2 * Copyright (c) 2005-2006 Apple Computer, Inc. All rights reserved.
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
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.
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
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.
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
31 * APPLE NOTE: This file is compiled even if dtrace is unconfig'd. A symbol
32 * from this file (_dtrace_register_anon_DOF) always needs to be exported for
33 * an external kext to link against.
38 #define MACH__POSIX_C_SOURCE_PRIVATE 1 /* pulls in suitable savearea from mach/ppc/thread_status.h */
39 #include <kern/thread.h>
40 #include <mach/thread_status.h>
44 #include <sys/malloc.h>
47 #include <sys/proc_internal.h>
48 #include <sys/kauth.h>
50 #include <sys/systm.h>
51 #include <sys/dtrace.h>
52 #include <sys/dtrace_impl.h>
53 #include <libkern/OSAtomic.h>
54 #include <kern/kern_types.h>
55 #include <kern/timer_call.h>
56 #include <kern/thread_call.h>
57 #include <kern/task.h>
58 #include <kern/sched_prim.h>
59 #include <kern/queue.h>
60 #include <miscfs/devfs/devfs.h>
61 #include <kern/kalloc.h>
63 #include <mach/vm_param.h>
64 #include <mach/mach_vm.h>
65 #include <mach/task.h>
67 #include <vm/vm_map.h> /* All the bits we care about are guarded by MACH_KERNEL_PRIVATE :-( */
69 /* missing prototypes, not exported by Mach */
70 extern kern_return_t
task_suspend_internal(task_t
);
71 extern kern_return_t
task_resume_internal(task_t
);
76 /* Solaris proc_t is the struct. Darwin's proc_t is a pointer to it. */
77 #define proc_t struct proc /* Steer clear of the Darwin typedef for proc_t */
79 /* Not called from probe context */
85 if ((p
= proc_find(pid
)) == PROC_NULL
) {
89 task_suspend_internal(p
->task
);
93 lck_mtx_lock(&p
->p_dtrace_sprlock
);
98 /* Not called from probe context */
102 if (p
!= PROC_NULL
) {
103 lck_mtx_unlock(&p
->p_dtrace_sprlock
);
107 task_resume_internal(p
->task
);
117 // These are not exported from vm_map.h.
118 extern kern_return_t
vm_map_read_user(vm_map_t map
, vm_map_address_t src_addr
, void *dst_p
, vm_size_t size
);
119 extern kern_return_t
vm_map_write_user(vm_map_t map
, void *src_p
, vm_map_address_t dst_addr
, vm_size_t size
);
121 /* Not called from probe context */
123 uread(proc_t
*p
, void *buf
, user_size_t len
, user_addr_t a
)
127 ASSERT(p
!= PROC_NULL
);
128 ASSERT(p
->task
!= NULL
);
130 task_t task
= p
->task
;
133 * Grab a reference to the task vm_map_t to make sure
134 * the map isn't pulled out from under us.
136 * Because the proc_lock is not held at all times on all code
137 * paths leading here, it is possible for the proc to have
138 * exited. If the map is null, fail.
140 vm_map_t map
= get_task_map_reference(task
);
142 ret
= vm_map_read_user( map
, (vm_map_address_t
)a
, buf
, (vm_size_t
)len
);
143 vm_map_deallocate(map
);
145 ret
= KERN_TERMINATED
;
151 /* Not called from probe context */
153 uwrite(proc_t
*p
, void *buf
, user_size_t len
, user_addr_t a
)
158 ASSERT(p
->task
!= NULL
);
160 task_t task
= p
->task
;
163 * Grab a reference to the task vm_map_t to make sure
164 * the map isn't pulled out from under us.
166 * Because the proc_lock is not held at all times on all code
167 * paths leading here, it is possible for the proc to have
168 * exited. If the map is null, fail.
170 vm_map_t map
= get_task_map_reference(task
);
172 /* Find the memory permissions. */
173 uint32_t nestingDepth
=999999;
174 vm_region_submap_short_info_data_64_t info
;
175 mach_msg_type_number_t count
= VM_REGION_SUBMAP_SHORT_INFO_COUNT_64
;
176 mach_vm_address_t address
= (mach_vm_address_t
)a
;
177 mach_vm_size_t sizeOfRegion
= (mach_vm_size_t
)len
;
179 ret
= mach_vm_region_recurse(map
, &address
, &sizeOfRegion
, &nestingDepth
, (vm_region_recurse_info_t
)&info
, &count
);
180 if (ret
!= KERN_SUCCESS
)
185 if (!(info
.protection
& VM_PROT_WRITE
)) {
186 /* Save the original protection values for restoration later */
187 reprotect
= info
.protection
;
189 if (info
.max_protection
& VM_PROT_WRITE
) {
190 /* The memory is not currently writable, but can be made writable. */
191 ret
= mach_vm_protect (map
, (mach_vm_offset_t
)a
, (mach_vm_size_t
)len
, 0, reprotect
| VM_PROT_WRITE
);
194 * The memory is not currently writable, and cannot be made writable. We need to COW this memory.
196 * Strange, we can't just say "reprotect | VM_PROT_COPY", that fails.
198 ret
= mach_vm_protect (map
, (mach_vm_offset_t
)a
, (mach_vm_size_t
)len
, 0, VM_PROT_COPY
| VM_PROT_READ
| VM_PROT_WRITE
);
201 if (ret
!= KERN_SUCCESS
)
205 /* The memory was already writable. */
206 reprotect
= VM_PROT_NONE
;
209 ret
= vm_map_write_user( map
,
214 if (ret
!= KERN_SUCCESS
)
217 if (reprotect
!= VM_PROT_NONE
) {
218 ASSERT(reprotect
& VM_PROT_EXECUTE
);
219 ret
= mach_vm_protect (map
, (mach_vm_offset_t
)a
, (mach_vm_size_t
)len
, 0, reprotect
);
223 vm_map_deallocate(map
);
225 ret
= KERN_TERMINATED
;
237 dtrace_cpu_t
*cpu_list
;
238 cpu_core_t
*cpu_core
; /* XXX TLB lockdown? */
245 * dtrace_CRED() can be called from probe context. We cannot simply call kauth_cred_get() since
246 * that function may try to resolve a lazy credential binding, which entails taking the proc_lock.
251 struct uthread
*uthread
= get_bsdthread_info(current_thread());
256 return uthread
->uu_ucred
; /* May return NOCRED which is defined to be 0 */
259 #define HAS_ALLPRIVS(cr) priv_isfullset(&CR_OEPRIV(cr))
260 #define HAS_PRIVILEGE(cr, pr) ((pr) == PRIV_ALL ? \
262 PRIV_ISASSERT(&CR_OEPRIV(cr), pr))
264 int PRIV_POLICY_CHOICE(void* cred
, int priv
, int all
)
266 #pragma unused(priv, all)
267 return kauth_cred_issuser(cred
); /* XXX TODO: How is this different from PRIV_POLICY_ONLY? */
271 PRIV_POLICY_ONLY(void *cr
, int priv
, int boolean
)
273 #pragma unused(priv, boolean)
274 return kauth_cred_issuser(cr
); /* XXX TODO: HAS_PRIVILEGE(cr, priv); */
277 /* XXX Get around const poisoning using structure assigns */
279 crgetgid(const cred_t
*cr
) { cred_t copy_cr
= *cr
; return kauth_cred_getgid(©_cr
); }
282 crgetuid(const cred_t
*cr
) { cred_t copy_cr
= *cr
; return kauth_cred_getuid(©_cr
); }
288 typedef struct wrap_timer_call
{
289 /* node attributes */
295 struct timer_call call
;
297 /* next item in the linked list */
298 LIST_ENTRY(wrap_timer_call
) entries
;
301 #define WAKEUP_REAPER 0x7FFFFFFFFFFFFFFFLL
302 #define NEARLY_FOREVER 0x7FFFFFFFFFFFFFFELL
304 /* CPU going online/offline notifications */
305 void (*dtrace_cpu_state_changed_hook
)(int, boolean_t
) = NULL
;
306 void dtrace_cpu_state_changed(int, boolean_t
);
309 dtrace_install_cpu_hooks(void) {
310 dtrace_cpu_state_changed_hook
= dtrace_cpu_state_changed
;
314 dtrace_cpu_state_changed(int cpuid
, boolean_t is_running
) {
315 #pragma unused(cpuid)
316 wrap_timer_call_t
*wrapTC
= NULL
;
317 boolean_t suspend
= (is_running
? FALSE
: TRUE
);
320 /* Ensure that we're not going to leave the CPU */
321 s
= dtrace_interrupt_disable();
322 assert(cpuid
== cpu_number());
324 LIST_FOREACH(wrapTC
, &(cpu_list
[cpu_number()].cpu_cyc_list
), entries
) {
325 assert(wrapTC
->cpuid
== cpu_number());
327 assert(!wrapTC
->suspended
);
328 /* If this fails, we'll panic anyway, so let's do this now. */
329 if (!timer_call_cancel(&wrapTC
->call
))
330 panic("timer_call_set_suspend() failed to cancel a timer call");
331 wrapTC
->suspended
= TRUE
;
333 /* Rearm the timer, but ensure it was suspended first. */
334 assert(wrapTC
->suspended
);
335 clock_deadline_for_periodic_event(wrapTC
->when
.cyt_interval
, mach_absolute_time(),
337 timer_call_enter1(&wrapTC
->call
, (void*) wrapTC
, wrapTC
->deadline
,
338 TIMER_CALL_SYS_CRITICAL
| TIMER_CALL_LOCAL
);
339 wrapTC
->suspended
= FALSE
;
344 /* Restore the previous interrupt state. */
345 dtrace_interrupt_enable(s
);
349 _timer_call_apply_cyclic( void *ignore
, void *vTChdl
)
351 #pragma unused(ignore)
352 wrap_timer_call_t
*wrapTC
= (wrap_timer_call_t
*)vTChdl
;
354 (*(wrapTC
->hdlr
.cyh_func
))( wrapTC
->hdlr
.cyh_arg
);
356 clock_deadline_for_periodic_event( wrapTC
->when
.cyt_interval
, mach_absolute_time(), &(wrapTC
->deadline
) );
357 timer_call_enter1( &(wrapTC
->call
), (void *)wrapTC
, wrapTC
->deadline
, TIMER_CALL_SYS_CRITICAL
| TIMER_CALL_LOCAL
);
361 timer_call_add_cyclic(wrap_timer_call_t
*wrapTC
, cyc_handler_t
*handler
, cyc_time_t
*when
)
366 timer_call_setup( &(wrapTC
->call
), _timer_call_apply_cyclic
, NULL
);
367 wrapTC
->hdlr
= *handler
;
368 wrapTC
->when
= *when
;
370 nanoseconds_to_absolutetime( wrapTC
->when
.cyt_interval
, (uint64_t *)&wrapTC
->when
.cyt_interval
);
372 now
= mach_absolute_time();
373 wrapTC
->deadline
= now
;
375 clock_deadline_for_periodic_event( wrapTC
->when
.cyt_interval
, now
, &(wrapTC
->deadline
) );
377 /* Insert the timer to the list of the running timers on this CPU, and start it. */
378 s
= dtrace_interrupt_disable();
379 wrapTC
->cpuid
= cpu_number();
380 LIST_INSERT_HEAD(&cpu_list
[wrapTC
->cpuid
].cpu_cyc_list
, wrapTC
, entries
);
381 timer_call_enter1(&wrapTC
->call
, (void*) wrapTC
, wrapTC
->deadline
,
382 TIMER_CALL_SYS_CRITICAL
| TIMER_CALL_LOCAL
);
383 wrapTC
->suspended
= FALSE
;
384 dtrace_interrupt_enable(s
);
386 return (cyclic_id_t
)wrapTC
;
390 * Executed on the CPU the timer is running on.
393 timer_call_remove_cyclic(cyclic_id_t cyclic
)
395 wrap_timer_call_t
*wrapTC
= (wrap_timer_call_t
*)cyclic
;
398 assert(cpu_number() == wrapTC
->cpuid
);
400 if (!timer_call_cancel(&wrapTC
->call
))
401 panic("timer_call_remove_cyclic() failed to cancel a timer call");
403 LIST_REMOVE(wrapTC
, entries
);
407 timer_call_get_cyclic_arg(cyclic_id_t cyclic
)
409 wrap_timer_call_t
*wrapTC
= (wrap_timer_call_t
*)cyclic
;
411 return (wrapTC
? wrapTC
->hdlr
.cyh_arg
: NULL
);
415 cyclic_timer_add(cyc_handler_t
*handler
, cyc_time_t
*when
)
417 wrap_timer_call_t
*wrapTC
= _MALLOC(sizeof(wrap_timer_call_t
), M_TEMP
, M_ZERO
| M_WAITOK
);
421 return timer_call_add_cyclic( wrapTC
, handler
, when
);
425 cyclic_timer_remove(cyclic_id_t cyclic
)
427 ASSERT( cyclic
!= CYCLIC_NONE
);
429 /* Removing a timer call must be done on the CPU the timer is running on. */
430 wrap_timer_call_t
*wrapTC
= (wrap_timer_call_t
*) cyclic
;
431 dtrace_xcall(wrapTC
->cpuid
, (dtrace_xcall_t
) timer_call_remove_cyclic
, (void*) cyclic
);
433 _FREE((void *)cyclic
, M_TEMP
);
437 _cyclic_add_omni(cyclic_id_list_t cyc_list
)
441 wrap_timer_call_t
*wrapTC
;
442 cyc_omni_handler_t
*omni
= (cyc_omni_handler_t
*)cyc_list
;
445 (omni
->cyo_online
)(omni
->cyo_arg
, CPU
, &cH
, &cT
);
447 t
= (char *)cyc_list
;
448 t
+= sizeof(cyc_omni_handler_t
);
449 cyc_list
= (cyclic_id_list_t
)(uintptr_t)t
;
451 t
+= sizeof(cyclic_id_t
)*NCPU
;
452 t
+= (sizeof(wrap_timer_call_t
))*cpu_number();
453 wrapTC
= (wrap_timer_call_t
*)(uintptr_t)t
;
455 cyc_list
[cpu_number()] = timer_call_add_cyclic(wrapTC
, &cH
, &cT
);
459 cyclic_add_omni(cyc_omni_handler_t
*omni
)
461 cyclic_id_list_t cyc_list
=
462 _MALLOC( (sizeof(wrap_timer_call_t
))*NCPU
+
463 sizeof(cyclic_id_t
)*NCPU
+
464 sizeof(cyc_omni_handler_t
), M_TEMP
, M_ZERO
| M_WAITOK
);
465 if (NULL
== cyc_list
)
466 return (cyclic_id_list_t
)CYCLIC_NONE
;
468 *(cyc_omni_handler_t
*)cyc_list
= *omni
;
469 dtrace_xcall(DTRACE_CPUALL
, (dtrace_xcall_t
)_cyclic_add_omni
, (void *)cyc_list
);
475 _cyclic_remove_omni(cyclic_id_list_t cyc_list
)
477 cyc_omni_handler_t
*omni
= (cyc_omni_handler_t
*)cyc_list
;
482 t
= (char *)cyc_list
;
483 t
+= sizeof(cyc_omni_handler_t
);
484 cyc_list
= (cyclic_id_list_t
)(uintptr_t)t
;
487 * If the processor was offline when dtrace started, we did not allocate
488 * a cyclic timer for this CPU.
490 if ((cid
= cyc_list
[cpu_number()]) != CYCLIC_NONE
) {
491 oarg
= timer_call_get_cyclic_arg(cid
);
492 timer_call_remove_cyclic(cid
);
493 (omni
->cyo_offline
)(omni
->cyo_arg
, CPU
, oarg
);
498 cyclic_remove_omni(cyclic_id_list_t cyc_list
)
500 ASSERT( cyc_list
!= (cyclic_id_list_t
)CYCLIC_NONE
);
502 dtrace_xcall(DTRACE_CPUALL
, (dtrace_xcall_t
)_cyclic_remove_omni
, (void *)cyc_list
);
503 _FREE(cyc_list
, M_TEMP
);
506 typedef struct wrap_thread_call
{
511 } wrap_thread_call_t
;
514 * _cyclic_apply will run on some thread under kernel_task. That's OK for the
515 * cleaner and the deadman, but too distant in time and place for the profile provider.
518 _cyclic_apply( void *ignore
, void *vTChdl
)
520 #pragma unused(ignore)
521 wrap_thread_call_t
*wrapTC
= (wrap_thread_call_t
*)vTChdl
;
523 (*(wrapTC
->hdlr
.cyh_func
))( wrapTC
->hdlr
.cyh_arg
);
525 clock_deadline_for_periodic_event( wrapTC
->when
.cyt_interval
, mach_absolute_time(), &(wrapTC
->deadline
) );
526 (void)thread_call_enter1_delayed( wrapTC
->TChdl
, (void *)wrapTC
, wrapTC
->deadline
);
528 /* Did cyclic_remove request a wakeup call when this thread call was re-armed? */
529 if (wrapTC
->when
.cyt_interval
== WAKEUP_REAPER
)
530 thread_wakeup((event_t
)wrapTC
);
534 cyclic_add(cyc_handler_t
*handler
, cyc_time_t
*when
)
538 wrap_thread_call_t
*wrapTC
= _MALLOC(sizeof(wrap_thread_call_t
), M_TEMP
, M_ZERO
| M_WAITOK
);
542 wrapTC
->TChdl
= thread_call_allocate( _cyclic_apply
, NULL
);
543 wrapTC
->hdlr
= *handler
;
544 wrapTC
->when
= *when
;
546 ASSERT(when
->cyt_when
== 0);
547 ASSERT(when
->cyt_interval
< WAKEUP_REAPER
);
549 nanoseconds_to_absolutetime(wrapTC
->when
.cyt_interval
, (uint64_t *)&wrapTC
->when
.cyt_interval
);
551 now
= mach_absolute_time();
552 wrapTC
->deadline
= now
;
554 clock_deadline_for_periodic_event( wrapTC
->when
.cyt_interval
, now
, &(wrapTC
->deadline
) );
555 (void)thread_call_enter1_delayed( wrapTC
->TChdl
, (void *)wrapTC
, wrapTC
->deadline
);
557 return (cyclic_id_t
)wrapTC
;
561 noop_cyh_func(void * ignore
)
563 #pragma unused(ignore)
567 cyclic_remove(cyclic_id_t cyclic
)
569 wrap_thread_call_t
*wrapTC
= (wrap_thread_call_t
*)cyclic
;
571 ASSERT(cyclic
!= CYCLIC_NONE
);
573 while (!thread_call_cancel(wrapTC
->TChdl
)) {
574 int ret
= assert_wait(wrapTC
, THREAD_UNINT
);
575 ASSERT(ret
== THREAD_WAITING
);
577 wrapTC
->when
.cyt_interval
= WAKEUP_REAPER
;
579 ret
= thread_block(THREAD_CONTINUE_NULL
);
580 ASSERT(ret
== THREAD_AWAKENED
);
583 if (thread_call_free(wrapTC
->TChdl
))
584 _FREE(wrapTC
, M_TEMP
);
586 /* Gut this cyclic and move on ... */
587 wrapTC
->hdlr
.cyh_func
= noop_cyh_func
;
588 wrapTC
->when
.cyt_interval
= NEARLY_FOREVER
;
593 * timeout / untimeout (converted to dtrace_timeout / dtrace_untimeout due to name collision)
597 dtrace_timeout(void (*func
)(void *, void *), void* arg
, uint64_t nanos
)
600 thread_call_t call
= thread_call_allocate(func
, NULL
);
602 nanoseconds_to_absolutetime(nanos
, &nanos
);
605 * This method does not use clock_deadline_for_periodic_event() because it is a one-shot,
606 * and clock drift on later invocations is not a worry.
608 uint64_t deadline
= mach_absolute_time() + nanos
;
609 /* DRK: consider using a lower priority callout here */
610 thread_call_enter_delayed(call
, deadline
);
619 ddi_report_dev(dev_info_t
*devi
)
624 #define NSOFT_STATES 32 /* XXX No more than 32 clients at a time, please. */
625 static void *soft
[NSOFT_STATES
];
628 ddi_soft_state_init(void **state_p
, size_t size
, size_t n_items
)
630 #pragma unused(n_items)
633 for (i
= 0; i
< NSOFT_STATES
; ++i
) soft
[i
] = _MALLOC(size
, M_TEMP
, M_ZERO
| M_WAITOK
);
634 *(size_t *)state_p
= size
;
639 ddi_soft_state_zalloc(void *state
, int item
)
641 #pragma unused(state)
642 if (item
< NSOFT_STATES
)
649 ddi_get_soft_state(void *state
, int item
)
651 #pragma unused(state)
652 ASSERT(item
< NSOFT_STATES
);
657 ddi_soft_state_free(void *state
, int item
)
659 ASSERT(item
< NSOFT_STATES
);
660 bzero( soft
[item
], (size_t)state
);
665 ddi_soft_state_fini(void **state_p
)
667 #pragma unused(state_p)
670 for (i
= 0; i
< NSOFT_STATES
; ++i
) _FREE( soft
[i
], M_TEMP
);
673 static unsigned int gRegisteredProps
= 0;
675 char name
[32]; /* enough for "dof-data-" + digits */
680 kern_return_t
_dtrace_register_anon_DOF(char *, uchar_t
*, uint_t
);
683 _dtrace_register_anon_DOF(char *name
, uchar_t
*data
, uint_t nelements
)
685 if (gRegisteredProps
< sizeof(gPropTable
)/sizeof(gPropTable
[0])) {
686 int *p
= (int *)_MALLOC(nelements
*sizeof(int), M_TEMP
, M_WAITOK
);
691 strlcpy(gPropTable
[gRegisteredProps
].name
, name
, sizeof(gPropTable
[0].name
));
692 gPropTable
[gRegisteredProps
].nelements
= nelements
;
693 gPropTable
[gRegisteredProps
].data
= p
;
695 while (nelements
-- > 0) {
696 *p
++ = (int)(*data
++);
707 ddi_prop_lookup_int_array(dev_t match_dev
, dev_info_t
*dip
, uint_t flags
,
708 const char *name
, int **data
, uint_t
*nelements
)
710 #pragma unused(match_dev,dip,flags)
712 for (i
= 0; i
< gRegisteredProps
; ++i
)
714 if (0 == strncmp(name
, gPropTable
[i
].name
,
715 sizeof(gPropTable
[i
].name
))) {
716 *data
= gPropTable
[i
].data
;
717 *nelements
= gPropTable
[i
].nelements
;
725 ddi_prop_free(void *buf
)
732 ddi_driver_major(dev_info_t
*devi
) { return (int)major(CAST_DOWN_EXPLICIT(int,devi
)); }
735 ddi_create_minor_node(dev_info_t
*dip
, const char *name
, int spec_type
,
736 minor_t minor_num
, const char *node_type
, int flag
)
738 #pragma unused(spec_type,node_type,flag)
739 dev_t dev
= makedev( ddi_driver_major(dip
), minor_num
);
741 if (NULL
== devfs_make_node( dev
, DEVFS_CHAR
, UID_ROOT
, GID_WHEEL
, 0666, name
, 0 ))
748 ddi_remove_minor_node(dev_info_t
*dip
, char *name
)
750 #pragma unused(dip,name)
751 /* XXX called from dtrace_detach, so NOTREACHED for now. */
757 return (major_t
) major(d
);
763 return (minor_t
) minor(d
);
767 makedevice(major_t major
, minor_t minor
)
769 return makedev( major
, minor
);
772 int ddi_getprop(dev_t dev
, dev_info_t
*dip
, int flags
, const char *name
, int defvalue
)
774 #pragma unused(dev, dip, flags, name)
780 * Kernel Debug Interface
783 kdi_dtrace_set(kdi_dtrace_set_t ignore
)
785 #pragma unused(ignore)
786 return 0; /* Success */
789 extern void Debugger(const char*);
792 debug_enter(char *c
) { Debugger(c
); }
799 dt_kmem_alloc(size_t size
, int kmflag
)
801 #pragma unused(kmflag)
804 * We ignore the M_NOWAIT bit in kmflag (all of kmflag, in fact).
805 * Requests larger than 8K with M_NOWAIT fail in kalloc_canblock.
807 #if defined(DTRACE_MEMORY_ZONES)
808 return dtrace_alloc(size
);
815 dt_kmem_zalloc(size_t size
, int kmflag
)
817 #pragma unused(kmflag)
820 * We ignore the M_NOWAIT bit in kmflag (all of kmflag, in fact).
821 * Requests larger than 8K with M_NOWAIT fail in kalloc_canblock.
823 #if defined(DTRACE_MEMORY_ZONES)
824 void* buf
= dtrace_alloc(size
);
826 void* buf
= kalloc(size
);
838 dt_kmem_free(void *buf
, size_t size
)
842 * DTrace relies on this, its doing a lot of NULL frees.
843 * A null free causes the debug builds to panic.
845 if (buf
== NULL
) return;
849 #if defined(DTRACE_MEMORY_ZONES)
850 dtrace_free(buf
, size
);
859 * aligned kmem allocator
860 * align should be a power of two
863 void* dt_kmem_alloc_aligned(size_t size
, size_t align
, int kmflag
)
865 void *mem
, **addr_to_free
;
866 intptr_t mem_aligned
;
867 size_t *size_to_free
, hdr_size
;
869 /* Must be a power of two. */
871 assert((align
& (align
- 1)) == 0);
874 * We are going to add a header to the allocation. It contains
875 * the address to free and the total size of the buffer.
877 hdr_size
= sizeof(size_t) + sizeof(void*);
878 mem
= dt_kmem_alloc(size
+ align
+ hdr_size
, kmflag
);
882 mem_aligned
= (intptr_t) (((intptr_t) mem
+ align
+ hdr_size
) & ~(align
- 1));
884 /* Write the address to free in the header. */
885 addr_to_free
= (void**) (mem_aligned
- sizeof(void*));
888 /* Write the size to free in the header. */
889 size_to_free
= (size_t*) (mem_aligned
- hdr_size
);
890 *size_to_free
= size
+ align
+ hdr_size
;
892 return (void*) mem_aligned
;
895 void* dt_kmem_zalloc_aligned(size_t size
, size_t align
, int kmflag
)
899 buf
= dt_kmem_alloc_aligned(size
, align
, kmflag
);
909 void dt_kmem_free_aligned(void* buf
, size_t size
)
912 intptr_t ptr
= (intptr_t) buf
;
913 void **addr_to_free
= (void**) (ptr
- sizeof(void*));
914 size_t *size_to_free
= (size_t*) (ptr
- (sizeof(size_t) + sizeof(void*)));
919 dt_kmem_free(*addr_to_free
, *size_to_free
);
923 * dtrace wants to manage just a single block: dtrace_state_percpu_t * NCPU, and
924 * doesn't specify constructor, destructor, or reclaim methods.
925 * At present, it always zeroes the block it obtains from kmem_cache_alloc().
926 * We'll manage this constricted use of kmem_cache with ordinary _MALLOC and _FREE.
930 const char *name
, /* descriptive name for this cache */
931 size_t bufsize
, /* size of the objects it manages */
932 size_t align
, /* required object alignment */
933 int (*constructor
)(void *, void *, int), /* object constructor */
934 void (*destructor
)(void *, void *), /* object destructor */
935 void (*reclaim
)(void *), /* memory reclaim callback */
936 void *private, /* pass-thru arg for constr/destr/reclaim */
937 vmem_t
*vmp
, /* vmem source for slab allocation */
938 int cflags
) /* cache creation flags */
940 #pragma unused(name,align,constructor,destructor,reclaim,private,vmp,cflags)
941 return (kmem_cache_t
*)bufsize
; /* A cookie that tracks the single object size. */
945 kmem_cache_alloc(kmem_cache_t
*cp
, int kmflag
)
947 #pragma unused(kmflag)
948 size_t bufsize
= (size_t)cp
;
949 return (void *)_MALLOC(bufsize
, M_TEMP
, M_WAITOK
);
953 kmem_cache_free(kmem_cache_t
*cp
, void *buf
)
960 kmem_cache_destroy(kmem_cache_t
*cp
)
968 extern void thread_call_setup(thread_call_t
, thread_call_func_t
, thread_call_param_t
); /* XXX MACH_KERNEL_PRIVATE */
971 _taskq_apply( task_func_t func
, thread_call_param_t arg
)
977 taskq_create(const char *name
, int nthreads
, pri_t pri
, int minalloc
,
978 int maxalloc
, uint_t flags
)
980 #pragma unused(name,nthreads,pri,minalloc,maxalloc,flags)
982 return (taskq_t
*)thread_call_allocate( (thread_call_func_t
)_taskq_apply
, NULL
);
986 taskq_dispatch(taskq_t
*tq
, task_func_t func
, void *arg
, uint_t flags
)
988 #pragma unused(flags)
989 thread_call_setup( (thread_call_t
) tq
, (thread_call_func_t
)_taskq_apply
, (thread_call_param_t
)func
);
990 thread_call_enter1( (thread_call_t
) tq
, (thread_call_param_t
)arg
);
991 return (taskqid_t
) tq
/* for lack of anything better */;
995 taskq_destroy(taskq_t
*tq
)
997 thread_call_cancel( (thread_call_t
) tq
);
998 thread_call_free( (thread_call_t
) tq
);
1004 * vmem (Solaris "slab" allocator) used by DTrace solely to hand out resource ids
1006 typedef unsigned int u_daddr_t
;
1009 /* By passing around blist *handles*, the underlying blist can be resized as needed. */
1015 vmem_create(const char *name
, void *base
, size_t size
, size_t quantum
, void *ignore5
,
1016 void *ignore6
, vmem_t
*source
, size_t qcache_max
, int vmflag
)
1018 #pragma unused(name,quantum,ignore5,ignore6,source,qcache_max,vmflag)
1020 struct blist_hdl
*p
= _MALLOC(sizeof(struct blist_hdl
), M_TEMP
, M_WAITOK
);
1022 ASSERT(quantum
== 1);
1023 ASSERT(NULL
== ignore5
);
1024 ASSERT(NULL
== ignore6
);
1025 ASSERT(NULL
== source
);
1026 ASSERT(0 == qcache_max
);
1027 ASSERT(vmflag
& VMC_IDENTIFIER
);
1029 size
= MIN(128, size
); /* Clamp to 128 initially, since the underlying data structure is pre-allocated */
1031 p
->blist
= bl
= blist_create( size
);
1032 blist_free(bl
, 0, size
);
1033 if (base
) blist_alloc( bl
, (daddr_t
)(uintptr_t)base
); /* Chomp off initial ID(s) */
1039 vmem_alloc(vmem_t
*vmp
, size_t size
, int vmflag
)
1041 #pragma unused(vmflag)
1042 struct blist_hdl
*q
= (struct blist_hdl
*)vmp
;
1043 blist_t bl
= q
->blist
;
1046 p
= blist_alloc(bl
, (daddr_t
)size
);
1048 if ((daddr_t
)-1 == p
) {
1049 blist_resize(&bl
, (bl
->bl_blocks
) << 1, 1);
1051 p
= blist_alloc(bl
, (daddr_t
)size
);
1052 if ((daddr_t
)-1 == p
)
1053 panic("vmem_alloc: failure after blist_resize!");
1056 return (void *)(uintptr_t)p
;
1060 vmem_free(vmem_t
*vmp
, void *vaddr
, size_t size
)
1062 struct blist_hdl
*p
= (struct blist_hdl
*)vmp
;
1064 blist_free( p
->blist
, (daddr_t
)(uintptr_t)vaddr
, (daddr_t
)size
);
1068 vmem_destroy(vmem_t
*vmp
)
1070 struct blist_hdl
*p
= (struct blist_hdl
*)vmp
;
1072 blist_destroy( p
->blist
);
1073 _FREE( p
, sizeof(struct blist_hdl
) );
1081 * dtrace_gethrestime() provides the "walltimestamp", a value that is anchored at
1082 * January 1, 1970. Because it can be called from probe context, it must take no locks.
1086 dtrace_gethrestime(void)
1089 clock_nsec_t nanosecs
;
1090 uint64_t secs64
, ns64
;
1092 clock_get_calendar_nanotime_nowait(&secs
, &nanosecs
);
1093 secs64
= (uint64_t)secs
;
1094 ns64
= (uint64_t)nanosecs
;
1096 ns64
= ns64
+ (secs64
* 1000000000LL);
1101 * dtrace_gethrtime() provides high-resolution timestamps with machine-dependent origin.
1102 * Hence its primary use is to specify intervals.
1106 dtrace_abs_to_nano(uint64_t elapsed
)
1108 static mach_timebase_info_data_t sTimebaseInfo
= { 0, 0 };
1111 * If this is the first time we've run, get the timebase.
1112 * We can use denom == 0 to indicate that sTimebaseInfo is
1113 * uninitialised because it makes no sense to have a zero
1114 * denominator in a fraction.
1117 if ( sTimebaseInfo
.denom
== 0 ) {
1118 (void) clock_timebase_info(&sTimebaseInfo
);
1122 * Convert to nanoseconds.
1123 * return (elapsed * (uint64_t)sTimebaseInfo.numer)/(uint64_t)sTimebaseInfo.denom;
1125 * Provided the final result is representable in 64 bits the following maneuver will
1126 * deliver that result without intermediate overflow.
1128 if (sTimebaseInfo
.denom
== sTimebaseInfo
.numer
)
1130 else if (sTimebaseInfo
.denom
== 1)
1131 return elapsed
* (uint64_t)sTimebaseInfo
.numer
;
1133 /* Decompose elapsed = eta32 * 2^32 + eps32: */
1134 uint64_t eta32
= elapsed
>> 32;
1135 uint64_t eps32
= elapsed
& 0x00000000ffffffffLL
;
1137 uint32_t numer
= sTimebaseInfo
.numer
, denom
= sTimebaseInfo
.denom
;
1139 /* Form product of elapsed64 (decomposed) and numer: */
1140 uint64_t mu64
= numer
* eta32
;
1141 uint64_t lambda64
= numer
* eps32
;
1143 /* Divide the constituents by denom: */
1144 uint64_t q32
= mu64
/denom
;
1145 uint64_t r32
= mu64
- (q32
* denom
); /* mu64 % denom */
1147 return (q32
<< 32) + ((r32
<< 32) + lambda64
)/denom
;
1152 dtrace_gethrtime(void)
1154 static uint64_t start
= 0;
1157 start
= mach_absolute_time();
1159 return dtrace_abs_to_nano(mach_absolute_time() - start
);
1163 * Atomicity and synchronization
1166 dtrace_cas32(uint32_t *target
, uint32_t cmp
, uint32_t new)
1168 if (OSCompareAndSwap( (UInt32
)cmp
, (UInt32
)new, (volatile UInt32
*)target
))
1171 return ~cmp
; /* Must return something *other* than cmp */
1175 dtrace_casptr(void *target
, void *cmp
, void *new)
1177 if (OSCompareAndSwapPtr( cmp
, new, (void**)target
))
1180 return (void *)(~(uintptr_t)cmp
); /* Must return something *other* than cmp */
1184 * Interrupt manipulation
1187 dtrace_interrupt_disable(void)
1189 return (dtrace_icookie_t
)ml_set_interrupts_enabled(FALSE
);
1193 dtrace_interrupt_enable(dtrace_icookie_t reenable
)
1195 (void)ml_set_interrupts_enabled((boolean_t
)reenable
);
1202 dtrace_sync_func(void) {}
1205 * dtrace_sync() is not called from probe context.
1210 dtrace_xcall(DTRACE_CPUALL
, (dtrace_xcall_t
)dtrace_sync_func
, NULL
);
1214 * The dtrace_copyin/out/instr and dtrace_fuword* routines can be called from probe context.
1217 extern kern_return_t
dtrace_copyio_preflight(addr64_t
);
1218 extern kern_return_t
dtrace_copyio_postflight(addr64_t
);
1221 dtrace_copycheck(user_addr_t uaddr
, uintptr_t kaddr
, size_t size
)
1223 #pragma unused(kaddr)
1225 vm_offset_t recover
= dtrace_set_thread_recover( current_thread(), 0 ); /* Snare any extant recovery point. */
1226 dtrace_set_thread_recover( current_thread(), recover
); /* Put it back. We *must not* re-enter and overwrite. */
1228 ASSERT(kaddr
+ size
>= kaddr
);
1230 if ( uaddr
+ size
< uaddr
|| /* Avoid address wrap. */
1231 KERN_FAILURE
== dtrace_copyio_preflight(uaddr
)) /* Machine specific setup/constraints. */
1233 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1234 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1241 dtrace_copyin(user_addr_t src
, uintptr_t dst
, size_t len
, volatile uint16_t *flags
)
1243 #pragma unused(flags)
1245 if (dtrace_copycheck( src
, dst
, len
)) {
1246 if (copyin((const user_addr_t
)src
, (char *)dst
, (vm_size_t
)len
)) {
1247 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1248 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= src
;
1250 dtrace_copyio_postflight(src
);
1255 dtrace_copyinstr(user_addr_t src
, uintptr_t dst
, size_t len
, volatile uint16_t *flags
)
1257 #pragma unused(flags)
1261 if (dtrace_copycheck( src
, dst
, len
)) {
1262 /* copyin as many as 'len' bytes. */
1263 int error
= copyinstr((const user_addr_t
)src
, (char *)dst
, (vm_size_t
)len
, &actual
);
1266 * ENAMETOOLONG is returned when 'len' bytes have been copied in but the NUL terminator was
1267 * not encountered. That does not require raising CPU_DTRACE_BADADDR, and we press on.
1268 * Note that we do *not* stuff a NUL terminator when returning ENAMETOOLONG, that's left
1271 if (error
&& error
!= ENAMETOOLONG
) {
1272 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1273 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= src
;
1275 dtrace_copyio_postflight(src
);
1280 dtrace_copyout(uintptr_t src
, user_addr_t dst
, size_t len
, volatile uint16_t *flags
)
1282 #pragma unused(flags)
1284 if (dtrace_copycheck( dst
, src
, len
)) {
1285 if (copyout((const void *)src
, dst
, (vm_size_t
)len
)) {
1286 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1287 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= dst
;
1289 dtrace_copyio_postflight(dst
);
1294 dtrace_copyoutstr(uintptr_t src
, user_addr_t dst
, size_t len
, volatile uint16_t *flags
)
1296 #pragma unused(flags)
1300 if (dtrace_copycheck( dst
, src
, len
)) {
1303 * ENAMETOOLONG is returned when 'len' bytes have been copied out but the NUL terminator was
1304 * not encountered. We raise CPU_DTRACE_BADADDR in that case.
1305 * Note that we do *not* stuff a NUL terminator when returning ENAMETOOLONG, that's left
1308 if (copyoutstr((const void *)src
, dst
, (size_t)len
, &actual
)) {
1309 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1310 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= dst
;
1312 dtrace_copyio_postflight(dst
);
1317 dtrace_fuword8(user_addr_t uaddr
)
1321 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT
);
1322 if (dtrace_copycheck( uaddr
, (uintptr_t)&ret
, sizeof(ret
))) {
1323 if (copyin((const user_addr_t
)uaddr
, (char *)&ret
, sizeof(ret
))) {
1324 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1325 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1327 dtrace_copyio_postflight(uaddr
);
1329 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT
);
1335 dtrace_fuword16(user_addr_t uaddr
)
1339 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT
);
1340 if (dtrace_copycheck( uaddr
, (uintptr_t)&ret
, sizeof(ret
))) {
1341 if (copyin((const user_addr_t
)uaddr
, (char *)&ret
, sizeof(ret
))) {
1342 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1343 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1345 dtrace_copyio_postflight(uaddr
);
1347 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT
);
1353 dtrace_fuword32(user_addr_t uaddr
)
1357 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT
);
1358 if (dtrace_copycheck( uaddr
, (uintptr_t)&ret
, sizeof(ret
))) {
1359 if (copyin((const user_addr_t
)uaddr
, (char *)&ret
, sizeof(ret
))) {
1360 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1361 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1363 dtrace_copyio_postflight(uaddr
);
1365 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT
);
1371 dtrace_fuword64(user_addr_t uaddr
)
1375 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT
);
1376 if (dtrace_copycheck( uaddr
, (uintptr_t)&ret
, sizeof(ret
))) {
1377 if (copyin((const user_addr_t
)uaddr
, (char *)&ret
, sizeof(ret
))) {
1378 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1379 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1381 dtrace_copyio_postflight(uaddr
);
1383 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT
);
1389 * Emulation of Solaris fuword / suword
1390 * Called from the fasttrap provider, so the use of copyin/out requires fewer safegaurds.
1394 fuword8(user_addr_t uaddr
, uint8_t *value
)
1396 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint8_t)) != 0) {
1404 fuword16(user_addr_t uaddr
, uint16_t *value
)
1406 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint16_t)) != 0) {
1414 fuword32(user_addr_t uaddr
, uint32_t *value
)
1416 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint32_t)) != 0) {
1424 fuword64(user_addr_t uaddr
, uint64_t *value
)
1426 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint64_t)) != 0) {
1434 fuword8_noerr(user_addr_t uaddr
, uint8_t *value
)
1436 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint8_t))) {
1442 fuword16_noerr(user_addr_t uaddr
, uint16_t *value
)
1444 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint16_t))) {
1450 fuword32_noerr(user_addr_t uaddr
, uint32_t *value
)
1452 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint32_t))) {
1458 fuword64_noerr(user_addr_t uaddr
, uint64_t *value
)
1460 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint64_t))) {
1466 suword64(user_addr_t addr
, uint64_t value
)
1468 if (copyout((const void *)&value
, addr
, sizeof(value
)) != 0) {
1476 suword32(user_addr_t addr
, uint32_t value
)
1478 if (copyout((const void *)&value
, addr
, sizeof(value
)) != 0) {
1486 suword16(user_addr_t addr
, uint16_t value
)
1488 if (copyout((const void *)&value
, addr
, sizeof(value
)) != 0) {
1496 suword8(user_addr_t addr
, uint8_t value
)
1498 if (copyout((const void *)&value
, addr
, sizeof(value
)) != 0) {
1509 extern boolean_t
dtrace_tally_fault(user_addr_t
);
1512 dtrace_tally_fault(user_addr_t uaddr
)
1514 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1515 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1516 return( DTRACE_CPUFLAG_ISSET(CPU_DTRACE_NOFAULT
) ? TRUE
: FALSE
);
1520 extern int prf(const char *, va_list, int, struct tty
*); /* bsd/kern/subr_prf.h */
1523 vuprintf(const char *format
, va_list ap
)
1525 return prf(format
, ap
, TOTTY
, NULL
);
1528 /* Not called from probe context */
1529 void cmn_err( int level
, const char *format
, ... )
1531 #pragma unused(level)
1534 va_start(alist
, format
);
1535 vuprintf(format
, alist
);
1542 * 2002-01-24 gvdl Initial implementation of strstr
1545 __private_extern__
const char *
1546 strstr(const char *in
, const char *str
)
1553 return (const char *) in
; // Trivial empty string case
1564 } while (strncmp(in
, str
, len
) != 0);
1566 return (const char *) (in
- 1);
1573 dtrace_caller(int ignore
)
1575 #pragma unused(ignore)
1576 return -1; /* Just as in Solaris dtrace_asm.s */
1580 dtrace_getstackdepth(int aframes
)
1582 struct frame
*fp
= (struct frame
*)__builtin_frame_address(0);
1583 struct frame
*nextfp
, *minfp
, *stacktop
;
1587 if ((on_intr
= CPU_ON_INTR(CPU
)) != 0)
1588 stacktop
= (struct frame
*)dtrace_get_cpu_int_stack_top();
1590 stacktop
= (struct frame
*)(dtrace_get_kernel_stack(current_thread()) + kernel_stack_size
);
1599 nextfp
= *(struct frame
**)fp
;
1601 if (nextfp
<= minfp
|| nextfp
>= stacktop
) {
1604 * Hop from interrupt stack to thread stack.
1606 vm_offset_t kstack_base
= dtrace_get_kernel_stack(current_thread());
1608 minfp
= (struct frame
*)kstack_base
;
1609 stacktop
= (struct frame
*)(kstack_base
+ kernel_stack_size
);
1621 if (depth
<= aframes
)
1624 return (depth
- aframes
);
1631 dtrace_vtime_enable(void) {}
1634 dtrace_vtime_disable(void) {}
1636 #else /* else ! CONFIG_DTRACE */
1638 #include <sys/types.h>
1639 #include <mach/vm_types.h>
1640 #include <mach/kmod.h>
1643 * This exists to prevent build errors when dtrace is unconfigured.
1646 kern_return_t
_dtrace_register_anon_DOF(char *, unsigned char *, uint32_t);
1648 kern_return_t
_dtrace_register_anon_DOF(char *arg1
, unsigned char *arg2
, uint32_t arg3
) {
1649 #pragma unused(arg1, arg2, arg3)
1651 return KERN_FAILURE
;
1654 #endif /* CONFIG_DTRACE */