
684 	int                     trap_pl = get_preemption_level();
685 
686 	thread = current_thread();
687 
688 	if (__improbable(is_saved_state32(state))) {
689 		panic("kernel_trap(%p) with 32-bit state", state);
690 	}
691 	saved_state = saved_state64(state);
692 
693 	/* Record cpu where state was captured */
694 	saved_state->isf.cpu = cpu_number();
695 
696 	vaddr = (user_addr_t)saved_state->cr2;
697 	type  = saved_state->isf.trapno;
698 	code  = (int)(saved_state->isf.err & 0xffff);
699 	intr  = (saved_state->isf.rflags & EFL_IF) != 0;        /* state of ints at trap */
700 	kern_ip = (vm_offset_t)saved_state->isf.rip;
701 
702 	is_user = (vaddr < VM_MAX_USER_PAGE_ADDRESS);
703 
704 #if DEVELOPMENT || DEBUG
705 	uint32_t traptrace_index = traptrace_start(type, kern_ip, mach_absolute_time(), saved_state->rbp);
706 #endif
707 
708 #if CONFIG_DTRACE
709 	/*
710 	 * Is there a DTrace hook?
711 	 */
712 	if (__improbable(tempDTraceTrapHook != NULL)) {
713 		if (tempDTraceTrapHook(type, state, lo_spp, 0) == KERN_SUCCESS) {
714 			/*
715 			 * If it succeeds, we are done...
716 			 */
717 			goto common_return;
718 		}
719 	}
720 
721 	/* Handle traps originated from probe context. */
722 	if (thread != THREAD_NULL && thread->t_dtrace_inprobe) {
723 		if (dtrace_handle_trap(type, state)) {
724 			goto common_return;
725 		}
726 	}
727 
728 #endif /* CONFIG_DTRACE */
729 
730 	/*
731 	 * we come here with interrupts off as we don't want to recurse
732 	 * on preemption below.  but we do want to re-enable interrupts
733 	 * as soon we possibly can to hold latency down
734 	 */
735 	if (__improbable(T_PREEMPT == type)) {
736 		ast_taken_kernel();
737 
738 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
739 		    (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86, type)) | DBG_FUNC_NONE,
740 		    0, 0, 0, VM_KERNEL_UNSLIDE(kern_ip), 0);
741 
742 		goto common_return;
743 	}
744 
745 	user_addr_t     kd_vaddr = is_user ? vaddr : VM_KERNEL_UNSLIDE(vaddr);
746 	KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
747 	    (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86, type)) | DBG_FUNC_NONE,
748 	    (unsigned)(kd_vaddr >> 32), (unsigned)kd_vaddr, is_user,
749 	    VM_KERNEL_UNSLIDE(kern_ip), 0);
750 
751 
752 	if (T_PAGE_FAULT == type) {
753 		/*
754 		 * assume we're faulting in the kernel map
755 		 */
756 		map = kernel_map;
757 
758 		if (__probable((thread != THREAD_NULL) && (thread->map != kernel_map) &&
759 		    (vaddr < VM_MAX_USER_PAGE_ADDRESS))) {
760 			/* fault occurred in userspace */
761 			map = thread->map;
762 
763 			/* Intercept a potential Supervisor Mode Execute
764 			 * Protection fault. These criteria identify
765 			 * both NX faults and SMEP faults, but both
766 			 * are fatal. We avoid checking PTEs (racy).
767 			 * (The VM could just redrive a SMEP fault, hence
768 			 * the intercept).
769 			 */
770 			if (__improbable((code == (T_PF_PROT | T_PF_EXECUTE)) &&
771 			    (pmap_smep_enabled) && (saved_state->isf.rip == vaddr))) {
772 				goto debugger_entry;
773 			}
774 
775 			/*
776 			 * Additionally check for SMAP faults...
777 			 * which are characterized by page-present and
778 			 * the AC bit unset (i.e. not from copyin/out path).
779 			 */
780 			if (__improbable(code & T_PF_PROT &&
781 			    pmap_smap_enabled &&
782 			    (saved_state->isf.rflags & EFL_AC) == 0)) {
783 				goto debugger_entry;
784 			}
785 
786 			/*
787 			 * If we're not sharing cr3 with the user
788 			 * and we faulted in copyio,
789 			 * then switch cr3 here and dismiss the fault.
790 			 */
791 			if (no_shared_cr3 &&
792 			    (thread->machine.specFlags & CopyIOActive) &&
793 			    map->pmap->pm_cr3 != get_cr3_base()) {
794 				pmap_assert(current_cpu_datap()->cpu_pmap_pcid_enabled == FALSE);
795 				set_cr3_raw(map->pmap->pm_cr3);
796 				return;
797 			}
798 			if (__improbable(vaddr < PAGE_SIZE) &&
799 			    ((thread->machine.specFlags & CopyIOActive) == 0)) {
800 				goto debugger_entry;
801 			}
802 		}
803 	}
804 
805 	(void) ml_set_interrupts_enabled(intr);
806 
807 	switch (type) {
808 	case T_NO_FPU:
809 		fpnoextflt();
810 		goto common_return;
811 
812 	case T_FPU_FAULT:
813 		fpextovrflt();
814 		goto common_return;
815 
816 	case T_FLOATING_POINT_ERROR:
817 		fpexterrflt();
818 		goto common_return;
819 
820 	case T_SSE_FLOAT_ERROR:
821 		fpSSEexterrflt();
822 		goto common_return;
823 
824 	case T_INVALID_OPCODE:
825 		if (handle_kernel_breakpoint(saved_state, &reason, &trapcomment)) {
826 			goto common_return;
827 		}
828 		fpUDflt(kern_ip);
829 		goto debugger_entry;
830 
831 	case T_DEBUG:
832 		/*
833 		 * Re-enable LBR tracing for core/panic files if necessary. i386_lbr_enable confirms LBR should be re-enabled.
834 		 */
835 		i386_lbr_enable();
836 		if ((saved_state->isf.rflags & EFL_TF) == 0 && NO_WATCHPOINTS) {
837 			/* We've somehow encountered a debug
838 			 * register match that does not belong
839 			 * to the kernel debugger.
840 			 * This isn't supposed to happen.
841 			 */
842 			reset_dr7();
843 			goto common_return;
844 		}
845 		goto debugger_entry;
846 	case T_INT3:
847 		goto debugger_entry;
848 	case T_PAGE_FAULT:
849 
850 #if CONFIG_DTRACE
851 		if (thread != THREAD_NULL && thread->t_dtrace_inprobe) { /* Executing under dtrace_probe? */
852 			if (dtrace_tally_fault(vaddr)) { /* Should a fault under dtrace be ignored? */
853 				/*
854 				 * DTrace has "anticipated" the possibility of this fault, and has
855 				 * established the suitable recovery state. Drop down now into the
856 				 * recovery handling code in "case T_GENERAL_PROTECTION:".
857 				 */
858 				goto FALL_THROUGH;
859 			}
860 		}
861 #endif /* CONFIG_DTRACE */
862 
863 		prot = VM_PROT_READ;
864 
865 		if (code & T_PF_WRITE) {
866 			prot |= VM_PROT_WRITE;
867 		}
868 		if (code & T_PF_EXECUTE) {
869 			prot |= VM_PROT_EXECUTE;
870 		}
871 
872 		/**
873 		 * vm_fault() can be called with preemption disabled (and indeed this is expected for
874 		 * certain copyio() scenarios), but can't safely be called with interrupts disabled
875 		 * once the system has gone multi-threaded.  Other than some early-boot situations
876 		 * such as startup kext loading, kernel paging operations should never be triggered
877 		 * by non-interruptible code in the first place, so a fault from such a context will
878 		 * ultimately produce a kernel page fault panic anyway.  In these cases, skip calling
879 		 * vm_fault() to avoid masking the real kernel panic with a failed VM locking assertion.
880 		 */
881 		if (__improbable(!(intr ||
882 		    startup_phase < STARTUP_SUB_EARLY_BOOT ||
883 		    current_cpu_datap()->cpu_hibernate))) {
884 			fault_result = result = KERN_FAILURE;
885 			goto FALL_THROUGH;
886 		}
887 
888 		// VM will query this property when deciding to throttle this fault, we don't want to
889 		// throttle kernel faults for copyio faults. The presence of a recovery entry is used as a
890 		// proxy for being in copyio code.
891 		rp = find_recovery_entry(kern_ip);
892 		const bool was_recover = thread->recover;
893 		thread->recover = was_recover || (rp != NULL);
894 
895 		fault_result = result = vm_fault(map,
896 		    vaddr,
897 		    prot,
898 		    FALSE, VM_KERN_MEMORY_NONE,
899 		    THREAD_UNINT, NULL, 0);
900 
901 		thread->recover = was_recover;
902 		if (result == KERN_SUCCESS) {
903 			goto common_return;
904 		}
905 		/*
906 		 * fall through
907 		 */
908 FALL_THROUGH:
909 
910 	case T_GENERAL_PROTECTION:
911 		/*
912 		 * If there is a failure recovery address
913 		 * for this fault, go there.
914 		 */
915 		if ((rp != NULL) || (rp = find_recovery_entry(kern_ip))) {
916 			set_recovery_ip(saved_state, rp->recover_addr);
917 			goto common_return;
918 		}
919 
920 		/*
921 		 * Unanticipated page-fault errors in kernel
922 		 * should not happen.
923 		 *
924 		 * fall through...
925 		 */
926 		OS_FALLTHROUGH;
927 	default:
928 		/*
929 		 * Exception 15 is reserved but some chips may generate it
930 		 * spuriously. Seen at startup on AMD Athlon-64.
931 		 */
932 		if (type == 15) {
933 			kprintf("kernel_trap() ignoring spurious trap 15\n");
934 			goto common_return;
935 		}
936 debugger_entry:
937 		/* Ensure that the i386_kernel_state at the base of the
938 		 * current thread's stack (if any) is synchronized with the
939 		 * context at the moment of the trap, to facilitate
940 		 * access through the debugger.
941 		 */
942 		sync_iss_to_iks(state);
943 #if  MACH_KDP
944 		if (kdp_i386_trap(type, saved_state, result, (vm_offset_t)vaddr)) {
945 			goto common_return;
946 		}
947 #endif
948 	}
949 	if (type == T_PAGE_FAULT) {
950 		panic_fault_address = vaddr;
951 	}
952 	pal_cli();
953 
954 	panic_trap(saved_state, trapcomment, reason, trap_pl, fault_result);
955 	/*
956 	 * NO RETURN
957 	 */
958 
959 common_return:
960 #if DEVELOPMENT || DEBUG
961 	if (traptrace_index != TRAPTRACE_INVALID_INDEX) {
962 		traptrace_end(traptrace_index, mach_absolute_time());
963 	}
964 #endif
965 	return;
966 }
967 
968 static void
set_recovery_ip(x86_saved_state64_t * saved_state,vm_offset_t ip)969 set_recovery_ip(x86_saved_state64_t  *saved_state, vm_offset_t ip)
970 {
971 	saved_state->isf.rip = ip;
972 }
973 
974 static void
panic_trap(x86_saved_state64_t * regs,uint16_t trapcomment,const char * trapreason,uint32_t pl,kern_return_t fault_result)975 panic_trap(
976 	x86_saved_state64_t    *regs,
977 	uint16_t                trapcomment,
978 	const char             *trapreason,
979 	uint32_t                pl,
980 	kern_return_t           fault_result)
981 {
982 	char            trapbuf[64];
983 	pal_cr_t        cr0, cr2, cr3, cr4;
984 	boolean_t       potential_smep_fault = FALSE, potential_kernel_NX_fault = FALSE;
985 	boolean_t       potential_smap_fault = FALSE;
986 
987 	pal_get_control_registers( &cr0, &cr2, &cr3, &cr4 );
988 	assert(ml_get_interrupts_enabled() == FALSE);
989 	current_cpu_datap()->cpu_fatal_trap_state = regs;
990 	/*
991 	 * Issue an I/O port read if one has been requested - this is an
992 	 * event logic analyzers can use as a trigger point.
993 	 */
994 	panic_notify();
995 
996 	kprintf("CPU %d panic trap number 0x%x, rip 0x%016llx\n",
997 	    cpu_number(), regs->isf.trapno, regs->isf.rip);
998 	kprintf("cr0 0x%016llx cr2 0x%016llx cr3 0x%016llx cr4 0x%016llx\n",
999 	    cr0, cr2, cr3, cr4);
1000 
1001 	if ((regs->isf.trapno == T_PAGE_FAULT) && (regs->isf.err == (T_PF_PROT | T_PF_EXECUTE)) && (regs->isf.rip == regs->cr2)) {
1002 		if (pmap_smep_enabled && (regs->isf.rip < VM_MAX_USER_PAGE_ADDRESS)) {
1003 			potential_smep_fault = TRUE;
1004 		} else if (regs->isf.rip >= VM_MIN_KERNEL_AND_KEXT_ADDRESS) {
1005 			potential_kernel_NX_fault = TRUE;
1006 		}
1007 	} else if (pmap_smap_enabled &&
1008 	    regs->isf.trapno == T_PAGE_FAULT &&
1009 	    regs->isf.err & T_PF_PROT &&
1010 	    regs->cr2 < VM_MAX_USER_PAGE_ADDRESS &&
1011 	    regs->isf.rip >= VM_MIN_KERNEL_AND_KEXT_ADDRESS) {
1012 		potential_smap_fault = TRUE;
1013 	}
1014 
1015 	if (trapreason == NULL) {
1016 		const char *traptype = "Unknown";
1017 
1018 		if (regs->isf.trapno < TRAP_TYPES) {
1019 			traptype = trap_type[regs->isf.trapno];
1020 		}
1021 
1022 		trapreason = "Kernel trap";
1023 
1024 		if (trapcomment == 0) {
1025 			snprintf(trapbuf, sizeof(trapbuf),
1026 			    "type = %d=%s, ",
1027 			    regs->isf.trapno, traptype);
1028 		} else {
1029 			snprintf(trapbuf, sizeof(trapbuf),
1030 			    "type = %d=%s #%#04hx, ",
1031 			    regs->isf.trapno, traptype, trapcomment);
1032 		}
1033 	} else {
1034 		trapbuf[0] = '\0';
1035 	}
1036 
1037 #undef panic
1038 	panic("%s at 0x%016llx, %sregisters:\n"
1039 	    "CR0: 0x%016llx, CR2: 0x%016llx, CR3: 0x%016llx, CR4: 0x%016llx\n"
1040 	    "RAX: 0x%016llx, RBX: 0x%016llx, RCX: 0x%016llx, RDX: 0x%016llx\n"
1041 	    "RSP: 0x%016llx, RBP: 0x%016llx, RSI: 0x%016llx, RDI: 0x%016llx\n"
1042 	    "R8:  0x%016llx, R9:  0x%016llx, R10: 0x%016llx, R11: 0x%016llx\n"
1043 	    "R12: 0x%016llx, R13: 0x%016llx, R14: 0x%016llx, R15: 0x%016llx\n"
1044 	    "RFL: 0x%016llx, RIP: 0x%016llx, CS:  0x%016llx, SS:  0x%016llx\n"
1045 	    "Fault CR2: 0x%016llx, Error code: 0x%016llx, Fault CPU: 0x%x%s%s%s%s, PL: %d, VF: %d\n",
1046 	    trapreason, regs->isf.rip, trapbuf,
1047 	    cr0, cr2, cr3, cr4,
1048 	    regs->rax, regs->rbx, regs->rcx, regs->rdx,
1049 	    regs->isf.rsp, regs->rbp, regs->rsi, regs->rdi,
1050 	    regs->r8, regs->r9, regs->r10, regs->r11,
1051 	    regs->r12, regs->r13, regs->r14, regs->r15,
1052 	    regs->isf.rflags, regs->isf.rip, regs->isf.cs & 0xFFFF,
1053 	    regs->isf.ss & 0xFFFF, regs->cr2, regs->isf.err, regs->isf.cpu,
1054 	    virtualized ? " VMM" : "",
1055 	    potential_kernel_NX_fault ? " Kernel NX fault" : "",
1056 	    potential_smep_fault ? " SMEP/User NX fault" : "",
1057 	    potential_smap_fault ? " SMAP fault" : "",
1058 	    pl,
1059 	    fault_result);
1060 }
1061 
1062 #if CONFIG_DTRACE
1063 extern kern_return_t dtrace_user_probe(x86_saved_state_t *);
1064 #endif
1065 
1066 #if DEBUG
1067 uint32_t fsigs[2];
1068 uint32_t fsigns, fsigcs;
1069 #endif
1070 
1071 /*
1072  *	Trap from user mode.
1073  */
1074 void
user_trap(x86_saved_state_t * saved_state)1075 user_trap(
1076 	x86_saved_state_t *saved_state)
1077 {
1078 	int                     exc;
1079 	int                     err;
1080 	mach_exception_code_t   code;
1081 	mach_exception_subcode_t subcode;
1082 	int                     type;
1083 	user_addr_t             vaddr;
1084 	vm_prot_t               prot;
1085 	thread_t                thread = current_thread();
1086 	kern_return_t           kret;
1087 	user_addr_t             rip;
1088 	unsigned long           dr6 = 0; /* 32 bit for i386, 64 bit for x86_64 */
1089 	int                     current_cpu = cpu_number();
1090 #if DEVELOPMENT || DEBUG
1091 	bool                    inspect_cacheline = false;
1092 	uint32_t                traptrace_index;
1093 #endif
1094 	assert((is_saved_state32(saved_state) && !thread_is_64bit_addr(thread)) ||
1095 	    (is_saved_state64(saved_state) && thread_is_64bit_addr(thread)));
1096 
1097 	if (is_saved_state64(saved_state)) {
1098 		x86_saved_state64_t     *regs;
1099 
1100 		regs = saved_state64(saved_state);
1101 
1102 		/* Record cpu where state was captured */
1103 		regs->isf.cpu = current_cpu;
1104 
1105 		type = regs->isf.trapno;
1106 		err  = (int)regs->isf.err & 0xffff;
1107 		vaddr = (user_addr_t)regs->cr2;
1108 		rip   = (user_addr_t)regs->isf.rip;
1109 #if DEVELOPMENT || DEBUG
1110 		traptrace_index = traptrace_start(type, rip, mach_absolute_time(), regs->rbp);
1111 #endif
1112 	} else {
1113 		x86_saved_state32_t     *regs;
1114 
1115 		regs = saved_state32(saved_state);
1116 
1117 		/* Record cpu where state was captured */
1118 		regs->cpu = current_cpu;
1119 
1120 		type  = regs->trapno;
1121 		err   = regs->err & 0xffff;
1122 		vaddr = (user_addr_t)regs->cr2;
1123 		rip   = (user_addr_t)regs->eip;
1124 #if DEVELOPMENT || DEBUG
1125 		traptrace_index = traptrace_start(type, rip, mach_absolute_time(), regs->ebp);
1126 #endif
1127 	}
1128 
1129 #if DEVELOPMENT || DEBUG
1130 	/*
1131 	 * Copy the cacheline of code into the thread's instruction stream save area
1132 	 * before enabling interrupts (the assumption is that we have not otherwise faulted or
1133 	 * trapped since the original cache line stores).  If the saved code is not valid,
1134 	 * we'll catch it below when we process the copyin() for unhandled faults.
1135 	 */
1136 	if (thread->machine.insn_copy_optout == false &&
1137 	    (type == T_PAGE_FAULT || type == T_INVALID_OPCODE || type == T_GENERAL_PROTECTION)) {
1138 #define CACHELINE_SIZE 64
1139 		THREAD_TO_PCB(thread)->insn_cacheline[CACHELINE_SIZE] = (uint8_t)(rip & (CACHELINE_SIZE - 1));
1140 		bcopy(&cpu_shadowp(current_cpu)->cpu_rtimes[0],
1141 		    &THREAD_TO_PCB(thread)->insn_cacheline[0],
1142 		    sizeof(THREAD_TO_PCB(thread)->insn_cacheline) - 1);
1143 		inspect_cacheline = true;
1144 	}
1145 #endif
1146 
1147 	if (type == T_DEBUG) {
1148 		if (thread->machine.ids) {
1149 			unsigned long clear = 0;
1150 			/* Stash and clear this processor's DR6 value, in the event
1151 			 * this was a debug register match
1152 			 */
1153 			__asm__ volatile ("mov %%db6, %0" : "=r" (dr6));
1154 			__asm__ volatile ("mov %0, %%db6" : : "r" (clear));
1155 		}
1156 		/* [Re]Enable LBRs *BEFORE* enabling interrupts to ensure we hit the right CPU */
1157 		i386_lbr_enable();
1158 	}
1159 
1160 	if (type == T_PAGE_FAULT) {
1161 		thread_reset_pcs_will_fault(thread);
1162 	}
1163 
1164 	pal_sti();
1165 
1166 	KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
1167 	    (MACHDBG_CODE(DBG_MACH_EXCP_UTRAP_x86, type)) | DBG_FUNC_NONE,
1168 	    (unsigned)(vaddr >> 32), (unsigned)vaddr,
1169 	    (unsigned)(rip >> 32), (unsigned)rip, 0);
1170 
1171 	code = 0;
1172 	subcode = 0;
1173 	exc = 0;
1174 
1175 #if CONFIG_DTRACE
1176 	/*
1177 	 * DTrace does not consume all user traps, only INT_3's for now.
1178 	 * Avoid needlessly calling tempDTraceTrapHook here, and let the
1179 	 * INT_3 case handle them.
1180 	 */
1181 #endif
1182 
1183 	DEBUG_KPRINT_SYSCALL_MASK(1,
1184 	    "user_trap: type=0x%x(%s) err=0x%x cr2=%p rip=%p\n",
1185 	    type, trap_type[type], err, (void *)(long) vaddr, (void *)(long) rip);
1186 
1187 	switch (type) {
1188 	case T_DIVIDE_ERROR:
1189 		exc = EXC_ARITHMETIC;
1190 		code = EXC_I386_DIV;
1191 		break;
1192 
1193 	case T_DEBUG:
1194 	{
1195 		pcb_t   pcb;
1196 		/*
1197 		 * Update the PCB with this processor's DR6 value
1198 		 * in the event this was a debug register match.
1199 		 */
1200 		pcb = THREAD_TO_PCB(thread);
1201 		if (pcb->ids) {
1202 			/*
1203 			 * We can get and set the status register
1204 			 * in 32-bit mode even on a 64-bit thread
1205 			 * because the high order bits are not
1206 			 * used on x86_64
1207 			 */
1208 			if (thread_is_64bit_addr(thread)) {
1209 				x86_debug_state64_t *ids = pcb->ids;
1210 				ids->dr6 = dr6;
1211 			} else {         /* 32 bit thread */
1212 				x86_debug_state32_t *ids = pcb->ids;
1213 				ids->dr6 = (uint32_t) dr6;
1214 			}
1215 		}
1216 		exc = EXC_BREAKPOINT;
1217 		code = EXC_I386_SGL;
1218 		break;
1219 	}
1220 	case T_INT3:
1221 #if CONFIG_DTRACE
1222 		if (dtrace_user_probe(saved_state) == KERN_SUCCESS) {
1223 			return; /* If it succeeds, we are done... */
1224 		}
1225 #endif
1226 		exc = EXC_BREAKPOINT;
1227 		code = EXC_I386_BPT;
1228 		break;
1229 
1230 	case T_OVERFLOW:
1231 		exc = EXC_ARITHMETIC;
1232 		code = EXC_I386_INTO;
1233 		break;
1234 
1235 	case T_OUT_OF_BOUNDS:
1236 		exc = EXC_SOFTWARE;
1237 		code = EXC_I386_BOUND;
1238 		break;
1239 
1240 	case T_INVALID_OPCODE:
1241 		if (fpUDflt(rip) == 1) {
1242 			exc = EXC_BAD_INSTRUCTION;
1243 			code = EXC_I386_INVOP;
1244 		}
1245 		break;
1246 
1247 	case T_NO_FPU:
1248 		fpnoextflt();
1249 		break;
1250 
1251 	case T_FPU_FAULT:
1252 		fpextovrflt();
1253 		/*
1254 		 * Raise exception.
1255 		 */
1256 		exc = EXC_BAD_ACCESS;
1257 		code = VM_PROT_READ | VM_PROT_EXECUTE;
1258 		subcode = 0;
1259 		break;
1260 
1261 	case T_INVALID_TSS:     /* invalid TSS == iret with NT flag set */
1262 		exc = EXC_BAD_INSTRUCTION;
1263 		code = EXC_I386_INVTSSFLT;
1264 		subcode = err;
1265 		break;
1266 
1267 	case T_SEGMENT_NOT_PRESENT:
1268 		exc = EXC_BAD_INSTRUCTION;
1269 		code = EXC_I386_SEGNPFLT;
1270 		subcode = err;
1271 		break;
1272 
1273 	case T_STACK_FAULT:
1274 		exc = EXC_BAD_INSTRUCTION;
1275 		code = EXC_I386_STKFLT;
1276 		subcode = err;
1277 		break;
1278 
1279 	case T_GENERAL_PROTECTION:
1280 		/*
1281 		 * There's a wide range of circumstances which generate this
1282 		 * class of exception. From user-space, many involve bad
1283 		 * addresses (such as a non-canonical 64-bit address).
1284 		 * So we map this to EXC_BAD_ACCESS (and thereby SIGSEGV).
1285 		 * The trouble is cr2 doesn't contain the faulting address;
1286 		 * we'd need to decode the faulting instruction to really
1287 		 * determine this. We'll leave that to debuggers.
1288 		 * However, attempted execution of privileged instructions
1289 		 * (e.g. cli) also generate GP faults and so we map these to
1290 		 * to EXC_BAD_ACCESS (and thence SIGSEGV) also - rather than
1291 		 * EXC_BAD_INSTRUCTION which is more accurate. We just can't
1292 		 * win!
1293 		 */
1294 		exc = EXC_BAD_ACCESS;
1295 		code = EXC_I386_GPFLT;
1296 		subcode = err;
1297 		break;
1298 
1299 	case T_PAGE_FAULT:
1300 	{
1301 		prot = VM_PROT_READ;
1302 
1303 		if (err & T_PF_WRITE) {
1304 			prot |= VM_PROT_WRITE;
1305 		}
1306 		if (__improbable(err & T_PF_EXECUTE)) {
1307 			prot |= VM_PROT_EXECUTE;
1308 		}
1309 #if DEVELOPMENT || DEBUG
1310 		bool do_simd_hash = thread_fpsimd_hash_enabled();
1311 		uint32_t fsig = 0;
1312 		fsig = do_simd_hash ? thread_fpsimd_hash(thread) : 0;
1313 #if DEBUG
1314 		fsigs[0] = fsig;
1315 #endif
1316 #endif
1317 		kret = vm_fault(thread->map,
1318 		    vaddr,
1319 		    prot, FALSE, VM_KERN_MEMORY_NONE,
1320 		    THREAD_ABORTSAFE, NULL, 0);
1321 #if DEVELOPMENT || DEBUG
1322 		if (do_simd_hash && fsig) {
1323 			uint32_t fsig2 = thread_fpsimd_hash(thread);
1324 #if DEBUG
1325 			fsigcs++;
1326 			fsigs[1] = fsig2;
1327 #endif
1328 			if (fsig != fsig2) {
1329 				panic("FP/SIMD state hash mismatch across fault thread: %p 0x%x->0x%x", thread, fsig, fsig2);
1330 			}
1331 		} else {
1332 #if DEBUG
1333 			fsigns++;
1334 #endif
1335 		}
1336 #endif
1337 		if (__probable((kret == KERN_SUCCESS) || (kret == KERN_ABORTED))) {
1338 			break;
1339 		} else if (__improbable(kret == KERN_FAILURE)) {
1340 			/*
1341 			 * For a user trap, vm_fault() should never return KERN_FAILURE.
1342 			 * If it does, we're leaking preemption disables somewhere in the kernel.
1343 			 */
1344 			panic("vm_fault() KERN_FAILURE from user fault on thread %p", thread);
1345 		}
1346 
1347 		/* PAL debug hook (empty on x86) */
1348 		pal_dbg_page_fault(thread, vaddr, kret);
1349 		exc = EXC_BAD_ACCESS;
1350 		code = kret;
1351 		subcode = vaddr;
1352 	}
1353 	break;
1354 
1355 	case T_SSE_FLOAT_ERROR:
1356 		fpSSEexterrflt();
1357 		exc = EXC_ARITHMETIC;
1358 		code = EXC_I386_SSEEXTERR;
1359 		subcode = ((struct x86_fx_thread_state *)thread->machine.ifps)->fx_MXCSR;
1360 		break;
1361 
1362 
1363 	case T_FLOATING_POINT_ERROR:
1364 		fpexterrflt();
1365 		exc = EXC_ARITHMETIC;
1366 		code = EXC_I386_EXTERR;
1367 		subcode = ((struct x86_fx_thread_state *)thread->machine.ifps)->fx_status;
1368 		break;
1369 
1370 	case T_DTRACE_RET:
1371 #if CONFIG_DTRACE
1372 		if (dtrace_user_probe(saved_state) == KERN_SUCCESS) {
1373 			return; /* If it succeeds, we are done... */
1374 		}
1375 #endif
1376 		/*
1377 		 * If we get an INT 0x7f when we do not expect to,
1378 		 * treat it as an illegal instruction
1379 		 */
1380 		exc = EXC_BAD_INSTRUCTION;
1381 		code = EXC_I386_INVOP;
1382 		break;
1383 
1384 	default:
1385 		panic("Unexpected user trap, type %d", type);
1386 	}
1387 
1388 	if (type == T_PAGE_FAULT) {
1389 		thread_reset_pcs_done_faulting(thread);
1390 	}
1391 
1392 	if (exc != 0) {
1393 		uint16_t cs;
1394 		boolean_t intrs;
1395 
1396 		if (is_saved_state64(saved_state)) {
1397 			cs = saved_state64(saved_state)->isf.cs;
1398 		} else {
1399 			cs = saved_state32(saved_state)->cs;
1400 		}
1401 
1402 		if (last_branch_enabled_modes == LBR_ENABLED_USERMODE) {
1403 			intrs = ml_set_interrupts_enabled(FALSE);
1404 			/*
1405 			 * This is a bit racy (it's possible for this thread to migrate to another CPU, then
1406 			 * migrate back, but that seems rather rare in practice), but good enough to ensure
1407 			 * the LBRs are saved before proceeding with exception/signal dispatch.
1408 			 */
1409 			if (current_cpu == cpu_number()) {
1410 				i386_lbr_synch(thread);
1411 			}
1412 			ml_set_interrupts_enabled(intrs);
1413 		}
1414 
1415 		/*
1416 		 * Do not try to copyin from the instruction stream if the page fault was due
1417 		 * to an access to rip and was unhandled.
1418 		 * Do not deal with cases when %cs != USER[64]_CS
1419 		 * And of course there's no need to copy the instruction stream if the boot-arg
1420 		 * was set to 0.
1421 		 */
1422 		if (thread->machine.insn_copy_optout == false && insn_copyin_count > 0 &&
1423 		    (cs == USER64_CS || cs == USER_CS) && (type != T_PAGE_FAULT || vaddr != rip)) {
1424 #if DEVELOPMENT || DEBUG
1425 			copy_instruction_stream(thread, rip, type, inspect_cacheline);
1426 #else
1427 			copy_instruction_stream(thread, rip, type);
1428 #endif
1429 		}
1430 
1431 #if DEVELOPMENT || DEBUG
1432 		if (traptrace_index != TRAPTRACE_INVALID_INDEX) {
1433 			traptrace_end(traptrace_index, mach_absolute_time());
1434 		}
1435 #endif
1436 		/*
1437 		 * Note: Codepaths that directly return from user_trap() have pending
1438 		 * ASTs processed in locore
1439 		 */
1440 		i386_exception(exc, code, subcode);
1441 		/* NOTREACHED */
1442 	} else {
1443 #if DEVELOPMENT || DEBUG
1444 		if (traptrace_index != TRAPTRACE_INVALID_INDEX) {
1445 			traptrace_end(traptrace_index, mach_absolute_time());
1446 		}
1447 #endif
1448 	}
1449 }
1450 
1451 /*
1452  * Copyin up to x86_INSTRUCTION_STATE_MAX_INSN_BYTES bytes from the page that includes `rip`,
1453  * ensuring that we stay on the same page, clipping the start or end, as needed.
1454  * Add the clipped amount back at the start or end, depending on where it fits.
1455  * Consult the variable populated by the boot-arg `insn_capcnt'
1456  */
1457 static __attribute__((noinline)) void
copy_instruction_stream(thread_t thread,uint64_t rip,int __unused trap_code,bool inspect_cacheline)1458 copy_instruction_stream(thread_t thread, uint64_t rip, int __unused trap_code
1459 #if DEVELOPMENT || DEBUG
1460     , bool inspect_cacheline
1461 #endif
1462     )
1463 {
1464 #if x86_INSTRUCTION_STATE_MAX_INSN_BYTES > 4096
1465 #error x86_INSTRUCTION_STATE_MAX_INSN_BYTES cannot exceed a page in size.
1466 #endif
1467 	pcb_t pcb = THREAD_TO_PCB(thread);
1468 	vm_map_offset_t pagemask = ~vm_map_page_mask(current_map());
1469 	vm_map_offset_t rip_page = rip & pagemask;
1470 	vm_map_offset_t start_addr;
1471 	vm_map_offset_t insn_offset;
1472 	vm_map_offset_t end_addr = rip + (insn_copyin_count / 2);
1473 	void *stack_buffer;
1474 	int copyin_err = 0;
1475 #if defined(MACH_BSD) && (DEVELOPMENT || DEBUG)
1476 	void *procname;
1477 #endif
1478 
1479 #if DEVELOPMENT || DEBUG
1480 	assert(insn_copyin_count <= x86_INSTRUCTION_STATE_MAX_INSN_BYTES);
1481 #else
1482 	if (insn_copyin_count > x86_INSTRUCTION_STATE_MAX_INSN_BYTES ||
1483 	    insn_copyin_count < 64 /* CACHELINE_SIZE */) {
1484 		return;
1485 	}
1486 #endif
1487 
1488 #pragma clang diagnostic push
1489 #pragma clang diagnostic ignored "-Walloca"
1490 	stack_buffer = __builtin_alloca(insn_copyin_count);
1491 #pragma clang diagnostic pop
1492 
1493 	if (rip >= (insn_copyin_count / 2)) {
1494 		start_addr = rip - (insn_copyin_count / 2);
1495 	} else {
1496 		start_addr = 0;
1497 	}
1498 
1499 	if (start_addr < rip_page) {
1500 		insn_offset = (insn_copyin_count / 2) - (rip_page - start_addr);
1501 		end_addr += (rip_page - start_addr);
1502 		start_addr = rip_page;
1503 	} else if (end_addr >= (rip_page + (~pagemask + 1))) {
1504 		start_addr -= (end_addr - (rip_page + (~pagemask + 1))); /* Adjust start address backward */
1505 		/* Adjust instruction offset due to start address change */
1506 		insn_offset = (insn_copyin_count / 2) + (end_addr - (rip_page + (~pagemask + 1)));
1507 		end_addr = rip_page + (~pagemask + 1);  /* clip to the start of the next page (non-inclusive */
1508 	} else {
1509 		insn_offset = insn_copyin_count / 2;
1510 	}
1511 
1512 	disable_preemption();   /* Prevent copyin from faulting in the instruction stream */
1513 	if (
1514 #if DEVELOPMENT || DEBUG
1515 		(insnstream_force_cacheline_mismatch < 2) &&
1516 #endif
1517 		((end_addr > start_addr) && (copyin_err = copyin(start_addr, stack_buffer, end_addr - start_addr)) == 0)) {
1518 		enable_preemption();
1519 
1520 		if (pcb->insn_state == 0) {
1521 			pcb->insn_state = kalloc_data(sizeof(x86_instruction_state_t), Z_WAITOK);
1522 		}
1523 
1524 		if (pcb->insn_state != 0) {
1525 			bcopy(stack_buffer, pcb->insn_state->insn_bytes, end_addr - start_addr);
1526 			bzero(&pcb->insn_state->insn_bytes[end_addr - start_addr],
1527 			    insn_copyin_count - (end_addr - start_addr));
1528 
1529 			pcb->insn_state->insn_stream_valid_bytes = (int)(end_addr - start_addr);
1530 			pcb->insn_state->insn_offset = (int)insn_offset;
1531 
1532 #if DEVELOPMENT || DEBUG
1533 			/* Now try to validate the cacheline we read at early-fault time matches the code
1534 			 * copied in. Before we do that, we have to make sure the buffer contains a valid
1535 			 * cacheline by looking for the 2 sentinel values written in the event the cacheline
1536 			 * could not be copied.
1537 			 */
1538 #define CACHELINE_DATA_NOT_PRESENT 0xdeadc0debeefcafeULL
1539 #define CACHELINE_MASK (CACHELINE_SIZE - 1)
1540 
1541 			if (inspect_cacheline &&
1542 			    (*(uint64_t *)(uintptr_t)&pcb->insn_cacheline[0] != CACHELINE_DATA_NOT_PRESENT &&
1543 			    *(uint64_t *)(uintptr_t)&pcb->insn_cacheline[8] != CACHELINE_DATA_NOT_PRESENT)) {
1544 				/*
1545 				 * The position of the cacheline in the instruction buffer is at offset
1546 				 * insn_offset - (rip & CACHELINE_MASK)
1547 				 */
1548 				if (__improbable((rip & CACHELINE_MASK) > insn_offset)) {
1549 					printf("thread %p code cacheline @ %p clipped wrt copied-in code (offset %d)\n",
1550 					    thread, (void *)(rip & ~CACHELINE_MASK), (int)(rip & CACHELINE_MASK));
1551 				} else if (bcmp(&pcb->insn_state->insn_bytes[insn_offset - (rip & CACHELINE_MASK)],
1552 				    &pcb->insn_cacheline[0], CACHELINE_SIZE) != 0
1553 				    || insnstream_force_cacheline_mismatch
1554 				    ) {
1555 #if x86_INSTRUCTION_STATE_CACHELINE_SIZE != CACHELINE_SIZE
1556 #error cacheline size mismatch
1557 #endif
1558 					bcopy(&pcb->insn_cacheline[0], &pcb->insn_state->insn_cacheline[0],
1559 					    x86_INSTRUCTION_STATE_CACHELINE_SIZE);
1560 					/* Mark the instruction stream as being out-of-synch */
1561 					pcb->insn_state->out_of_synch = 1;
1562 
1563 					printf("thread %p code cacheline @ %p mismatches with copied-in code [trap 0x%x]\n",
1564 					    thread, (void *)(rip & ~CACHELINE_MASK), trap_code);
1565 					for (int i = 0; i < 8; i++) {
1566 						printf("\t[%d] cl=0x%08llx vs. ci=0x%08llx\n", i, *(uint64_t *)(uintptr_t)&pcb->insn_cacheline[i * 8],
1567 						    *(uint64_t *)(uintptr_t)&pcb->insn_state->insn_bytes[(i * 8) + insn_offset - (rip & CACHELINE_MASK)]);
1568 					}
1569 					if (panic_on_cacheline_mismatch) {
1570 						panic("Cacheline mismatch while processing unhandled exception.");
1571 					}
1572 				} else {
1573 					pcb->insn_state->out_of_synch = 0;
1574 				}
1575 			} else if (inspect_cacheline) {
1576 				printf("thread %p could not capture code cacheline at fault IP %p [offset %d]\n",
1577 				    (void *)thread, (void *)rip, (int)(insn_offset - (rip & CACHELINE_MASK)));
1578 				pcb->insn_state->out_of_synch = 0;
1579 			}
1580 #else
1581 			pcb->insn_state->out_of_synch = 0;
1582 #endif /* DEVELOPMENT || DEBUG */
1583 
1584 #if defined(MACH_BSD) && (DEVELOPMENT || DEBUG)
1585 			if (panic_on_trap_procname[0] != 0) {
1586 				task_t task = get_threadtask(thread);
1587 				char procnamebuf[65] = {0};
1588 
1589 				if (get_bsdtask_info(task) != NULL) {
1590 					procname = proc_name_address(get_bsdtask_info(task));
1591 					strlcpy(procnamebuf, procname, sizeof(procnamebuf));
1592 
1593 					if (strcasecmp(panic_on_trap_procname, procnamebuf) == 0 &&
1594 					    ((1U << trap_code) & panic_on_trap_mask) != 0) {
1595 						panic("Panic requested on trap type 0x%x for process `%s'", trap_code,
1596 						    panic_on_trap_procname);
1597 						/*NORETURN*/
1598 					}
1599 				}
1600 			}
1601 #endif /* MACH_BSD && (DEVELOPMENT || DEBUG) */
1602 		}
1603 	} else {
1604 		enable_preemption();
1605 
1606 		pcb->insn_state_copyin_failure_errorcode = copyin_err;
1607 #if DEVELOPMENT || DEBUG
1608 		if (inspect_cacheline && pcb->insn_state == 0) {
1609 			pcb->insn_state = kalloc_data(sizeof(x86_instruction_state_t), Z_WAITOK);
1610 		}
1611 		if (pcb->insn_state != 0) {
1612 			pcb->insn_state->insn_stream_valid_bytes = 0;
1613 			pcb->insn_state->insn_offset = 0;
1614 
1615 			if (inspect_cacheline &&
1616 			    (*(uint64_t *)(uintptr_t)&pcb->insn_cacheline[0] != CACHELINE_DATA_NOT_PRESENT &&
1617 			    *(uint64_t *)(uintptr_t)&pcb->insn_cacheline[8] != CACHELINE_DATA_NOT_PRESENT)) {
1618 				/*
1619 				 * We can still copy the cacheline into the instruction state structure
1620 				 * if it contains valid data
1621 				 */
1622 				pcb->insn_state->out_of_synch = 1;
1623 				bcopy(&pcb->insn_cacheline[0], &pcb->insn_state->insn_cacheline[0],
1624 				    x86_INSTRUCTION_STATE_CACHELINE_SIZE);
1625 			}
1626 		}
1627 #endif /* DEVELOPMENT || DEBUG */
1628 	}
1629 }
1630 
1631 /*
1632  * Handle exceptions for i386.
1633  *
1634  * If we are an AT bus machine, we must turn off the AST for a
1635  * delayed floating-point exception.
1636  *
1637  * If we are providing floating-point emulation, we may have
1638  * to retrieve the real register values from the floating point
1639  * emulator.
1640  */
1641 void
i386_exception(int exc,mach_exception_code_t code,mach_exception_subcode_t subcode)1642 i386_exception(
1643 	int     exc,
1644 	mach_exception_code_t code,
1645 	mach_exception_subcode_t subcode)
1646 {
1647 	mach_exception_data_type_t   codes[EXCEPTION_CODE_MAX];
1648 
1649 	DEBUG_KPRINT_SYSCALL_MACH("i386_exception: exc=%d code=0x%llx subcode=0x%llx\n",
1650 	    exc, code, subcode);
1651 	codes[0] = code;                /* new exception interface */
1652 	codes[1] = subcode;
1653 	exception_triage(exc, codes, 2);
1654 	/*NOTREACHED*/
1655 }
1656 
1657 
1658 /* Synchronize a thread's x86_kernel_state (if any) with the given
1659  * x86_saved_state_t obtained from the trap/IPI handler; called in
1660  * kernel_trap() prior to entering the debugger, and when receiving
1661  * an "MP_KDP" IPI. Called with null saved_state if an incoming IPI
1662  * was detected from the kernel while spinning with interrupts masked.
1663  */
1664 
1665 void
sync_iss_to_iks(x86_saved_state_t * saved_state)1666 sync_iss_to_iks(x86_saved_state_t *saved_state)
1667 {
1668 	struct x86_kernel_state *iks = NULL;
1669 	vm_offset_t kstack;
1670 	boolean_t record_active_regs = FALSE;
1671 
1672 	/* The PAL may have a special way to sync registers */
1673 	if (saved_state && saved_state->flavor == THREAD_STATE_NONE) {
1674 		pal_get_kern_regs( saved_state );
1675 	}
1676 
1677 	if (current_thread() != NULL &&
1678 	    (kstack = current_thread()->kernel_stack) != 0) {
1679 		x86_saved_state64_t     *regs = saved_state64(saved_state);
1680 
1681 		iks = STACK_IKS(kstack);
1682 
1683 		/* Did we take the trap/interrupt in kernel mode? */
1684 		if (saved_state == NULL || /* NULL => polling in kernel */
1685 		    regs == USER_REGS64(current_thread())) {
1686 			record_active_regs = TRUE;
1687 		} else {
1688 			iks->k_rbx = regs->rbx;
1689 			iks->k_rsp = regs->isf.rsp;
1690 			iks->k_rbp = regs->rbp;
1691 			iks->k_r12 = regs->r12;
1692 			iks->k_r13 = regs->r13;
1693 			iks->k_r14 = regs->r14;
1694 			iks->k_r15 = regs->r15;
1695 			iks->k_rip = regs->isf.rip;
1696 		}
1697 	}
1698 
1699 	if (record_active_regs == TRUE) {
1700 		/* Show the trap handler path */
1701 		__asm__ volatile ("movq %%rbx, %0" : "=m" (iks->k_rbx));
1702 		__asm__ volatile ("movq %%rsp, %0" : "=m" (iks->k_rsp));
1703 		__asm__ volatile ("movq %%rbp, %0" : "=m" (iks->k_rbp));
1704 		__asm__ volatile ("movq %%r12, %0" : "=m" (iks->k_r12));
1705 		__asm__ volatile ("movq %%r13, %0" : "=m" (iks->k_r13));
1706 		__asm__ volatile ("movq %%r14, %0" : "=m" (iks->k_r14));
1707 		__asm__ volatile ("movq %%r15, %0" : "=m" (iks->k_r15));
1708 		/* "Current" instruction pointer */
1709 		__asm__ volatile ("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
1710                                   : "=m" (iks->k_rip)
1711                                   :
1712                                   : "rax");
1713 	}
1714 }
1715 
1716 /*
1717  * This is used by the NMI interrupt handler (from mp.c) to
1718  * uncondtionally sync the trap handler context to the IKS
1719  * irrespective of whether the NMI was fielded in kernel
1720  * or user space.
1721  */
1722 void
sync_iss_to_iks_unconditionally(__unused x86_saved_state_t * saved_state)1723 sync_iss_to_iks_unconditionally(__unused x86_saved_state_t *saved_state)
1724 {
1725 	struct x86_kernel_state *iks;
1726 	vm_offset_t kstack;
1727 
1728 	if ((kstack = current_thread()->kernel_stack) != 0) {
1729 		iks = STACK_IKS(kstack);
1730 		/* Display the trap handler path */
1731 		__asm__ volatile ("movq %%rbx, %0" : "=m" (iks->k_rbx));
1732 		__asm__ volatile ("movq %%rsp, %0" : "=m" (iks->k_rsp));
1733 		__asm__ volatile ("movq %%rbp, %0" : "=m" (iks->k_rbp));
1734 		__asm__ volatile ("movq %%r12, %0" : "=m" (iks->k_r12));
1735 		__asm__ volatile ("movq %%r13, %0" : "=m" (iks->k_r13));
1736 		__asm__ volatile ("movq %%r14, %0" : "=m" (iks->k_r14));
1737 		__asm__ volatile ("movq %%r15, %0" : "=m" (iks->k_r15));
1738 		/* "Current" instruction pointer */
1739 		__asm__ volatile ("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:" : "=m" (iks->k_rip)::"rax");
1740 	}
1741 }
1742 
1743 #if DEBUG
1744 #define TERI 1
1745 #endif
1746 
1747 #if TERI
1748 extern void     thread_exception_return_internal(void) __dead2;
1749 
1750 void
thread_exception_return(void)1751 thread_exception_return(void)
1752 {
1753 	thread_t thread = current_thread();
1754 	task_t   task   = current_task();
1755 
1756 	ml_set_interrupts_enabled(FALSE);
1757 	if (thread_is_64bit_addr(thread) != task_has_64Bit_addr(task)) {
1758 		panic("Task/thread bitness mismatch %p %p, task: %d, thread: %d",
1759 		    thread, task, thread_is_64bit_addr(thread), task_has_64Bit_addr(task));
1760 	}
1761 
1762 	if (thread_is_64bit_addr(thread)) {
1763 		if ((gdt_desc_p(USER64_CS)->access & ACC_PL_U) == 0) {
1764 			panic("64-GDT mismatch %p, descriptor: %p", thread, gdt_desc_p(USER64_CS));
1765 		}
1766 	} else {
1767 		if ((gdt_desc_p(USER_CS)->access & ACC_PL_U) == 0) {
1768 			panic("32-GDT mismatch %p, descriptor: %p", thread, gdt_desc_p(USER_CS));
1769 		}
1770 	}
1771 	assert(get_preemption_level() == 0);
1772 	thread_exception_return_internal();
1773 }
1774 #endif
1775 
1776 #if DEVELOPMENT || DEBUG
1777 static int trap_handled;
1778 
1779 static const char *
handle_recoverable_kernel_trap(__unused void * tstate,uint16_t comment)1780 handle_recoverable_kernel_trap(
1781 	__unused void     *tstate,
1782 	uint16_t          comment)
1783 {
1784 	assert(comment == TEST_RECOVERABLE_SOFT_TRAP);
1785 
1786 	printf("Recoverable trap handled.\n");
1787 	trap_handled = 1;
1788 
1789 	return NULL;
1790 }
1791 
1792 KERNEL_BRK_DESCRIPTOR_DEFINE(test_desc,
1793     .type                = TRAP_TELEMETRY_TYPE_KERNEL_BRK_TEST,
1794     .base                = TEST_RECOVERABLE_SOFT_TRAP,
1795     .max                 = TEST_RECOVERABLE_SOFT_TRAP,
1796     .options             = BRK_TELEMETRY_OPTIONS_RECOVERABLE_DEFAULT(
1797 	    /* enable_telemetry */ false),
1798     .handle_breakpoint   = handle_recoverable_kernel_trap);
1799 
1800 static int
recoverable_kernel_trap_test(__unused int64_t in,int64_t * out)1801 recoverable_kernel_trap_test(__unused int64_t in, int64_t *out)
1802 {
1803 	ml_recoverable_trap(TEST_RECOVERABLE_SOFT_TRAP);
1804 
1805 	*out = trap_handled;
1806 	return 0;
1807 }
1808 
1809 SYSCTL_TEST_REGISTER(recoverable_kernel_trap, recoverable_kernel_trap_test);
1810 #endif
1811