4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
25 * Copyright 2007 Sun Microsystems, Inc. All rights reserved.
26 * Use is subject to license terms.
30 * Copyright (c) 2011, Joyent, Inc. All rights reserved.
33 #ifndef _SYS_DTRACE_IMPL_H
34 #define _SYS_DTRACE_IMPL_H
41 * DTrace Dynamic Tracing Software: Kernel Implementation Interfaces
43 * Note: The contents of this file are private to the implementation of the
44 * Solaris system and DTrace subsystem and are subject to change at any time
45 * without notice. Applications and drivers using these interfaces will fail
46 * to run on future releases. These interfaces should not be used for any
47 * purpose except those expressly outlined in dtrace(7D) and libdtrace(3LIB).
48 * Please refer to the "Solaris Dynamic Tracing Guide" for more information.
51 #include <sys/dtrace.h>
54 typedef uint32_t pc_t;
56 typedef uintptr_t pc_t;
58 typedef u_long greg_t;
62 * DTrace Implementation Constants and Typedefs
64 #define DTRACE_MAXPROPLEN 128
65 #define DTRACE_DYNVAR_CHUNKSIZE 256
69 struct dtrace_predicate;
71 struct dtrace_provider;
74 typedef struct dtrace_probe dtrace_probe_t;
75 typedef struct dtrace_ecb dtrace_ecb_t;
76 typedef struct dtrace_predicate dtrace_predicate_t;
77 typedef struct dtrace_action dtrace_action_t;
78 typedef struct dtrace_provider dtrace_provider_t;
79 typedef struct dtrace_meta dtrace_meta_t;
80 typedef struct dtrace_state dtrace_state_t;
81 typedef uint32_t dtrace_optid_t;
82 typedef uint32_t dtrace_specid_t;
83 typedef uint64_t dtrace_genid_t;
88 * The probe is the fundamental unit of the DTrace architecture. Probes are
89 * created by DTrace providers, and managed by the DTrace framework. A probe
90 * is identified by a unique <provider, module, function, name> tuple, and has
91 * a unique probe identifier assigned to it. (Some probes are not associated
92 * with a specific point in text; these are called _unanchored probes_ and have
93 * no module or function associated with them.) Probes are represented as a
94 * dtrace_probe structure. To allow quick lookups based on each element of the
95 * probe tuple, probes are hashed by each of provider, module, function and
96 * name. (If a lookup is performed based on a regular expression, a
97 * dtrace_probekey is prepared, and a linear search is performed.) Each probe
98 * is additionally pointed to by a linear array indexed by its identifier. The
99 * identifier is the provider's mechanism for indicating to the DTrace
100 * framework that a probe has fired: the identifier is passed as the first
101 * argument to dtrace_probe(), where it is then mapped into the corresponding
102 * dtrace_probe structure. From the dtrace_probe structure, dtrace_probe() can
103 * iterate over the probe's list of enabling control blocks; see "DTrace
104 * Enabling Control Blocks", below.)
106 struct dtrace_probe {
107 dtrace_id_t dtpr_id; /* probe identifier */
108 dtrace_ecb_t *dtpr_ecb; /* ECB list; see below */
109 dtrace_ecb_t *dtpr_ecb_last; /* last ECB in list */
110 void *dtpr_arg; /* provider argument */
111 dtrace_cacheid_t dtpr_predcache; /* predicate cache ID */
112 int dtpr_aframes; /* artificial frames */
113 dtrace_provider_t *dtpr_provider; /* pointer to provider */
114 char *dtpr_mod; /* probe's module name */
115 char *dtpr_func; /* probe's function name */
116 char *dtpr_name; /* probe's name */
117 dtrace_probe_t *dtpr_nextmod; /* next in module hash */
118 dtrace_probe_t *dtpr_prevmod; /* previous in module hash */
119 dtrace_probe_t *dtpr_nextfunc; /* next in function hash */
120 dtrace_probe_t *dtpr_prevfunc; /* previous in function hash */
121 dtrace_probe_t *dtpr_nextname; /* next in name hash */
122 dtrace_probe_t *dtpr_prevname; /* previous in name hash */
123 dtrace_genid_t dtpr_gen; /* probe generation ID */
126 typedef int dtrace_probekey_f(const char *, const char *, int);
128 typedef struct dtrace_probekey {
129 char *dtpk_prov; /* provider name to match */
130 dtrace_probekey_f *dtpk_pmatch; /* provider matching function */
131 char *dtpk_mod; /* module name to match */
132 dtrace_probekey_f *dtpk_mmatch; /* module matching function */
133 char *dtpk_func; /* func name to match */
134 dtrace_probekey_f *dtpk_fmatch; /* func matching function */
135 char *dtpk_name; /* name to match */
136 dtrace_probekey_f *dtpk_nmatch; /* name matching function */
137 dtrace_id_t dtpk_id; /* identifier to match */
140 typedef struct dtrace_hashbucket {
141 struct dtrace_hashbucket *dthb_next; /* next on hash chain */
142 dtrace_probe_t *dthb_chain; /* chain of probes */
143 int dthb_len; /* number of probes here */
144 } dtrace_hashbucket_t;
146 typedef struct dtrace_hash {
147 dtrace_hashbucket_t **dth_tab; /* hash table */
148 int dth_size; /* size of hash table */
149 int dth_mask; /* mask to index into table */
150 int dth_nbuckets; /* total number of buckets */
151 uintptr_t dth_nextoffs; /* offset of next in probe */
152 uintptr_t dth_prevoffs; /* offset of prev in probe */
153 uintptr_t dth_stroffs; /* offset of str in probe */
157 * DTrace Enabling Control Blocks
159 * When a provider wishes to fire a probe, it calls into dtrace_probe(),
160 * passing the probe identifier as the first argument. As described above,
161 * dtrace_probe() maps the identifier into a pointer to a dtrace_probe_t
162 * structure. This structure contains information about the probe, and a
163 * pointer to the list of Enabling Control Blocks (ECBs). Each ECB points to
164 * DTrace consumer state, and contains an optional predicate, and a list of
165 * actions. (Shown schematically below.) The ECB abstraction allows a single
166 * probe to be multiplexed across disjoint consumers, or across disjoint
167 * enablings of a single probe within one consumer.
169 * Enabling Control Block
171 * +------------------------+
172 * | dtrace_epid_t ---------+--------------> Enabled Probe ID (EPID)
173 * | dtrace_state_t * ------+--------------> State associated with this ECB
174 * | dtrace_predicate_t * --+---------+
175 * | dtrace_action_t * -----+----+ |
176 * | dtrace_ecb_t * ---+ | | | Predicate (if any)
177 * +-------------------+----+ | | dtrace_predicate_t
178 * | | +---> +--------------------+
179 * | | | dtrace_difo_t * ---+----> DIFO
180 * | | +--------------------+
182 * Next ECB | | Action
183 * (if any) | | dtrace_action_t
184 * : +--> +-------------------+
185 * : | dtrace_actkind_t -+------> kind
186 * v | dtrace_difo_t * --+------> DIFO (if any)
187 * | dtrace_recdesc_t -+------> record descr.
188 * | dtrace_action_t * +------+
189 * +-------------------+ |
191 * +-------------------------------+ (if any)
195 * +--> +-------------------+
196 * | dtrace_actkind_t -+------> kind
197 * | dtrace_difo_t * --+------> DIFO (if any)
198 * | dtrace_action_t * +------+
199 * +-------------------+ |
201 * +-------------------------------+ (if any)
207 * dtrace_probe() iterates over the ECB list. If the ECB needs less space
208 * than is available in the principal buffer, the ECB is processed: if the
209 * predicate is non-NULL, the DIF object is executed. If the result is
210 * non-zero, the action list is processed, with each action being executed
211 * accordingly. When the action list has been completely executed, processing
212 * advances to the next ECB. processing advances to the next ECB. If the
213 * result is non-zero; For each ECB, it first determines the The ECB
214 * abstraction allows disjoint consumers to multiplex on single probes.
217 dtrace_epid_t dte_epid; /* enabled probe ID */
218 uint32_t dte_alignment; /* required alignment */
219 size_t dte_needed; /* bytes needed */
220 size_t dte_size; /* total size of payload */
221 dtrace_predicate_t *dte_predicate; /* predicate, if any */
222 dtrace_action_t *dte_action; /* actions, if any */
223 dtrace_ecb_t *dte_next; /* next ECB on probe */
224 dtrace_state_t *dte_state; /* pointer to state */
225 uint32_t dte_cond; /* security condition */
226 dtrace_probe_t *dte_probe; /* pointer to probe */
227 dtrace_action_t *dte_action_last; /* last action on ECB */
228 uint64_t dte_uarg; /* library argument */
231 struct dtrace_predicate {
232 dtrace_difo_t *dtp_difo; /* DIF object */
233 dtrace_cacheid_t dtp_cacheid; /* cache identifier */
234 int dtp_refcnt; /* reference count */
237 struct dtrace_action {
238 dtrace_actkind_t dta_kind; /* kind of action */
239 uint16_t dta_intuple; /* boolean: in aggregation */
240 uint32_t dta_refcnt; /* reference count */
241 dtrace_difo_t *dta_difo; /* pointer to DIFO */
242 dtrace_recdesc_t dta_rec; /* record description */
243 dtrace_action_t *dta_prev; /* previous action */
244 dtrace_action_t *dta_next; /* next action */
247 typedef struct dtrace_aggregation {
248 dtrace_action_t dtag_action; /* action; must be first */
249 dtrace_aggid_t dtag_id; /* identifier */
250 dtrace_ecb_t *dtag_ecb; /* corresponding ECB */
251 dtrace_action_t *dtag_first; /* first action in tuple */
252 uint32_t dtag_base; /* base of aggregation */
253 uint8_t dtag_hasarg; /* boolean: has argument */
254 uint64_t dtag_initial; /* initial value */
255 void (*dtag_aggregate)(uint64_t *, uint64_t, uint64_t);
256 } dtrace_aggregation_t;
261 * Principal buffers, aggregation buffers, and speculative buffers are all
262 * managed with the dtrace_buffer structure. By default, this structure
263 * includes twin data buffers -- dtb_tomax and dtb_xamot -- that serve as the
264 * active and passive buffers, respectively. For speculative buffers,
265 * dtb_xamot will be NULL; for "ring" and "fill" buffers, dtb_xamot will point
266 * to a scratch buffer. For all buffer types, the dtrace_buffer structure is
267 * always allocated on a per-CPU basis; a single dtrace_buffer structure is
268 * never shared among CPUs. (That is, there is never true sharing of the
269 * dtrace_buffer structure; to prevent false sharing of the structure, it must
270 * always be aligned to the coherence granularity -- generally 64 bytes.)
272 * One of the critical design decisions of DTrace is that a given ECB always
273 * stores the same quantity and type of data. This is done to assure that the
274 * only metadata required for an ECB's traced data is the EPID. That is, from
275 * the EPID, the consumer can determine the data layout. (The data buffer
276 * layout is shown schematically below.) By assuring that one can determine
277 * data layout from the EPID, the metadata stream can be separated from the
278 * data stream -- simplifying the data stream enormously.
280 * base of data buffer ---> +------+--------------------+------+
281 * | EPID | data | EPID |
282 * +------+--------+------+----+------+
283 * | data | EPID | data |
284 * +---------------+------+-----------+
286 * +------+--------------------+------+
288 * +------+--------------------+ |
298 * limit of data buffer ---> +----------------------------------+
300 * When evaluating an ECB, dtrace_probe() determines if the ECB's needs of the
301 * principal buffer (both scratch and payload) exceed the available space. If
302 * the ECB's needs exceed available space (and if the principal buffer policy
303 * is the default "switch" policy), the ECB is dropped, the buffer's drop count
304 * is incremented, and processing advances to the next ECB. If the ECB's needs
305 * can be met with the available space, the ECB is processed, but the offset in
306 * the principal buffer is only advanced if the ECB completes processing
309 * When a buffer is to be switched (either because the buffer is the principal
310 * buffer with a "switch" policy or because it is an aggregation buffer), a
311 * cross call is issued to the CPU associated with the buffer. In the cross
312 * call context, interrupts are disabled, and the active and the inactive
313 * buffers are atomically switched. This involves switching the data pointers,
314 * copying the various state fields (offset, drops, errors, etc.) into their
315 * inactive equivalents, and clearing the state fields. Because interrupts are
316 * disabled during this procedure, the switch is guaranteed to appear atomic to
319 * DTrace Ring Buffering
321 * To process a ring buffer correctly, one must know the oldest valid record.
322 * Processing starts at the oldest record in the buffer and continues until
323 * the end of the buffer is reached. Processing then resumes starting with
324 * the record stored at offset 0 in the buffer, and continues until the
325 * youngest record is processed. If trace records are of a fixed-length,
326 * determining the oldest record is trivial:
328 * - If the ring buffer has not wrapped, the oldest record is the record
329 * stored at offset 0.
331 * - If the ring buffer has wrapped, the oldest record is the record stored
332 * at the current offset.
334 * With variable length records, however, just knowing the current offset
335 * doesn't suffice for determining the oldest valid record: assuming that one
336 * allows for arbitrary data, one has no way of searching forward from the
337 * current offset to find the oldest valid record. (That is, one has no way
338 * of separating data from metadata.) It would be possible to simply refuse to
339 * process any data in the ring buffer between the current offset and the
340 * limit, but this leaves (potentially) an enormous amount of otherwise valid
343 * To effect ring buffering, we track two offsets in the buffer: the current
344 * offset and the _wrapped_ offset. If a request is made to reserve some
345 * amount of data, and the buffer has wrapped, the wrapped offset is
346 * incremented until the wrapped offset minus the current offset is greater
347 * than or equal to the reserve request. This is done by repeatedly looking
348 * up the ECB corresponding to the EPID at the current wrapped offset, and
349 * incrementing the wrapped offset by the size of the data payload
350 * corresponding to that ECB. If this offset is greater than or equal to the
351 * limit of the data buffer, the wrapped offset is set to 0. Thus, the
352 * current offset effectively "chases" the wrapped offset around the buffer.
355 * base of data buffer ---> +------+--------------------+------+
356 * | EPID | data | EPID |
357 * +------+--------+------+----+------+
358 * | data | EPID | data |
359 * +---------------+------+-----------+
361 * +------+---------------------------+
363 * current offset ---> +------+---------------------------+
365 * wrapped offset ---> +------+--------------------+------+
366 * | EPID | data | EPID |
367 * +------+--------+------+----+------+
368 * | data | EPID | data |
369 * +---------------+------+-----------+
372 * . ... valid data ... .
375 * +------+-------------+------+------+
376 * | EPID | data | EPID | data |
377 * +------+------------++------+------+
378 * | data, cont. | leftover |
379 * limit of data buffer ---> +-------------------+--------------+
381 * If the amount of requested buffer space exceeds the amount of space
382 * available between the current offset and the end of the buffer:
384 * (1) all words in the data buffer between the current offset and the limit
385 * of the data buffer (marked "leftover", above) are set to
388 * (2) the wrapped offset is set to zero
390 * (3) the iteration process described above occurs until the wrapped offset
391 * is greater than the amount of desired space.
393 * The wrapped offset is implemented by (re-)using the inactive offset.
394 * In a "switch" buffer policy, the inactive offset stores the offset in
395 * the inactive buffer; in a "ring" buffer policy, it stores the wrapped
398 * DTrace Scratch Buffering
400 * Some ECBs may wish to allocate dynamically-sized temporary scratch memory.
401 * To accommodate such requests easily, scratch memory may be allocated in
402 * the buffer beyond the current offset plus the needed memory of the current
403 * ECB. If there isn't sufficient room in the buffer for the requested amount
404 * of scratch space, the allocation fails and an error is generated. Scratch
405 * memory is tracked in the dtrace_mstate_t and is automatically freed when
406 * the ECB ceases processing. Note that ring buffers cannot allocate their
407 * scratch from the principal buffer -- lest they needlessly overwrite older,
408 * valid data. Ring buffers therefore have their own dedicated scratch buffer
409 * from which scratch is allocated.
411 #define DTRACEBUF_RING 0x0001 /* bufpolicy set to "ring" */
412 #define DTRACEBUF_FILL 0x0002 /* bufpolicy set to "fill" */
413 #define DTRACEBUF_NOSWITCH 0x0004 /* do not switch buffer */
414 #define DTRACEBUF_WRAPPED 0x0008 /* ring buffer has wrapped */
415 #define DTRACEBUF_DROPPED 0x0010 /* drops occurred */
416 #define DTRACEBUF_ERROR 0x0020 /* errors occurred */
417 #define DTRACEBUF_FULL 0x0040 /* "fill" buffer is full */
418 #define DTRACEBUF_CONSUMED 0x0080 /* buffer has been consumed */
419 #define DTRACEBUF_INACTIVE 0x0100 /* buffer is not yet active */
421 typedef struct dtrace_buffer {
422 uint64_t dtb_offset; /* current offset in buffer */
423 uint64_t dtb_size; /* size of buffer */
424 uint32_t dtb_flags; /* flags */
425 uint32_t dtb_drops; /* number of drops */
426 caddr_t dtb_tomax; /* active buffer */
427 caddr_t dtb_xamot; /* inactive buffer */
428 uint32_t dtb_xamot_flags; /* inactive flags */
429 uint32_t dtb_xamot_drops; /* drops in inactive buffer */
430 uint64_t dtb_xamot_offset; /* offset in inactive buffer */
431 uint32_t dtb_errors; /* number of errors */
432 uint32_t dtb_xamot_errors; /* errors in inactive buffer */
434 uint64_t dtb_pad1; /* pad out to 64 bytes */
436 uint64_t dtb_switched; /* time of last switch */
437 uint64_t dtb_interval; /* observed switch interval */
438 uint64_t dtb_pad2[6]; /* pad to avoid false sharing */
442 * DTrace Aggregation Buffers
444 * Aggregation buffers use much of the same mechanism as described above
445 * ("DTrace Buffers"). However, because an aggregation is fundamentally a
446 * hash, there exists dynamic metadata associated with an aggregation buffer
447 * that is not associated with other kinds of buffers. This aggregation
448 * metadata is _only_ relevant for the in-kernel implementation of
449 * aggregations; it is not actually relevant to user-level consumers. To do
450 * this, we allocate dynamic aggregation data (hash keys and hash buckets)
451 * starting below the _limit_ of the buffer, and we allocate data from the
452 * _base_ of the buffer. When the aggregation buffer is copied out, _only_ the
453 * data is copied out; the metadata is simply discarded. Schematically,
454 * aggregation buffers look like:
456 * base of data buffer ---> +-------+------+-----------+-------+
457 * | aggid | key | value | aggid |
458 * +-------+------+-----------+-------+
460 * +-------+-------+-----+------------+
461 * | value | aggid | key | value |
462 * +-------+------++-----+------+-----+
463 * | aggid | key | value | |
464 * +-------+------+-------------+ |
474 * | || +------------+
476 * +---------------------+ |
478 * | (dtrace_aggkey structures) |
480 * +----------------------------------+
482 * | (dtrace_aggbuffer structure) |
484 * limit of data buffer ---> +----------------------------------+
487 * As implied above, just as we assure that ECBs always store a constant
488 * amount of data, we assure that a given aggregation -- identified by its
489 * aggregation ID -- always stores data of a constant quantity and type.
490 * As with EPIDs, this allows the aggregation ID to serve as the metadata for a
493 * Note that the size of the dtrace_aggkey structure must be sizeof (uintptr_t)
494 * aligned. (If this the structure changes such that this becomes false, an
495 * assertion will fail in dtrace_aggregate().)
497 typedef struct dtrace_aggkey {
498 uint32_t dtak_hashval; /* hash value */
499 uint32_t dtak_action:4; /* action -- 4 bits */
500 uint32_t dtak_size:28; /* size -- 28 bits */
501 caddr_t dtak_data; /* data pointer */
502 struct dtrace_aggkey *dtak_next; /* next in hash chain */
505 typedef struct dtrace_aggbuffer {
506 uintptr_t dtagb_hashsize; /* number of buckets */
507 uintptr_t dtagb_free; /* free list of keys */
508 dtrace_aggkey_t **dtagb_hash; /* hash table */
509 } dtrace_aggbuffer_t;
512 * DTrace Speculations
514 * Speculations have a per-CPU buffer and a global state. Once a speculation
515 * buffer has been comitted or discarded, it cannot be reused until all CPUs
516 * have taken the same action (commit or discard) on their respective
517 * speculative buffer. However, because DTrace probes may execute in arbitrary
518 * context, other CPUs cannot simply be cross-called at probe firing time to
519 * perform the necessary commit or discard. The speculation states thus
520 * optimize for the case that a speculative buffer is only active on one CPU at
521 * the time of a commit() or discard() -- for if this is the case, other CPUs
522 * need not take action, and the speculation is immediately available for
523 * reuse. If the speculation is active on multiple CPUs, it must be
524 * asynchronously cleaned -- potentially leading to a higher rate of dirty
525 * speculative drops. The speculation states are as follows:
527 * DTRACESPEC_INACTIVE <= Initial state; inactive speculation
528 * DTRACESPEC_ACTIVE <= Allocated, but not yet speculatively traced to
529 * DTRACESPEC_ACTIVEONE <= Speculatively traced to on one CPU
530 * DTRACESPEC_ACTIVEMANY <= Speculatively traced to on more than one CPU
531 * DTRACESPEC_COMMITTING <= Currently being commited on one CPU
532 * DTRACESPEC_COMMITTINGMANY <= Currently being commited on many CPUs
533 * DTRACESPEC_DISCARDING <= Currently being discarded on many CPUs
535 * The state transition diagram is as follows:
537 * +----------------------------------------------------------+
540 * | +-------------------| COMMITTING |<-----------------+ |
541 * | | +------------+ | |
542 * | | copied spec. ^ commit() on | | discard() on
543 * | | into principal | active CPU | | active CPU
546 * +----------+ +--------+ +-----------+
547 * | INACTIVE |---------------->| ACTIVE |--------------->| ACTIVEONE |
548 * +----------+ speculation() +--------+ speculate() +-----------+
550 * | | | discard() | |
551 * | | asynchronously | discard() on | | speculate()
552 * | | cleaned V inactive CPU | | on inactive
553 * | | +------------+ | | CPU
554 * | +-------------------| DISCARDING |<-----------------+ |
556 * | asynchronously ^ |
557 * | copied spec. | discard() |
558 * | into principal +------------------------+ |
560 * +----------------+ commit() +------------+
561 * | COMMITTINGMANY |<----------------------------------| ACTIVEMANY |
562 * +----------------+ +------------+
564 typedef enum dtrace_speculation_state {
565 DTRACESPEC_INACTIVE = 0,
567 DTRACESPEC_ACTIVEONE,
568 DTRACESPEC_ACTIVEMANY,
569 DTRACESPEC_COMMITTING,
570 DTRACESPEC_COMMITTINGMANY,
571 DTRACESPEC_DISCARDING
572 } dtrace_speculation_state_t;
574 typedef struct dtrace_speculation {
575 dtrace_speculation_state_t dtsp_state; /* current speculation state */
576 int dtsp_cleaning; /* non-zero if being cleaned */
577 dtrace_buffer_t *dtsp_buffer; /* speculative buffer */
578 } dtrace_speculation_t;
581 * DTrace Dynamic Variables
583 * The dynamic variable problem is obviously decomposed into two subproblems:
584 * allocating new dynamic storage, and freeing old dynamic storage. The
585 * presence of the second problem makes the first much more complicated -- or
586 * rather, the absence of the second renders the first trivial. This is the
587 * case with aggregations, for which there is effectively no deallocation of
588 * dynamic storage. (Or more accurately, all dynamic storage is deallocated
589 * when a snapshot is taken of the aggregation.) As DTrace dynamic variables
590 * allow for both dynamic allocation and dynamic deallocation, the
591 * implementation of dynamic variables is quite a bit more complicated than
592 * that of their aggregation kin.
594 * We observe that allocating new dynamic storage is tricky only because the
595 * size can vary -- the allocation problem is much easier if allocation sizes
596 * are uniform. We further observe that in D, the size of dynamic variables is
597 * actually _not_ dynamic -- dynamic variable sizes may be determined by static
598 * analysis of DIF text. (This is true even of putatively dynamically-sized
599 * objects like strings and stacks, the sizes of which are dictated by the
600 * "stringsize" and "stackframes" variables, respectively.) We exploit this by
601 * performing this analysis on all DIF before enabling any probes. For each
602 * dynamic load or store, we calculate the dynamically-allocated size plus the
603 * size of the dtrace_dynvar structure plus the storage required to key the
604 * data. For all DIF, we take the largest value and dub it the _chunksize_.
605 * We then divide dynamic memory into two parts: a hash table that is wide
606 * enough to have every chunk in its own bucket, and a larger region of equal
607 * chunksize units. Whenever we wish to dynamically allocate a variable, we
608 * always allocate a single chunk of memory. Depending on the uniformity of
609 * allocation, this will waste some amount of memory -- but it eliminates the
610 * non-determinism inherent in traditional heap fragmentation.
612 * Dynamic objects are allocated by storing a non-zero value to them; they are
613 * deallocated by storing a zero value to them. Dynamic variables are
614 * complicated enormously by being shared between CPUs. In particular,
615 * consider the following scenario:
618 * +---------------------------------+ +---------------------------------+
620 * | allocates dynamic object a[123] | | |
621 * | by storing the value 345 to it | | |
623 * | | | wishing to load from object |
624 * | | | a[123], performs lookup in |
625 * | | | dynamic variable space |
627 * | deallocates object a[123] by | | |
628 * | storing 0 to it | | |
630 * | allocates dynamic object b[567] | | performs load from a[123] |
631 * | by storing the value 789 to it | | |
635 * This is obviously a race in the D program, but there are nonetheless only
636 * two valid values for CPU B's load from a[123]: 345 or 0. Most importantly,
637 * CPU B may _not_ see the value 789 for a[123].
639 * There are essentially two ways to deal with this:
641 * (1) Explicitly spin-lock variables. That is, if CPU B wishes to load
642 * from a[123], it needs to lock a[123] and hold the lock for the
643 * duration that it wishes to manipulate it.
645 * (2) Avoid reusing freed chunks until it is known that no CPU is referring
648 * The implementation of (1) is rife with complexity, because it requires the
649 * user of a dynamic variable to explicitly decree when they are done using it.
650 * Were all variables by value, this perhaps wouldn't be debilitating -- but
651 * dynamic variables of non-scalar types are tracked by reference. That is, if
652 * a dynamic variable is, say, a string, and that variable is to be traced to,
653 * say, the principal buffer, the DIF emulation code returns to the main
654 * dtrace_probe() loop a pointer to the underlying storage, not the contents of
655 * the storage. Further, code calling on DIF emulation would have to be aware
656 * that the DIF emulation has returned a reference to a dynamic variable that
657 * has been potentially locked. The variable would have to be unlocked after
658 * the main dtrace_probe() loop is finished with the variable, and the main
659 * dtrace_probe() loop would have to be careful to not call any further DIF
660 * emulation while the variable is locked to avoid deadlock. More generally,
661 * if one were to implement (1), DIF emulation code dealing with dynamic
662 * variables could only deal with one dynamic variable at a time (lest deadlock
663 * result). To sum, (1) exports too much subtlety to the users of dynamic
664 * variables -- increasing maintenance burden and imposing serious constraints
665 * on future DTrace development.
667 * The implementation of (2) is also complex, but the complexity is more
668 * manageable. We need to be sure that when a variable is deallocated, it is
669 * not placed on a traditional free list, but rather on a _dirty_ list. Once a
670 * variable is on a dirty list, it cannot be found by CPUs performing a
671 * subsequent lookup of the variable -- but it may still be in use by other
672 * CPUs. To assure that all CPUs that may be seeing the old variable have
673 * cleared out of probe context, a dtrace_sync() can be issued. Once the
674 * dtrace_sync() has completed, it can be known that all CPUs are done
675 * manipulating the dynamic variable -- the dirty list can be atomically
676 * appended to the free list. Unfortunately, there's a slight hiccup in this
677 * mechanism: dtrace_sync() may not be issued from probe context. The
678 * dtrace_sync() must be therefore issued asynchronously from non-probe
679 * context. For this we rely on the DTrace cleaner, a cyclic that runs at the
680 * "cleanrate" frequency. To ease this implementation, we define several chunk
683 * - Dirty. Deallocated chunks, not yet cleaned. Not available.
685 * - Rinsing. Formerly dirty chunks that are currently being asynchronously
686 * cleaned. Not available, but will be shortly. Dynamic variable
687 * allocation may not spin or block for availability, however.
689 * - Clean. Clean chunks, ready for allocation -- but not on the free list.
691 * - Free. Available for allocation.
693 * Moreover, to avoid absurd contention, _each_ of these lists is implemented
694 * on a per-CPU basis. This is only for performance, not correctness; chunks
695 * may be allocated from another CPU's free list. The algorithm for allocation
698 * (1) Attempt to atomically allocate from current CPU's free list. If list
699 * is non-empty and allocation is successful, allocation is complete.
701 * (2) If the clean list is non-empty, atomically move it to the free list,
704 * (3) If the dynamic variable space is in the CLEAN state, look for free
705 * and clean lists on other CPUs by setting the current CPU to the next
706 * CPU, and reattempting (1). If the next CPU is the current CPU (that
707 * is, if all CPUs have been checked), atomically switch the state of
708 * the dynamic variable space based on the following:
710 * - If no free chunks were found and no dirty chunks were found,
711 * atomically set the state to EMPTY.
713 * - If dirty chunks were found, atomically set the state to DIRTY.
715 * - If rinsing chunks were found, atomically set the state to RINSING.
717 * (4) Based on state of dynamic variable space state, increment appropriate
718 * counter to indicate dynamic drops (if in EMPTY state) vs. dynamic
719 * dirty drops (if in DIRTY state) vs. dynamic rinsing drops (if in
720 * RINSING state). Fail the allocation.
722 * The cleaning cyclic operates with the following algorithm: for all CPUs
723 * with a non-empty dirty list, atomically move the dirty list to the rinsing
724 * list. Perform a dtrace_sync(). For all CPUs with a non-empty rinsing list,
725 * atomically move the rinsing list to the clean list. Perform another
726 * dtrace_sync(). By this point, all CPUs have seen the new clean list; the
727 * state of the dynamic variable space can be restored to CLEAN.
729 * There exist two final races that merit explanation. The first is a simple
733 * +---------------------------------+ +---------------------------------+
735 * | allocates dynamic object a[123] | | allocates dynamic object a[123] |
736 * | by storing the value 345 to it | | by storing the value 567 to it |
741 * Again, this is a race in the D program. It can be resolved by having a[123]
742 * hold the value 345 or a[123] hold the value 567 -- but it must be true that
743 * a[123] have only _one_ of these values. (That is, the racing CPUs may not
744 * put the same element twice on the same hash chain.) This is resolved
745 * simply: before the allocation is undertaken, the start of the new chunk's
746 * hash chain is noted. Later, after the allocation is complete, the hash
747 * chain is atomically switched to point to the new element. If this fails
748 * (because of either concurrent allocations or an allocation concurrent with a
749 * deletion), the newly allocated chunk is deallocated to the dirty list, and
750 * the whole process of looking up (and potentially allocating) the dynamic
751 * variable is reattempted.
753 * The final race is a simple deallocation race:
756 * +---------------------------------+ +---------------------------------+
758 * | deallocates dynamic object | | deallocates dynamic object |
759 * | a[123] by storing the value 0 | | a[123] by storing the value 0 |
760 * | to it | | to it |
765 * Once again, this is a race in the D program, but it is one that we must
766 * handle without corrupting the underlying data structures. Because
767 * deallocations require the deletion of a chunk from the middle of a hash
768 * chain, we cannot use a single-word atomic operation to remove it. For this,
769 * we add a spin lock to the hash buckets that is _only_ used for deallocations
770 * (allocation races are handled as above). Further, this spin lock is _only_
771 * held for the duration of the delete; before control is returned to the DIF
772 * emulation code, the hash bucket is unlocked.
774 typedef struct dtrace_key {
775 uint64_t dttk_value; /* data value or data pointer */
776 uint64_t dttk_size; /* 0 if by-val, >0 if by-ref */
779 typedef struct dtrace_tuple {
780 uint32_t dtt_nkeys; /* number of keys in tuple */
781 uint32_t dtt_pad; /* padding */
782 dtrace_key_t dtt_key[1]; /* array of tuple keys */
785 typedef struct dtrace_dynvar {
786 uint64_t dtdv_hashval; /* hash value -- 0 if free */
787 struct dtrace_dynvar *dtdv_next; /* next on list or hash chain */
788 void *dtdv_data; /* pointer to data */
789 dtrace_tuple_t dtdv_tuple; /* tuple key */
792 typedef enum dtrace_dynvar_op {
794 DTRACE_DYNVAR_NOALLOC,
795 DTRACE_DYNVAR_DEALLOC
796 } dtrace_dynvar_op_t;
798 typedef struct dtrace_dynhash {
799 dtrace_dynvar_t *dtdh_chain; /* hash chain for this bucket */
800 uintptr_t dtdh_lock; /* deallocation lock */
802 uintptr_t dtdh_pad[6]; /* pad to avoid false sharing */
804 uintptr_t dtdh_pad[14]; /* pad to avoid false sharing */
808 typedef struct dtrace_dstate_percpu {
809 dtrace_dynvar_t *dtdsc_free; /* free list for this CPU */
810 dtrace_dynvar_t *dtdsc_dirty; /* dirty list for this CPU */
811 dtrace_dynvar_t *dtdsc_rinsing; /* rinsing list for this CPU */
812 dtrace_dynvar_t *dtdsc_clean; /* clean list for this CPU */
813 uint64_t dtdsc_drops; /* number of capacity drops */
814 uint64_t dtdsc_dirty_drops; /* number of dirty drops */
815 uint64_t dtdsc_rinsing_drops; /* number of rinsing drops */
817 uint64_t dtdsc_pad; /* pad to avoid false sharing */
819 uint64_t dtdsc_pad[2]; /* pad to avoid false sharing */
821 } dtrace_dstate_percpu_t;
823 typedef enum dtrace_dstate_state {
824 DTRACE_DSTATE_CLEAN = 0,
827 DTRACE_DSTATE_RINSING
828 } dtrace_dstate_state_t;
830 typedef struct dtrace_dstate {
831 void *dtds_base; /* base of dynamic var. space */
832 size_t dtds_size; /* size of dynamic var. space */
833 size_t dtds_hashsize; /* number of buckets in hash */
834 size_t dtds_chunksize; /* size of each chunk */
835 dtrace_dynhash_t *dtds_hash; /* pointer to hash table */
836 dtrace_dstate_state_t dtds_state; /* current dynamic var. state */
837 dtrace_dstate_percpu_t *dtds_percpu; /* per-CPU dyn. var. state */
841 * DTrace Variable State
843 * The DTrace variable state tracks user-defined variables in its dtrace_vstate
844 * structure. Each DTrace consumer has exactly one dtrace_vstate structure,
845 * but some dtrace_vstate structures may exist without a corresponding DTrace
846 * consumer (see "DTrace Helpers", below). As described in <sys/dtrace.h>,
847 * user-defined variables can have one of three scopes:
849 * DIFV_SCOPE_GLOBAL => global scope
850 * DIFV_SCOPE_THREAD => thread-local scope (i.e. "self->" variables)
851 * DIFV_SCOPE_LOCAL => clause-local scope (i.e. "this->" variables)
853 * The variable state tracks variables by both their scope and their allocation
856 * - The dtvs_globals and dtvs_locals members each point to an array of
857 * dtrace_statvar structures. These structures contain both the variable
858 * metadata (dtrace_difv structures) and the underlying storage for all
859 * statically allocated variables, including statically allocated
860 * DIFV_SCOPE_GLOBAL variables and all DIFV_SCOPE_LOCAL variables.
862 * - The dtvs_tlocals member points to an array of dtrace_difv structures for
863 * DIFV_SCOPE_THREAD variables. As such, this array tracks _only_ the
864 * variable metadata for DIFV_SCOPE_THREAD variables; the underlying storage
865 * is allocated out of the dynamic variable space.
867 * - The dtvs_dynvars member is the dynamic variable state associated with the
868 * variable state. The dynamic variable state (described in "DTrace Dynamic
869 * Variables", above) tracks all DIFV_SCOPE_THREAD variables and all
870 * dynamically-allocated DIFV_SCOPE_GLOBAL variables.
872 typedef struct dtrace_statvar {
873 uint64_t dtsv_data; /* data or pointer to it */
874 size_t dtsv_size; /* size of pointed-to data */
875 int dtsv_refcnt; /* reference count */
876 dtrace_difv_t dtsv_var; /* variable metadata */
879 typedef struct dtrace_vstate {
880 dtrace_state_t *dtvs_state; /* back pointer to state */
881 dtrace_statvar_t **dtvs_globals; /* statically-allocated glbls */
882 int dtvs_nglobals; /* number of globals */
883 dtrace_difv_t *dtvs_tlocals; /* thread-local metadata */
884 int dtvs_ntlocals; /* number of thread-locals */
885 dtrace_statvar_t **dtvs_locals; /* clause-local data */
886 int dtvs_nlocals; /* number of clause-locals */
887 dtrace_dstate_t dtvs_dynvars; /* dynamic variable state */
891 * DTrace Machine State
893 * In the process of processing a fired probe, DTrace needs to track and/or
894 * cache some per-CPU state associated with that particular firing. This is
895 * state that is always discarded after the probe firing has completed, and
896 * much of it is not specific to any DTrace consumer, remaining valid across
897 * all ECBs. This state is tracked in the dtrace_mstate structure.
899 #define DTRACE_MSTATE_ARGS 0x00000001
900 #define DTRACE_MSTATE_PROBE 0x00000002
901 #define DTRACE_MSTATE_EPID 0x00000004
902 #define DTRACE_MSTATE_TIMESTAMP 0x00000008
903 #define DTRACE_MSTATE_STACKDEPTH 0x00000010
904 #define DTRACE_MSTATE_CALLER 0x00000020
905 #define DTRACE_MSTATE_IPL 0x00000040
906 #define DTRACE_MSTATE_FLTOFFS 0x00000080
907 #define DTRACE_MSTATE_WALLTIMESTAMP 0x00000100
908 #define DTRACE_MSTATE_USTACKDEPTH 0x00000200
909 #define DTRACE_MSTATE_UCALLER 0x00000400
911 typedef struct dtrace_mstate {
912 uintptr_t dtms_scratch_base; /* base of scratch space */
913 uintptr_t dtms_scratch_ptr; /* current scratch pointer */
914 size_t dtms_scratch_size; /* scratch size */
915 uint32_t dtms_present; /* variables that are present */
916 uint64_t dtms_arg[5]; /* cached arguments */
917 dtrace_epid_t dtms_epid; /* current EPID */
918 uint64_t dtms_timestamp; /* cached timestamp */
919 hrtime_t dtms_walltimestamp; /* cached wall timestamp */
920 int dtms_stackdepth; /* cached stackdepth */
921 int dtms_ustackdepth; /* cached ustackdepth */
922 struct dtrace_probe *dtms_probe; /* current probe */
923 uintptr_t dtms_caller; /* cached caller */
924 uint64_t dtms_ucaller; /* cached user-level caller */
925 int dtms_ipl; /* cached interrupt pri lev */
926 int dtms_fltoffs; /* faulting DIFO offset */
927 uintptr_t dtms_strtok; /* saved strtok() pointer */
928 uint32_t dtms_access; /* memory access rights */
929 dtrace_difo_t *dtms_difo; /* current dif object */
932 #define DTRACE_COND_OWNER 0x1
933 #define DTRACE_COND_USERMODE 0x2
934 #define DTRACE_COND_ZONEOWNER 0x4
936 #define DTRACE_PROBEKEY_MAXDEPTH 8 /* max glob recursion depth */
939 * Access flag used by dtrace_mstate.dtms_access.
941 #define DTRACE_ACCESS_KERNEL 0x1 /* the priv to read kmem */
947 * Each DTrace consumer is in one of several states, which (for purposes of
948 * avoiding yet-another overloading of the noun "state") we call the current
949 * _activity_. The activity transitions on dtrace_go() (from DTRACIOCGO), on
950 * dtrace_stop() (from DTRACIOCSTOP) and on the exit() action. Activities may
951 * only transition in one direction; the activity transition diagram is a
952 * directed acyclic graph. The activity transition diagram is as follows:
955 * +----------+ +--------+ +--------+
956 * | INACTIVE |------------------>| WARMUP |------------------>| ACTIVE |
957 * +----------+ dtrace_go(), +--------+ dtrace_go(), +--------+
958 * before BEGIN | after BEGIN | | |
960 * exit() action | | | |
961 * from BEGIN ECB | | | |
964 * +----------+ exit() action | | |
965 * +-----------------------------| DRAINING |<-------------------+ | |
968 * | dtrace_stop(), | | |
972 * | +---------+ +----------+ | |
973 * | | STOPPED |<----------------| COOLDOWN |<----------------------+ |
974 * | +---------+ dtrace_stop(), +----------+ dtrace_stop(), |
975 * | after END before END |
978 * +----------------------------->| KILLED |<--------------------------+
979 * deadman timeout or +--------+ deadman timeout or
980 * killed consumer killed consumer
982 * Note that once a DTrace consumer has stopped tracing, there is no way to
983 * restart it; if a DTrace consumer wishes to restart tracing, it must reopen
984 * the DTrace pseudodevice.
986 typedef enum dtrace_activity {
987 DTRACE_ACTIVITY_INACTIVE = 0, /* not yet running */
988 DTRACE_ACTIVITY_WARMUP, /* while starting */
989 DTRACE_ACTIVITY_ACTIVE, /* running */
990 DTRACE_ACTIVITY_DRAINING, /* before stopping */
991 DTRACE_ACTIVITY_COOLDOWN, /* while stopping */
992 DTRACE_ACTIVITY_STOPPED, /* after stopping */
993 DTRACE_ACTIVITY_KILLED /* killed */
997 * DTrace Helper Implementation
999 * A description of the helper architecture may be found in <sys/dtrace.h>.
1000 * Each process contains a pointer to its helpers in its p_dtrace_helpers
1001 * member. This is a pointer to a dtrace_helpers structure, which contains an
1002 * array of pointers to dtrace_helper structures, helper variable state (shared
1003 * among a process's helpers) and a generation count. (The generation count is
1004 * used to provide an identifier when a helper is added so that it may be
1005 * subsequently removed.) The dtrace_helper structure is self-explanatory,
1006 * containing pointers to the objects needed to execute the helper. Note that
1007 * helpers are _duplicated_ across fork(2), and destroyed on exec(2). No more
1008 * than dtrace_helpers_max are allowed per-process.
1010 #define DTRACE_HELPER_ACTION_USTACK 0
1011 #define DTRACE_NHELPER_ACTIONS 1
1013 typedef struct dtrace_helper_action {
1014 int dtha_generation; /* helper action generation */
1015 int dtha_nactions; /* number of actions */
1016 dtrace_difo_t *dtha_predicate; /* helper action predicate */
1017 dtrace_difo_t **dtha_actions; /* array of actions */
1018 struct dtrace_helper_action *dtha_next; /* next helper action */
1019 } dtrace_helper_action_t;
1021 typedef struct dtrace_helper_provider {
1022 int dthp_generation; /* helper provider generation */
1023 uint32_t dthp_ref; /* reference count */
1024 dof_helper_t dthp_prov; /* DOF w/ provider and probes */
1025 } dtrace_helper_provider_t;
1027 typedef struct dtrace_helpers {
1028 dtrace_helper_action_t **dthps_actions; /* array of helper actions */
1029 dtrace_vstate_t dthps_vstate; /* helper action var. state */
1030 dtrace_helper_provider_t **dthps_provs; /* array of providers */
1031 uint_t dthps_nprovs; /* count of providers */
1032 uint_t dthps_maxprovs; /* provider array size */
1033 int dthps_generation; /* current generation */
1034 pid_t dthps_pid; /* pid of associated proc */
1035 int dthps_deferred; /* helper in deferred list */
1036 struct dtrace_helpers *dthps_next; /* next pointer */
1037 struct dtrace_helpers *dthps_prev; /* prev pointer */
1041 * DTrace Helper Action Tracing
1043 * Debugging helper actions can be arduous. To ease the development and
1044 * debugging of helpers, DTrace contains a tracing-framework-within-a-tracing-
1045 * framework: helper tracing. If dtrace_helptrace_enabled is non-zero (which
1046 * it is by default on DEBUG kernels), all helper activity will be traced to a
1047 * global, in-kernel ring buffer. Each entry includes a pointer to the specific
1048 * helper, the location within the helper, and a trace of all local variables.
1049 * The ring buffer may be displayed in a human-readable format with the
1050 * ::dtrace_helptrace mdb(1) dcmd.
1052 #define DTRACE_HELPTRACE_NEXT (-1)
1053 #define DTRACE_HELPTRACE_DONE (-2)
1054 #define DTRACE_HELPTRACE_ERR (-3)
1056 typedef struct dtrace_helptrace {
1057 dtrace_helper_action_t *dtht_helper; /* helper action */
1058 int dtht_where; /* where in helper action */
1059 int dtht_nlocals; /* number of locals */
1060 int dtht_fault; /* type of fault (if any) */
1061 int dtht_fltoffs; /* DIF offset */
1062 uint64_t dtht_illval; /* faulting value */
1063 uint64_t dtht_locals[1]; /* local variables */
1064 } dtrace_helptrace_t;
1067 * DTrace Credentials
1069 * In probe context, we have limited flexibility to examine the credentials
1070 * of the DTrace consumer that created a particular enabling. We use
1071 * the Least Privilege interfaces to cache the consumer's cred pointer and
1072 * some facts about that credential in a dtrace_cred_t structure. These
1073 * can limit the consumer's breadth of visibility and what actions the
1074 * consumer may take.
1076 #define DTRACE_CRV_ALLPROC 0x01
1077 #define DTRACE_CRV_KERNEL 0x02
1078 #define DTRACE_CRV_ALLZONE 0x04
1080 #define DTRACE_CRV_ALL (DTRACE_CRV_ALLPROC | DTRACE_CRV_KERNEL | \
1083 #define DTRACE_CRA_PROC 0x0001
1084 #define DTRACE_CRA_PROC_CONTROL 0x0002
1085 #define DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER 0x0004
1086 #define DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE 0x0008
1087 #define DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG 0x0010
1088 #define DTRACE_CRA_KERNEL 0x0020
1089 #define DTRACE_CRA_KERNEL_DESTRUCTIVE 0x0040
1091 #define DTRACE_CRA_ALL (DTRACE_CRA_PROC | \
1092 DTRACE_CRA_PROC_CONTROL | \
1093 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER | \
1094 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE | \
1095 DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG | \
1096 DTRACE_CRA_KERNEL | \
1097 DTRACE_CRA_KERNEL_DESTRUCTIVE)
1099 typedef struct dtrace_cred {
1101 uint8_t dcr_destructive;
1102 uint8_t dcr_visible;
1103 uint16_t dcr_action;
1107 * DTrace Consumer State
1109 * Each DTrace consumer has an associated dtrace_state structure that contains
1110 * its in-kernel DTrace state -- including options, credentials, statistics and
1111 * pointers to ECBs, buffers, speculations and formats. A dtrace_state
1112 * structure is also allocated for anonymous enablings. When anonymous state
1113 * is grabbed, the grabbing consumers dts_anon pointer is set to the grabbed
1114 * dtrace_state structure.
1116 struct dtrace_state {
1118 dev_t dts_dev; /* device */
1120 struct cdev *dts_dev; /* device */
1122 int dts_necbs; /* total number of ECBs */
1123 dtrace_ecb_t **dts_ecbs; /* array of ECBs */
1124 dtrace_epid_t dts_epid; /* next EPID to allocate */
1125 size_t dts_needed; /* greatest needed space */
1126 struct dtrace_state *dts_anon; /* anon. state, if grabbed */
1127 dtrace_activity_t dts_activity; /* current activity */
1128 dtrace_vstate_t dts_vstate; /* variable state */
1129 dtrace_buffer_t *dts_buffer; /* principal buffer */
1130 dtrace_buffer_t *dts_aggbuffer; /* aggregation buffer */
1131 dtrace_speculation_t *dts_speculations; /* speculation array */
1132 int dts_nspeculations; /* number of speculations */
1133 int dts_naggregations; /* number of aggregations */
1134 dtrace_aggregation_t **dts_aggregations; /* aggregation array */
1136 vmem_t *dts_aggid_arena; /* arena for aggregation IDs */
1138 struct unrhdr *dts_aggid_arena; /* arena for aggregation IDs */
1140 uint64_t dts_errors; /* total number of errors */
1141 uint32_t dts_speculations_busy; /* number of spec. busy */
1142 uint32_t dts_speculations_unavail; /* number of spec unavail */
1143 uint32_t dts_stkstroverflows; /* stack string tab overflows */
1144 uint32_t dts_dblerrors; /* errors in ERROR probes */
1145 uint32_t dts_reserve; /* space reserved for END */
1146 hrtime_t dts_laststatus; /* time of last status */
1148 cyclic_id_t dts_cleaner; /* cleaning cyclic */
1149 cyclic_id_t dts_deadman; /* deadman cyclic */
1151 struct callout dts_cleaner; /* Cleaning callout. */
1152 struct callout dts_deadman; /* Deadman callout. */
1154 hrtime_t dts_alive; /* time last alive */
1155 char dts_speculates; /* boolean: has speculations */
1156 char dts_destructive; /* boolean: has dest. actions */
1157 int dts_nformats; /* number of formats */
1158 char **dts_formats; /* format string array */
1159 dtrace_optval_t dts_options[DTRACEOPT_MAX]; /* options */
1160 dtrace_cred_t dts_cred; /* credentials */
1161 size_t dts_nretained; /* number of retained enabs */
1164 struct dtrace_provider {
1165 dtrace_pattr_t dtpv_attr; /* provider attributes */
1166 dtrace_ppriv_t dtpv_priv; /* provider privileges */
1167 dtrace_pops_t dtpv_pops; /* provider operations */
1168 char *dtpv_name; /* provider name */
1169 void *dtpv_arg; /* provider argument */
1170 hrtime_t dtpv_defunct; /* when made defunct */
1171 struct dtrace_provider *dtpv_next; /* next provider */
1174 struct dtrace_meta {
1175 dtrace_mops_t dtm_mops; /* meta provider operations */
1176 char *dtm_name; /* meta provider name */
1177 void *dtm_arg; /* meta provider user arg */
1178 uint64_t dtm_count; /* no. of associated provs. */
1184 * A dtrace_enabling structure is used to track a collection of ECB
1185 * descriptions -- before they have been turned into actual ECBs. This is
1186 * created as a result of DOF processing, and is generally used to generate
1187 * ECBs immediately thereafter. However, enablings are also generally
1188 * retained should the probes they describe be created at a later time; as
1189 * each new module or provider registers with the framework, the retained
1190 * enablings are reevaluated, with any new match resulting in new ECBs. To
1191 * prevent probes from being matched more than once, the enabling tracks the
1192 * last probe generation matched, and only matches probes from subsequent
1195 typedef struct dtrace_enabling {
1196 dtrace_ecbdesc_t **dten_desc; /* all ECB descriptions */
1197 int dten_ndesc; /* number of ECB descriptions */
1198 int dten_maxdesc; /* size of ECB array */
1199 dtrace_vstate_t *dten_vstate; /* associated variable state */
1200 dtrace_genid_t dten_probegen; /* matched probe generation */
1201 dtrace_ecbdesc_t *dten_current; /* current ECB description */
1202 int dten_error; /* current error value */
1203 int dten_primed; /* boolean: set if primed */
1204 struct dtrace_enabling *dten_prev; /* previous enabling */
1205 struct dtrace_enabling *dten_next; /* next enabling */
1206 } dtrace_enabling_t;
1209 * DTrace Anonymous Enablings
1211 * Anonymous enablings are DTrace enablings that are not associated with a
1212 * controlling process, but rather derive their enabling from DOF stored as
1213 * properties in the dtrace.conf file. If there is an anonymous enabling, a
1214 * DTrace consumer state and enabling are created on attach. The state may be
1215 * subsequently grabbed by the first consumer specifying the "grabanon"
1216 * option. As long as an anonymous DTrace enabling exists, dtrace(7D) will
1219 typedef struct dtrace_anon {
1220 dtrace_state_t *dta_state; /* DTrace consumer state */
1221 dtrace_enabling_t *dta_enabling; /* pointer to enabling */
1222 processorid_t dta_beganon; /* which CPU BEGIN ran on */
1226 * DTrace Error Debugging
1229 #define DTRACE_ERRDEBUG
1232 #ifdef DTRACE_ERRDEBUG
1234 typedef struct dtrace_errhash {
1235 const char *dter_msg; /* error message */
1236 int dter_count; /* number of times seen */
1239 #define DTRACE_ERRHASHSZ 256 /* must be > number of err msgs */
1241 #endif /* DTRACE_ERRDEBUG */
1244 * DTrace Toxic Ranges
1246 * DTrace supports safe loads from probe context; if the address turns out to
1247 * be invalid, a bit will be set by the kernel indicating that DTrace
1248 * encountered a memory error, and DTrace will propagate the error to the user
1249 * accordingly. However, there may exist some regions of memory in which an
1250 * arbitrary load can change system state, and from which it is impossible to
1251 * recover from such a load after it has been attempted. Examples of this may
1252 * include memory in which programmable I/O registers are mapped (for which a
1253 * read may have some implications for the device) or (in the specific case of
1254 * UltraSPARC-I and -II) the virtual address hole. The platform is required
1255 * to make DTrace aware of these toxic ranges; DTrace will then check that
1256 * target addresses are not in a toxic range before attempting to issue a
1259 typedef struct dtrace_toxrange {
1260 uintptr_t dtt_base; /* base of toxic range */
1261 uintptr_t dtt_limit; /* limit of toxic range */
1262 } dtrace_toxrange_t;
1264 extern uint64_t dtrace_getarg(int, int);
1265 extern greg_t dtrace_getfp(void);
1266 extern int dtrace_getipl(void);
1267 extern uintptr_t dtrace_caller(int);
1268 extern uint32_t dtrace_cas32(uint32_t *, uint32_t, uint32_t);
1269 extern void *dtrace_casptr(volatile void *, volatile void *, volatile void *);
1270 extern void dtrace_copyin(uintptr_t, uintptr_t, size_t, volatile uint16_t *);
1271 extern void dtrace_copyinstr(uintptr_t, uintptr_t, size_t, volatile uint16_t *);
1272 extern void dtrace_copyout(uintptr_t, uintptr_t, size_t, volatile uint16_t *);
1273 extern void dtrace_copyoutstr(uintptr_t, uintptr_t, size_t,
1274 volatile uint16_t *);
1275 extern void dtrace_getpcstack(pc_t *, int, int, uint32_t *);
1276 extern ulong_t dtrace_getreg(struct trapframe *, uint_t);
1277 extern int dtrace_getstackdepth(int);
1278 extern void dtrace_getupcstack(uint64_t *, int);
1279 extern void dtrace_getufpstack(uint64_t *, uint64_t *, int);
1280 extern int dtrace_getustackdepth(void);
1281 extern uintptr_t dtrace_fulword(void *);
1282 extern uint8_t dtrace_fuword8(void *);
1283 extern uint16_t dtrace_fuword16(void *);
1284 extern uint32_t dtrace_fuword32(void *);
1285 extern uint64_t dtrace_fuword64(void *);
1286 extern void dtrace_probe_error(dtrace_state_t *, dtrace_epid_t, int, int,
1288 extern int dtrace_assfail(const char *, const char *, int);
1289 extern int dtrace_attached(void);
1291 extern hrtime_t dtrace_gethrestime(void);
1295 extern void dtrace_flush_windows(void);
1296 extern void dtrace_flush_user_windows(void);
1297 extern uint_t dtrace_getotherwin(void);
1298 extern uint_t dtrace_getfprs(void);
1300 extern void dtrace_copy(uintptr_t, uintptr_t, size_t);
1301 extern void dtrace_copystr(uintptr_t, uintptr_t, size_t, volatile uint16_t *);
1307 * DTrace calls ASSERT from probe context. To assure that a failed ASSERT
1308 * does not induce a markedly more catastrophic failure (e.g., one from which
1309 * a dump cannot be gleaned), DTrace must define its own ASSERT to be one that
1310 * may safely be called from probe context. This header file must thus be
1311 * included by any DTrace component that calls ASSERT from probe context, and
1312 * _only_ by those components. (The only exception to this is kernel
1313 * debugging infrastructure at user-level that doesn't depend on calling
1318 #define ASSERT(EX) ((void)((EX) || \
1319 dtrace_assfail(#EX, __FILE__, __LINE__)))
1321 #define ASSERT(X) ((void)0)
1328 #endif /* _SYS_DTRACE_IMPL_H */