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 :-( */
72 /* Solaris proc_t is the struct. Darwin's proc_t is a pointer to it. */
73 #define proc_t struct proc /* Steer clear of the Darwin typedef for proc_t */
75 /* Not called from probe context */
81 if ((p
= proc_find(pid
)) == PROC_NULL
) {
85 task_suspend_internal(p
->task
);
89 lck_mtx_lock(&p
->p_dtrace_sprlock
);
94 /* Not called from probe context */
99 lck_mtx_unlock(&p
->p_dtrace_sprlock
);
103 task_resume_internal(p
->task
);
113 // These are not exported from vm_map.h.
114 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
);
115 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
);
117 /* Not called from probe context */
119 uread(proc_t
*p
, void *buf
, user_size_t len
, user_addr_t a
)
123 ASSERT(p
!= PROC_NULL
);
124 ASSERT(p
->task
!= NULL
);
126 task_t task
= p
->task
;
129 * Grab a reference to the task vm_map_t to make sure
130 * the map isn't pulled out from under us.
132 * Because the proc_lock is not held at all times on all code
133 * paths leading here, it is possible for the proc to have
134 * exited. If the map is null, fail.
136 vm_map_t map
= get_task_map_reference(task
);
138 ret
= vm_map_read_user( map
, (vm_map_address_t
)a
, buf
, (vm_size_t
)len
);
139 vm_map_deallocate(map
);
141 ret
= KERN_TERMINATED
;
147 /* Not called from probe context */
149 uwrite(proc_t
*p
, void *buf
, user_size_t len
, user_addr_t a
)
154 ASSERT(p
->task
!= NULL
);
156 task_t task
= p
->task
;
159 * Grab a reference to the task vm_map_t to make sure
160 * the map isn't pulled out from under us.
162 * Because the proc_lock is not held at all times on all code
163 * paths leading here, it is possible for the proc to have
164 * exited. If the map is null, fail.
166 vm_map_t map
= get_task_map_reference(task
);
168 /* Find the memory permissions. */
169 uint32_t nestingDepth
=999999;
170 vm_region_submap_short_info_data_64_t info
;
171 mach_msg_type_number_t count
= VM_REGION_SUBMAP_SHORT_INFO_COUNT_64
;
172 mach_vm_address_t address
= (mach_vm_address_t
)a
;
173 mach_vm_size_t sizeOfRegion
= (mach_vm_size_t
)len
;
175 ret
= mach_vm_region_recurse(map
, &address
, &sizeOfRegion
, &nestingDepth
, (vm_region_recurse_info_t
)&info
, &count
);
176 if (ret
!= KERN_SUCCESS
)
181 if (!(info
.protection
& VM_PROT_WRITE
)) {
182 /* Save the original protection values for restoration later */
183 reprotect
= info
.protection
;
185 if (info
.max_protection
& VM_PROT_WRITE
) {
186 /* The memory is not currently writable, but can be made writable. */
187 ret
= mach_vm_protect (map
, (mach_vm_offset_t
)a
, (mach_vm_size_t
)len
, 0, reprotect
| VM_PROT_WRITE
);
190 * The memory is not currently writable, and cannot be made writable. We need to COW this memory.
192 * Strange, we can't just say "reprotect | VM_PROT_COPY", that fails.
194 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
);
197 if (ret
!= KERN_SUCCESS
)
201 /* The memory was already writable. */
202 reprotect
= VM_PROT_NONE
;
205 ret
= vm_map_write_user( map
,
210 if (ret
!= KERN_SUCCESS
)
213 if (reprotect
!= VM_PROT_NONE
) {
214 ASSERT(reprotect
& VM_PROT_EXECUTE
);
215 ret
= mach_vm_protect (map
, (mach_vm_offset_t
)a
, (mach_vm_size_t
)len
, 0, reprotect
);
219 vm_map_deallocate(map
);
221 ret
= KERN_TERMINATED
;
233 dtrace_cpu_t
*cpu_list
;
234 cpu_core_t
*cpu_core
; /* XXX TLB lockdown? */
241 * dtrace_CRED() can be called from probe context. We cannot simply call kauth_cred_get() since
242 * that function may try to resolve a lazy credential binding, which entails taking the proc_lock.
247 struct uthread
*uthread
= get_bsdthread_info(current_thread());
252 return uthread
->uu_ucred
; /* May return NOCRED which is defined to be 0 */
255 #define HAS_ALLPRIVS(cr) priv_isfullset(&CR_OEPRIV(cr))
256 #define HAS_PRIVILEGE(cr, pr) ((pr) == PRIV_ALL ? \
258 PRIV_ISASSERT(&CR_OEPRIV(cr), pr))
260 int PRIV_POLICY_CHOICE(void* cred
, int priv
, int all
)
262 #pragma unused(priv, all)
263 return kauth_cred_issuser(cred
); /* XXX TODO: How is this different from PRIV_POLICY_ONLY? */
267 PRIV_POLICY_ONLY(void *cr
, int priv
, int boolean
)
269 #pragma unused(priv, boolean)
270 return kauth_cred_issuser(cr
); /* XXX TODO: HAS_PRIVILEGE(cr, priv); */
273 /* XXX Get around const poisoning using structure assigns */
275 crgetgid(const cred_t
*cr
) { cred_t copy_cr
= *cr
; return kauth_cred_getgid(©_cr
); }
278 crgetuid(const cred_t
*cr
) { cred_t copy_cr
= *cr
; return kauth_cred_getuid(©_cr
); }
284 typedef struct wrap_timer_call
{
285 /* node attributes */
291 struct timer_call call
;
293 /* next item in the linked list */
294 LIST_ENTRY(wrap_timer_call
) entries
;
297 #define WAKEUP_REAPER 0x7FFFFFFFFFFFFFFFLL
298 #define NEARLY_FOREVER 0x7FFFFFFFFFFFFFFELL
301 typedef struct cyc_list
{
302 cyc_omni_handler_t cyl_omni
;
303 wrap_timer_call_t cyl_wrap_by_cpus
[];
306 /* CPU going online/offline notifications */
307 void (*dtrace_cpu_state_changed_hook
)(int, boolean_t
) = NULL
;
308 void dtrace_cpu_state_changed(int, boolean_t
);
311 dtrace_install_cpu_hooks(void) {
312 dtrace_cpu_state_changed_hook
= dtrace_cpu_state_changed
;
316 dtrace_cpu_state_changed(int cpuid
, boolean_t is_running
) {
317 #pragma unused(cpuid)
318 wrap_timer_call_t
*wrapTC
= NULL
;
319 boolean_t suspend
= (is_running
? FALSE
: TRUE
);
322 /* Ensure that we're not going to leave the CPU */
323 s
= dtrace_interrupt_disable();
324 assert(cpuid
== cpu_number());
326 LIST_FOREACH(wrapTC
, &(cpu_list
[cpu_number()].cpu_cyc_list
), entries
) {
327 assert(wrapTC
->cpuid
== cpu_number());
329 assert(!wrapTC
->suspended
);
330 /* If this fails, we'll panic anyway, so let's do this now. */
331 if (!timer_call_cancel(&wrapTC
->call
))
332 panic("timer_call_set_suspend() failed to cancel a timer call");
333 wrapTC
->suspended
= TRUE
;
335 /* Rearm the timer, but ensure it was suspended first. */
336 assert(wrapTC
->suspended
);
337 clock_deadline_for_periodic_event(wrapTC
->when
.cyt_interval
, mach_absolute_time(),
339 timer_call_enter1(&wrapTC
->call
, (void*) wrapTC
, wrapTC
->deadline
,
340 TIMER_CALL_SYS_CRITICAL
| TIMER_CALL_LOCAL
);
341 wrapTC
->suspended
= FALSE
;
346 /* Restore the previous interrupt state. */
347 dtrace_interrupt_enable(s
);
351 _timer_call_apply_cyclic( void *ignore
, void *vTChdl
)
353 #pragma unused(ignore)
354 wrap_timer_call_t
*wrapTC
= (wrap_timer_call_t
*)vTChdl
;
356 (*(wrapTC
->hdlr
.cyh_func
))( wrapTC
->hdlr
.cyh_arg
);
358 clock_deadline_for_periodic_event( wrapTC
->when
.cyt_interval
, mach_absolute_time(), &(wrapTC
->deadline
) );
359 timer_call_enter1( &(wrapTC
->call
), (void *)wrapTC
, wrapTC
->deadline
, TIMER_CALL_SYS_CRITICAL
| TIMER_CALL_LOCAL
);
363 timer_call_add_cyclic(wrap_timer_call_t
*wrapTC
, cyc_handler_t
*handler
, cyc_time_t
*when
)
368 timer_call_setup( &(wrapTC
->call
), _timer_call_apply_cyclic
, NULL
);
369 wrapTC
->hdlr
= *handler
;
370 wrapTC
->when
= *when
;
372 nanoseconds_to_absolutetime( wrapTC
->when
.cyt_interval
, (uint64_t *)&wrapTC
->when
.cyt_interval
);
374 now
= mach_absolute_time();
375 wrapTC
->deadline
= now
;
377 clock_deadline_for_periodic_event( wrapTC
->when
.cyt_interval
, now
, &(wrapTC
->deadline
) );
379 /* Insert the timer to the list of the running timers on this CPU, and start it. */
380 s
= dtrace_interrupt_disable();
381 wrapTC
->cpuid
= cpu_number();
382 LIST_INSERT_HEAD(&cpu_list
[wrapTC
->cpuid
].cpu_cyc_list
, wrapTC
, entries
);
383 timer_call_enter1(&wrapTC
->call
, (void*) wrapTC
, wrapTC
->deadline
,
384 TIMER_CALL_SYS_CRITICAL
| TIMER_CALL_LOCAL
);
385 wrapTC
->suspended
= FALSE
;
386 dtrace_interrupt_enable(s
);
388 return (cyclic_id_t
)wrapTC
;
392 * Executed on the CPU the timer is running on.
395 timer_call_remove_cyclic(wrap_timer_call_t
*wrapTC
)
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(wrap_timer_call_t
*wrapTC
)
409 return (wrapTC
? wrapTC
->hdlr
.cyh_arg
: NULL
);
413 cyclic_timer_add(cyc_handler_t
*handler
, cyc_time_t
*when
)
415 wrap_timer_call_t
*wrapTC
= _MALLOC(sizeof(wrap_timer_call_t
), M_TEMP
, M_ZERO
| M_WAITOK
);
419 return timer_call_add_cyclic( wrapTC
, handler
, when
);
423 cyclic_timer_remove(cyclic_id_t cyclic
)
425 ASSERT( cyclic
!= CYCLIC_NONE
);
427 /* Removing a timer call must be done on the CPU the timer is running on. */
428 wrap_timer_call_t
*wrapTC
= (wrap_timer_call_t
*) cyclic
;
429 dtrace_xcall(wrapTC
->cpuid
, (dtrace_xcall_t
) timer_call_remove_cyclic
, (void*) cyclic
);
431 _FREE((void *)cyclic
, M_TEMP
);
435 _cyclic_add_omni(cyc_list_t
*cyc_list
)
439 cyc_omni_handler_t
*omni
= &cyc_list
->cyl_omni
;
441 (omni
->cyo_online
)(omni
->cyo_arg
, CPU
, &cH
, &cT
);
443 wrap_timer_call_t
*wrapTC
= &cyc_list
->cyl_wrap_by_cpus
[cpu_number()];
444 timer_call_add_cyclic(wrapTC
, &cH
, &cT
);
448 cyclic_add_omni(cyc_omni_handler_t
*omni
)
450 cyc_list_t
*cyc_list
=
451 _MALLOC(sizeof(cyc_list_t
) + NCPU
* sizeof(wrap_timer_call_t
), M_TEMP
, M_ZERO
| M_WAITOK
);
453 if (NULL
== cyc_list
)
456 cyc_list
->cyl_omni
= *omni
;
458 dtrace_xcall(DTRACE_CPUALL
, (dtrace_xcall_t
)_cyclic_add_omni
, (void *)cyc_list
);
460 return (cyclic_id_list_t
)cyc_list
;
464 _cyclic_remove_omni(cyc_list_t
*cyc_list
)
466 cyc_omni_handler_t
*omni
= &cyc_list
->cyl_omni
;
468 wrap_timer_call_t
*wrapTC
;
471 * If the processor was offline when dtrace started, we did not allocate
472 * a cyclic timer for this CPU.
474 if ((wrapTC
= &cyc_list
->cyl_wrap_by_cpus
[cpu_number()]) != NULL
) {
475 oarg
= timer_call_get_cyclic_arg(wrapTC
);
476 timer_call_remove_cyclic(wrapTC
);
477 (omni
->cyo_offline
)(omni
->cyo_arg
, CPU
, oarg
);
482 cyclic_remove_omni(cyclic_id_list_t cyc_list
)
484 ASSERT(cyc_list
!= NULL
);
486 dtrace_xcall(DTRACE_CPUALL
, (dtrace_xcall_t
)_cyclic_remove_omni
, (void *)cyc_list
);
487 _FREE(cyc_list
, M_TEMP
);
490 typedef struct wrap_thread_call
{
495 } wrap_thread_call_t
;
498 * _cyclic_apply will run on some thread under kernel_task. That's OK for the
499 * cleaner and the deadman, but too distant in time and place for the profile provider.
502 _cyclic_apply( void *ignore
, void *vTChdl
)
504 #pragma unused(ignore)
505 wrap_thread_call_t
*wrapTC
= (wrap_thread_call_t
*)vTChdl
;
507 (*(wrapTC
->hdlr
.cyh_func
))( wrapTC
->hdlr
.cyh_arg
);
509 clock_deadline_for_periodic_event( wrapTC
->when
.cyt_interval
, mach_absolute_time(), &(wrapTC
->deadline
) );
510 (void)thread_call_enter1_delayed( wrapTC
->TChdl
, (void *)wrapTC
, wrapTC
->deadline
);
512 /* Did cyclic_remove request a wakeup call when this thread call was re-armed? */
513 if (wrapTC
->when
.cyt_interval
== WAKEUP_REAPER
)
514 thread_wakeup((event_t
)wrapTC
);
518 cyclic_add(cyc_handler_t
*handler
, cyc_time_t
*when
)
522 wrap_thread_call_t
*wrapTC
= _MALLOC(sizeof(wrap_thread_call_t
), M_TEMP
, M_ZERO
| M_WAITOK
);
526 wrapTC
->TChdl
= thread_call_allocate( _cyclic_apply
, NULL
);
527 wrapTC
->hdlr
= *handler
;
528 wrapTC
->when
= *when
;
530 ASSERT(when
->cyt_when
== 0);
531 ASSERT(when
->cyt_interval
< WAKEUP_REAPER
);
533 nanoseconds_to_absolutetime(wrapTC
->when
.cyt_interval
, (uint64_t *)&wrapTC
->when
.cyt_interval
);
535 now
= mach_absolute_time();
536 wrapTC
->deadline
= now
;
538 clock_deadline_for_periodic_event( wrapTC
->when
.cyt_interval
, now
, &(wrapTC
->deadline
) );
539 (void)thread_call_enter1_delayed( wrapTC
->TChdl
, (void *)wrapTC
, wrapTC
->deadline
);
541 return (cyclic_id_t
)wrapTC
;
545 noop_cyh_func(void * ignore
)
547 #pragma unused(ignore)
551 cyclic_remove(cyclic_id_t cyclic
)
553 wrap_thread_call_t
*wrapTC
= (wrap_thread_call_t
*)cyclic
;
555 ASSERT(cyclic
!= CYCLIC_NONE
);
557 while (!thread_call_cancel(wrapTC
->TChdl
)) {
558 int ret
= assert_wait(wrapTC
, THREAD_UNINT
);
559 ASSERT(ret
== THREAD_WAITING
);
561 wrapTC
->when
.cyt_interval
= WAKEUP_REAPER
;
563 ret
= thread_block(THREAD_CONTINUE_NULL
);
564 ASSERT(ret
== THREAD_AWAKENED
);
567 if (thread_call_free(wrapTC
->TChdl
))
568 _FREE(wrapTC
, M_TEMP
);
570 /* Gut this cyclic and move on ... */
571 wrapTC
->hdlr
.cyh_func
= noop_cyh_func
;
572 wrapTC
->when
.cyt_interval
= NEARLY_FOREVER
;
577 * timeout / untimeout (converted to dtrace_timeout / dtrace_untimeout due to name collision)
581 dtrace_timeout(void (*func
)(void *, void *), void* arg
, uint64_t nanos
)
584 thread_call_t call
= thread_call_allocate(func
, NULL
);
586 nanoseconds_to_absolutetime(nanos
, &nanos
);
589 * This method does not use clock_deadline_for_periodic_event() because it is a one-shot,
590 * and clock drift on later invocations is not a worry.
592 uint64_t deadline
= mach_absolute_time() + nanos
;
593 /* DRK: consider using a lower priority callout here */
594 thread_call_enter_delayed(call
, deadline
);
603 ddi_report_dev(dev_info_t
*devi
)
609 static unsigned int gRegisteredProps
= 0;
611 char name
[32]; /* enough for "dof-data-" + digits */
616 kern_return_t
_dtrace_register_anon_DOF(char *, uchar_t
*, uint_t
);
619 _dtrace_register_anon_DOF(char *name
, uchar_t
*data
, uint_t nelements
)
621 if (gRegisteredProps
< sizeof(gPropTable
)/sizeof(gPropTable
[0])) {
622 int *p
= (int *)_MALLOC(nelements
*sizeof(int), M_TEMP
, M_WAITOK
);
627 strlcpy(gPropTable
[gRegisteredProps
].name
, name
, sizeof(gPropTable
[0].name
));
628 gPropTable
[gRegisteredProps
].nelements
= nelements
;
629 gPropTable
[gRegisteredProps
].data
= p
;
631 while (nelements
-- > 0) {
632 *p
++ = (int)(*data
++);
643 ddi_prop_lookup_int_array(dev_t match_dev
, dev_info_t
*dip
, uint_t flags
,
644 const char *name
, int **data
, uint_t
*nelements
)
646 #pragma unused(match_dev,dip,flags)
648 for (i
= 0; i
< gRegisteredProps
; ++i
)
650 if (0 == strncmp(name
, gPropTable
[i
].name
,
651 sizeof(gPropTable
[i
].name
))) {
652 *data
= gPropTable
[i
].data
;
653 *nelements
= gPropTable
[i
].nelements
;
661 ddi_prop_free(void *buf
)
668 ddi_driver_major(dev_info_t
*devi
) { return (int)major(CAST_DOWN_EXPLICIT(int,devi
)); }
671 ddi_create_minor_node(dev_info_t
*dip
, const char *name
, int spec_type
,
672 minor_t minor_num
, const char *node_type
, int flag
)
674 #pragma unused(spec_type,node_type,flag)
675 dev_t dev
= makedev( ddi_driver_major(dip
), minor_num
);
677 if (NULL
== devfs_make_node( dev
, DEVFS_CHAR
, UID_ROOT
, GID_WHEEL
, 0666, name
, 0 ))
684 ddi_remove_minor_node(dev_info_t
*dip
, char *name
)
686 #pragma unused(dip,name)
687 /* XXX called from dtrace_detach, so NOTREACHED for now. */
693 return (major_t
) major(d
);
699 return (minor_t
) minor(d
);
703 makedevice(major_t major
, minor_t minor
)
705 return makedev( major
, minor
);
708 int ddi_getprop(dev_t dev
, dev_info_t
*dip
, int flags
, const char *name
, int defvalue
)
710 #pragma unused(dev, dip, flags, name)
716 * Kernel Debug Interface
719 kdi_dtrace_set(kdi_dtrace_set_t ignore
)
721 #pragma unused(ignore)
722 return 0; /* Success */
725 extern void Debugger(const char*);
728 debug_enter(char *c
) { Debugger(c
); }
735 dt_kmem_alloc(size_t size
, int kmflag
)
737 #pragma unused(kmflag)
740 * We ignore the M_NOWAIT bit in kmflag (all of kmflag, in fact).
741 * Requests larger than 8K with M_NOWAIT fail in kalloc_canblock.
743 #if defined(DTRACE_MEMORY_ZONES)
744 return dtrace_alloc(size
);
751 dt_kmem_zalloc(size_t size
, int kmflag
)
753 #pragma unused(kmflag)
756 * We ignore the M_NOWAIT bit in kmflag (all of kmflag, in fact).
757 * Requests larger than 8K with M_NOWAIT fail in kalloc_canblock.
759 #if defined(DTRACE_MEMORY_ZONES)
760 void* buf
= dtrace_alloc(size
);
762 void* buf
= kalloc(size
);
774 dt_kmem_free(void *buf
, size_t size
)
778 * DTrace relies on this, its doing a lot of NULL frees.
779 * A null free causes the debug builds to panic.
781 if (buf
== NULL
) return;
785 #if defined(DTRACE_MEMORY_ZONES)
786 dtrace_free(buf
, size
);
795 * aligned kmem allocator
796 * align should be a power of two
799 void* dt_kmem_alloc_aligned(size_t size
, size_t align
, int kmflag
)
801 void *mem
, **addr_to_free
;
802 intptr_t mem_aligned
;
803 size_t *size_to_free
, hdr_size
;
805 /* Must be a power of two. */
807 assert((align
& (align
- 1)) == 0);
810 * We are going to add a header to the allocation. It contains
811 * the address to free and the total size of the buffer.
813 hdr_size
= sizeof(size_t) + sizeof(void*);
814 mem
= dt_kmem_alloc(size
+ align
+ hdr_size
, kmflag
);
818 mem_aligned
= (intptr_t) (((intptr_t) mem
+ align
+ hdr_size
) & ~(align
- 1));
820 /* Write the address to free in the header. */
821 addr_to_free
= (void**) (mem_aligned
- sizeof(void*));
824 /* Write the size to free in the header. */
825 size_to_free
= (size_t*) (mem_aligned
- hdr_size
);
826 *size_to_free
= size
+ align
+ hdr_size
;
828 return (void*) mem_aligned
;
831 void* dt_kmem_zalloc_aligned(size_t size
, size_t align
, int kmflag
)
835 buf
= dt_kmem_alloc_aligned(size
, align
, kmflag
);
845 void dt_kmem_free_aligned(void* buf
, size_t size
)
848 intptr_t ptr
= (intptr_t) buf
;
849 void **addr_to_free
= (void**) (ptr
- sizeof(void*));
850 size_t *size_to_free
= (size_t*) (ptr
- (sizeof(size_t) + sizeof(void*)));
855 dt_kmem_free(*addr_to_free
, *size_to_free
);
859 * dtrace wants to manage just a single block: dtrace_state_percpu_t * NCPU, and
860 * doesn't specify constructor, destructor, or reclaim methods.
861 * At present, it always zeroes the block it obtains from kmem_cache_alloc().
862 * We'll manage this constricted use of kmem_cache with ordinary _MALLOC and _FREE.
866 const char *name
, /* descriptive name for this cache */
867 size_t bufsize
, /* size of the objects it manages */
868 size_t align
, /* required object alignment */
869 int (*constructor
)(void *, void *, int), /* object constructor */
870 void (*destructor
)(void *, void *), /* object destructor */
871 void (*reclaim
)(void *), /* memory reclaim callback */
872 void *private, /* pass-thru arg for constr/destr/reclaim */
873 vmem_t
*vmp
, /* vmem source for slab allocation */
874 int cflags
) /* cache creation flags */
876 #pragma unused(name,align,constructor,destructor,reclaim,private,vmp,cflags)
877 return (kmem_cache_t
*)bufsize
; /* A cookie that tracks the single object size. */
881 kmem_cache_alloc(kmem_cache_t
*cp
, int kmflag
)
883 #pragma unused(kmflag)
884 size_t bufsize
= (size_t)cp
;
885 return (void *)_MALLOC(bufsize
, M_TEMP
, M_WAITOK
);
889 kmem_cache_free(kmem_cache_t
*cp
, void *buf
)
896 kmem_cache_destroy(kmem_cache_t
*cp
)
904 extern void thread_call_setup(thread_call_t
, thread_call_func_t
, thread_call_param_t
); /* XXX MACH_KERNEL_PRIVATE */
907 _taskq_apply( task_func_t func
, thread_call_param_t arg
)
913 taskq_create(const char *name
, int nthreads
, pri_t pri
, int minalloc
,
914 int maxalloc
, uint_t flags
)
916 #pragma unused(name,nthreads,pri,minalloc,maxalloc,flags)
918 return (taskq_t
*)thread_call_allocate( (thread_call_func_t
)_taskq_apply
, NULL
);
922 taskq_dispatch(taskq_t
*tq
, task_func_t func
, void *arg
, uint_t flags
)
924 #pragma unused(flags)
925 thread_call_setup( (thread_call_t
) tq
, (thread_call_func_t
)_taskq_apply
, (thread_call_param_t
)func
);
926 thread_call_enter1( (thread_call_t
) tq
, (thread_call_param_t
)arg
);
927 return (taskqid_t
) tq
/* for lack of anything better */;
931 taskq_destroy(taskq_t
*tq
)
933 thread_call_cancel( (thread_call_t
) tq
);
934 thread_call_free( (thread_call_t
) tq
);
940 * vmem (Solaris "slab" allocator) used by DTrace solely to hand out resource ids
942 typedef unsigned int u_daddr_t
;
945 /* By passing around blist *handles*, the underlying blist can be resized as needed. */
951 vmem_create(const char *name
, void *base
, size_t size
, size_t quantum
, void *ignore5
,
952 void *ignore6
, vmem_t
*source
, size_t qcache_max
, int vmflag
)
954 #pragma unused(name,quantum,ignore5,ignore6,source,qcache_max,vmflag)
956 struct blist_hdl
*p
= _MALLOC(sizeof(struct blist_hdl
), M_TEMP
, M_WAITOK
);
958 ASSERT(quantum
== 1);
959 ASSERT(NULL
== ignore5
);
960 ASSERT(NULL
== ignore6
);
961 ASSERT(NULL
== source
);
962 ASSERT(0 == qcache_max
);
963 ASSERT(vmflag
& VMC_IDENTIFIER
);
965 size
= MIN(128, size
); /* Clamp to 128 initially, since the underlying data structure is pre-allocated */
967 p
->blist
= bl
= blist_create( size
);
968 blist_free(bl
, 0, size
);
969 if (base
) blist_alloc( bl
, (daddr_t
)(uintptr_t)base
); /* Chomp off initial ID(s) */
975 vmem_alloc(vmem_t
*vmp
, size_t size
, int vmflag
)
977 #pragma unused(vmflag)
978 struct blist_hdl
*q
= (struct blist_hdl
*)vmp
;
979 blist_t bl
= q
->blist
;
982 p
= blist_alloc(bl
, (daddr_t
)size
);
984 if ((daddr_t
)-1 == p
) {
985 blist_resize(&bl
, (bl
->bl_blocks
) << 1, 1);
987 p
= blist_alloc(bl
, (daddr_t
)size
);
988 if ((daddr_t
)-1 == p
)
989 panic("vmem_alloc: failure after blist_resize!");
992 return (void *)(uintptr_t)p
;
996 vmem_free(vmem_t
*vmp
, void *vaddr
, size_t size
)
998 struct blist_hdl
*p
= (struct blist_hdl
*)vmp
;
1000 blist_free( p
->blist
, (daddr_t
)(uintptr_t)vaddr
, (daddr_t
)size
);
1004 vmem_destroy(vmem_t
*vmp
)
1006 struct blist_hdl
*p
= (struct blist_hdl
*)vmp
;
1008 blist_destroy( p
->blist
);
1009 _FREE( p
, sizeof(struct blist_hdl
) );
1017 * dtrace_gethrestime() provides the "walltimestamp", a value that is anchored at
1018 * January 1, 1970. Because it can be called from probe context, it must take no locks.
1022 dtrace_gethrestime(void)
1025 clock_nsec_t nanosecs
;
1026 uint64_t secs64
, ns64
;
1028 clock_get_calendar_nanotime_nowait(&secs
, &nanosecs
);
1029 secs64
= (uint64_t)secs
;
1030 ns64
= (uint64_t)nanosecs
;
1032 ns64
= ns64
+ (secs64
* 1000000000LL);
1037 * dtrace_gethrtime() provides high-resolution timestamps with machine-dependent origin.
1038 * Hence its primary use is to specify intervals.
1042 dtrace_abs_to_nano(uint64_t elapsed
)
1044 static mach_timebase_info_data_t sTimebaseInfo
= { 0, 0 };
1047 * If this is the first time we've run, get the timebase.
1048 * We can use denom == 0 to indicate that sTimebaseInfo is
1049 * uninitialised because it makes no sense to have a zero
1050 * denominator in a fraction.
1053 if ( sTimebaseInfo
.denom
== 0 ) {
1054 (void) clock_timebase_info(&sTimebaseInfo
);
1058 * Convert to nanoseconds.
1059 * return (elapsed * (uint64_t)sTimebaseInfo.numer)/(uint64_t)sTimebaseInfo.denom;
1061 * Provided the final result is representable in 64 bits the following maneuver will
1062 * deliver that result without intermediate overflow.
1064 if (sTimebaseInfo
.denom
== sTimebaseInfo
.numer
)
1066 else if (sTimebaseInfo
.denom
== 1)
1067 return elapsed
* (uint64_t)sTimebaseInfo
.numer
;
1069 /* Decompose elapsed = eta32 * 2^32 + eps32: */
1070 uint64_t eta32
= elapsed
>> 32;
1071 uint64_t eps32
= elapsed
& 0x00000000ffffffffLL
;
1073 uint32_t numer
= sTimebaseInfo
.numer
, denom
= sTimebaseInfo
.denom
;
1075 /* Form product of elapsed64 (decomposed) and numer: */
1076 uint64_t mu64
= numer
* eta32
;
1077 uint64_t lambda64
= numer
* eps32
;
1079 /* Divide the constituents by denom: */
1080 uint64_t q32
= mu64
/denom
;
1081 uint64_t r32
= mu64
- (q32
* denom
); /* mu64 % denom */
1083 return (q32
<< 32) + ((r32
<< 32) + lambda64
)/denom
;
1088 dtrace_gethrtime(void)
1090 static uint64_t start
= 0;
1093 start
= mach_absolute_time();
1095 return dtrace_abs_to_nano(mach_absolute_time() - start
);
1099 * Atomicity and synchronization
1102 dtrace_cas32(uint32_t *target
, uint32_t cmp
, uint32_t new)
1104 if (OSCompareAndSwap( (UInt32
)cmp
, (UInt32
)new, (volatile UInt32
*)target
))
1107 return ~cmp
; /* Must return something *other* than cmp */
1111 dtrace_casptr(void *target
, void *cmp
, void *new)
1113 if (OSCompareAndSwapPtr( cmp
, new, (void**)target
))
1116 return (void *)(~(uintptr_t)cmp
); /* Must return something *other* than cmp */
1120 * Interrupt manipulation
1123 dtrace_interrupt_disable(void)
1125 return (dtrace_icookie_t
)ml_set_interrupts_enabled(FALSE
);
1129 dtrace_interrupt_enable(dtrace_icookie_t reenable
)
1131 (void)ml_set_interrupts_enabled((boolean_t
)reenable
);
1138 dtrace_sync_func(void) {}
1141 * dtrace_sync() is not called from probe context.
1146 dtrace_xcall(DTRACE_CPUALL
, (dtrace_xcall_t
)dtrace_sync_func
, NULL
);
1150 * The dtrace_copyin/out/instr and dtrace_fuword* routines can be called from probe context.
1153 extern kern_return_t
dtrace_copyio_preflight(addr64_t
);
1154 extern kern_return_t
dtrace_copyio_postflight(addr64_t
);
1157 dtrace_copycheck(user_addr_t uaddr
, uintptr_t kaddr
, size_t size
)
1159 #pragma unused(kaddr)
1161 vm_offset_t recover
= dtrace_set_thread_recover( current_thread(), 0 ); /* Snare any extant recovery point. */
1162 dtrace_set_thread_recover( current_thread(), recover
); /* Put it back. We *must not* re-enter and overwrite. */
1164 ASSERT(kaddr
+ size
>= kaddr
);
1166 if ( uaddr
+ size
< uaddr
|| /* Avoid address wrap. */
1167 KERN_FAILURE
== dtrace_copyio_preflight(uaddr
)) /* Machine specific setup/constraints. */
1169 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1170 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1177 dtrace_copyin(user_addr_t src
, uintptr_t dst
, size_t len
, volatile uint16_t *flags
)
1179 #pragma unused(flags)
1181 if (dtrace_copycheck( src
, dst
, len
)) {
1182 if (copyin((const user_addr_t
)src
, (char *)dst
, (vm_size_t
)len
)) {
1183 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1184 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= src
;
1186 dtrace_copyio_postflight(src
);
1191 dtrace_copyinstr(user_addr_t src
, uintptr_t dst
, size_t len
, volatile uint16_t *flags
)
1193 #pragma unused(flags)
1197 if (dtrace_copycheck( src
, dst
, len
)) {
1198 /* copyin as many as 'len' bytes. */
1199 int error
= copyinstr((const user_addr_t
)src
, (char *)dst
, (vm_size_t
)len
, &actual
);
1202 * ENAMETOOLONG is returned when 'len' bytes have been copied in but the NUL terminator was
1203 * not encountered. That does not require raising CPU_DTRACE_BADADDR, and we press on.
1204 * Note that we do *not* stuff a NUL terminator when returning ENAMETOOLONG, that's left
1207 if (error
&& error
!= ENAMETOOLONG
) {
1208 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1209 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= src
;
1211 dtrace_copyio_postflight(src
);
1216 dtrace_copyout(uintptr_t src
, user_addr_t dst
, size_t len
, volatile uint16_t *flags
)
1218 #pragma unused(flags)
1220 if (dtrace_copycheck( dst
, src
, len
)) {
1221 if (copyout((const void *)src
, dst
, (vm_size_t
)len
)) {
1222 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1223 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= dst
;
1225 dtrace_copyio_postflight(dst
);
1230 dtrace_copyoutstr(uintptr_t src
, user_addr_t dst
, size_t len
, volatile uint16_t *flags
)
1232 #pragma unused(flags)
1236 if (dtrace_copycheck( dst
, src
, len
)) {
1239 * ENAMETOOLONG is returned when 'len' bytes have been copied out but the NUL terminator was
1240 * not encountered. We raise CPU_DTRACE_BADADDR in that case.
1241 * Note that we do *not* stuff a NUL terminator when returning ENAMETOOLONG, that's left
1244 if (copyoutstr((const void *)src
, dst
, (size_t)len
, &actual
)) {
1245 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1246 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= dst
;
1248 dtrace_copyio_postflight(dst
);
1253 dtrace_fuword8(user_addr_t uaddr
)
1257 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT
);
1258 if (dtrace_copycheck( uaddr
, (uintptr_t)&ret
, sizeof(ret
))) {
1259 if (copyin((const user_addr_t
)uaddr
, (char *)&ret
, sizeof(ret
))) {
1260 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1261 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1263 dtrace_copyio_postflight(uaddr
);
1265 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT
);
1271 dtrace_fuword16(user_addr_t uaddr
)
1275 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT
);
1276 if (dtrace_copycheck( uaddr
, (uintptr_t)&ret
, sizeof(ret
))) {
1277 if (copyin((const user_addr_t
)uaddr
, (char *)&ret
, sizeof(ret
))) {
1278 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1279 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1281 dtrace_copyio_postflight(uaddr
);
1283 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT
);
1289 dtrace_fuword32(user_addr_t uaddr
)
1293 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT
);
1294 if (dtrace_copycheck( uaddr
, (uintptr_t)&ret
, sizeof(ret
))) {
1295 if (copyin((const user_addr_t
)uaddr
, (char *)&ret
, sizeof(ret
))) {
1296 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1297 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1299 dtrace_copyio_postflight(uaddr
);
1301 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT
);
1307 dtrace_fuword64(user_addr_t uaddr
)
1311 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT
);
1312 if (dtrace_copycheck( uaddr
, (uintptr_t)&ret
, sizeof(ret
))) {
1313 if (copyin((const user_addr_t
)uaddr
, (char *)&ret
, sizeof(ret
))) {
1314 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1315 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1317 dtrace_copyio_postflight(uaddr
);
1319 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT
);
1325 * Emulation of Solaris fuword / suword
1326 * Called from the fasttrap provider, so the use of copyin/out requires fewer safegaurds.
1330 fuword8(user_addr_t uaddr
, uint8_t *value
)
1332 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint8_t)) != 0) {
1340 fuword16(user_addr_t uaddr
, uint16_t *value
)
1342 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint16_t)) != 0) {
1350 fuword32(user_addr_t uaddr
, uint32_t *value
)
1352 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint32_t)) != 0) {
1360 fuword64(user_addr_t uaddr
, uint64_t *value
)
1362 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint64_t)) != 0) {
1370 fuword8_noerr(user_addr_t uaddr
, uint8_t *value
)
1372 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint8_t))) {
1378 fuword16_noerr(user_addr_t uaddr
, uint16_t *value
)
1380 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint16_t))) {
1386 fuword32_noerr(user_addr_t uaddr
, uint32_t *value
)
1388 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint32_t))) {
1394 fuword64_noerr(user_addr_t uaddr
, uint64_t *value
)
1396 if (copyin((const user_addr_t
)uaddr
, (char *)value
, sizeof(uint64_t))) {
1402 suword64(user_addr_t addr
, uint64_t value
)
1404 if (copyout((const void *)&value
, addr
, sizeof(value
)) != 0) {
1412 suword32(user_addr_t addr
, uint32_t value
)
1414 if (copyout((const void *)&value
, addr
, sizeof(value
)) != 0) {
1422 suword16(user_addr_t addr
, uint16_t value
)
1424 if (copyout((const void *)&value
, addr
, sizeof(value
)) != 0) {
1432 suword8(user_addr_t addr
, uint8_t value
)
1434 if (copyout((const void *)&value
, addr
, sizeof(value
)) != 0) {
1445 extern boolean_t
dtrace_tally_fault(user_addr_t
);
1448 dtrace_tally_fault(user_addr_t uaddr
)
1450 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR
);
1451 cpu_core
[CPU
->cpu_id
].cpuc_dtrace_illval
= uaddr
;
1452 return( DTRACE_CPUFLAG_ISSET(CPU_DTRACE_NOFAULT
) ? TRUE
: FALSE
);
1456 extern int prf(const char *, va_list, int, struct tty
*); /* bsd/kern/subr_prf.h */
1459 vuprintf(const char *format
, va_list ap
)
1461 return prf(format
, ap
, TOTTY
, NULL
);
1464 /* Not called from probe context */
1465 void cmn_err( int level
, const char *format
, ... )
1467 #pragma unused(level)
1470 va_start(alist
, format
);
1471 vuprintf(format
, alist
);
1478 * 2002-01-24 gvdl Initial implementation of strstr
1481 __private_extern__
const char *
1482 strstr(const char *in
, const char *str
)
1489 return (const char *) in
; // Trivial empty string case
1500 } while (strncmp(in
, str
, len
) != 0);
1502 return (const char *) (in
- 1);
1509 dtrace_caller(int ignore
)
1511 #pragma unused(ignore)
1512 return -1; /* Just as in Solaris dtrace_asm.s */
1516 dtrace_getstackdepth(int aframes
)
1518 struct frame
*fp
= (struct frame
*)__builtin_frame_address(0);
1519 struct frame
*nextfp
, *minfp
, *stacktop
;
1523 if ((on_intr
= CPU_ON_INTR(CPU
)) != 0)
1524 stacktop
= (struct frame
*)dtrace_get_cpu_int_stack_top();
1526 stacktop
= (struct frame
*)(dtrace_get_kernel_stack(current_thread()) + kernel_stack_size
);
1535 nextfp
= *(struct frame
**)fp
;
1537 if (nextfp
<= minfp
|| nextfp
>= stacktop
) {
1540 * Hop from interrupt stack to thread stack.
1542 vm_offset_t kstack_base
= dtrace_get_kernel_stack(current_thread());
1544 minfp
= (struct frame
*)kstack_base
;
1545 stacktop
= (struct frame
*)(kstack_base
+ kernel_stack_size
);
1557 if (depth
<= aframes
)
1560 return (depth
- aframes
);
1567 dtrace_vtime_enable(void) {}
1570 dtrace_vtime_disable(void) {}
1572 #else /* else ! CONFIG_DTRACE */
1574 #include <sys/types.h>
1575 #include <mach/vm_types.h>
1576 #include <mach/kmod.h>
1579 * This exists to prevent build errors when dtrace is unconfigured.
1582 kern_return_t
_dtrace_register_anon_DOF(char *, unsigned char *, uint32_t);
1584 kern_return_t
_dtrace_register_anon_DOF(char *arg1
, unsigned char *arg2
, uint32_t arg3
) {
1585 #pragma unused(arg1, arg2, arg3)
1587 return KERN_FAILURE
;
1590 #endif /* CONFIG_DTRACE */