xref: /illumos-gate/usr/src/uts/common/sys/dtrace_impl.h (revision 814a60b13c0ad90e5d2edfd29a7a84bbf416cc1a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #ifndef _SYS_DTRACE_IMPL_H
28 #define	_SYS_DTRACE_IMPL_H
29 
30 #pragma ident	"%Z%%M%	%I%	%E% SMI"
31 
32 #ifdef	__cplusplus
33 extern "C" {
34 #endif
35 
36 /*
37  * DTrace Dynamic Tracing Software: Kernel Implementation Interfaces
38  *
39  * Note: The contents of this file are private to the implementation of the
40  * Solaris system and DTrace subsystem and are subject to change at any time
41  * without notice.  Applications and drivers using these interfaces will fail
42  * to run on future releases.  These interfaces should not be used for any
43  * purpose except those expressly outlined in dtrace(7D) and libdtrace(3LIB).
44  * Please refer to the "Solaris Dynamic Tracing Guide" for more information.
45  */
46 
47 #include <sys/dtrace.h>
48 
49 /*
50  * DTrace Implementation Constants and Typedefs
51  */
52 #define	DTRACE_MAXPROPLEN		128
53 #define	DTRACE_DYNVAR_CHUNKSIZE		256
54 
55 struct dtrace_probe;
56 struct dtrace_ecb;
57 struct dtrace_predicate;
58 struct dtrace_action;
59 struct dtrace_provider;
60 struct dtrace_state;
61 
62 typedef struct dtrace_probe dtrace_probe_t;
63 typedef struct dtrace_ecb dtrace_ecb_t;
64 typedef struct dtrace_predicate dtrace_predicate_t;
65 typedef struct dtrace_action dtrace_action_t;
66 typedef struct dtrace_provider dtrace_provider_t;
67 typedef struct dtrace_meta dtrace_meta_t;
68 typedef struct dtrace_state dtrace_state_t;
69 typedef uint32_t dtrace_optid_t;
70 typedef uint32_t dtrace_specid_t;
71 typedef uint64_t dtrace_genid_t;
72 
73 /*
74  * DTrace Probes
75  *
76  * The probe is the fundamental unit of the DTrace architecture.  Probes are
77  * created by DTrace providers, and managed by the DTrace framework.  A probe
78  * is identified by a unique <provider, module, function, name> tuple, and has
79  * a unique probe identifier assigned to it.  (Some probes are not associated
80  * with a specific point in text; these are called _unanchored probes_ and have
81  * no module or function associated with them.)  Probes are represented as a
82  * dtrace_probe structure.  To allow quick lookups based on each element of the
83  * probe tuple, probes are hashed by each of provider, module, function and
84  * name.  (If a lookup is performed based on a regular expression, a
85  * dtrace_probekey is prepared, and a linear search is performed.) Each probe
86  * is additionally pointed to by a linear array indexed by its identifier.  The
87  * identifier is the provider's mechanism for indicating to the DTrace
88  * framework that a probe has fired:  the identifier is passed as the first
89  * argument to dtrace_probe(), where it is then mapped into the corresponding
90  * dtrace_probe structure.  From the dtrace_probe structure, dtrace_probe() can
91  * iterate over the probe's list of enabling control blocks; see "DTrace
92  * Enabling Control Blocks", below.)
93  */
94 struct dtrace_probe {
95 	dtrace_id_t dtpr_id;			/* probe identifier */
96 	dtrace_ecb_t *dtpr_ecb;			/* ECB list; see below */
97 	dtrace_ecb_t *dtpr_ecb_last;		/* last ECB in list */
98 	void *dtpr_arg;				/* provider argument */
99 	dtrace_cacheid_t dtpr_predcache;	/* predicate cache ID */
100 	int dtpr_aframes;			/* artificial frames */
101 	dtrace_provider_t *dtpr_provider;	/* pointer to provider */
102 	char *dtpr_mod;				/* probe's module name */
103 	char *dtpr_func;			/* probe's function name */
104 	char *dtpr_name;			/* probe's name */
105 	dtrace_probe_t *dtpr_nextmod;		/* next in module hash */
106 	dtrace_probe_t *dtpr_prevmod;		/* previous in module hash */
107 	dtrace_probe_t *dtpr_nextfunc;		/* next in function hash */
108 	dtrace_probe_t *dtpr_prevfunc;		/* previous in function hash */
109 	dtrace_probe_t *dtpr_nextname;		/* next in name hash */
110 	dtrace_probe_t *dtpr_prevname;		/* previous in name hash */
111 	dtrace_genid_t dtpr_gen;		/* probe generation ID */
112 };
113 
114 typedef int dtrace_probekey_f(const char *, const char *, int);
115 
116 typedef struct dtrace_probekey {
117 	const char *dtpk_prov;			/* provider name to match */
118 	dtrace_probekey_f *dtpk_pmatch;		/* provider matching function */
119 	const char *dtpk_mod;			/* module name to match */
120 	dtrace_probekey_f *dtpk_mmatch;		/* module matching function */
121 	const char *dtpk_func;			/* func name to match */
122 	dtrace_probekey_f *dtpk_fmatch;		/* func matching function */
123 	const char *dtpk_name;			/* name to match */
124 	dtrace_probekey_f *dtpk_nmatch;		/* name matching function */
125 	dtrace_id_t dtpk_id;			/* identifier to match */
126 } dtrace_probekey_t;
127 
128 typedef struct dtrace_hashbucket {
129 	struct dtrace_hashbucket *dthb_next;	/* next on hash chain */
130 	dtrace_probe_t *dthb_chain;		/* chain of probes */
131 	int dthb_len;				/* number of probes here */
132 } dtrace_hashbucket_t;
133 
134 typedef struct dtrace_hash {
135 	dtrace_hashbucket_t **dth_tab;		/* hash table */
136 	int dth_size;				/* size of hash table */
137 	int dth_mask;				/* mask to index into table */
138 	int dth_nbuckets;			/* total number of buckets */
139 	uintptr_t dth_nextoffs;			/* offset of next in probe */
140 	uintptr_t dth_prevoffs;			/* offset of prev in probe */
141 	uintptr_t dth_stroffs;			/* offset of str in probe */
142 } dtrace_hash_t;
143 
144 /*
145  * DTrace Enabling Control Blocks
146  *
147  * When a provider wishes to fire a probe, it calls into dtrace_probe(),
148  * passing the probe identifier as the first argument.  As described above,
149  * dtrace_probe() maps the identifier into a pointer to a dtrace_probe_t
150  * structure.  This structure contains information about the probe, and a
151  * pointer to the list of Enabling Control Blocks (ECBs).  Each ECB points to
152  * DTrace consumer state, and contains an optional predicate, and a list of
153  * actions.  (Shown schematically below.)  The ECB abstraction allows a single
154  * probe to be multiplexed across disjoint consumers, or across disjoint
155  * enablings of a single probe within one consumer.
156  *
157  *   Enabling Control Block
158  *        dtrace_ecb_t
159  * +------------------------+
160  * | dtrace_epid_t ---------+--------------> Enabled Probe ID (EPID)
161  * | dtrace_state_t * ------+--------------> State associated with this ECB
162  * | dtrace_predicate_t * --+---------+
163  * | dtrace_action_t * -----+----+    |
164  * | dtrace_ecb_t * ---+    |    |    |       Predicate (if any)
165  * +-------------------+----+    |    |       dtrace_predicate_t
166  *                     |         |    +---> +--------------------+
167  *                     |         |          | dtrace_difo_t * ---+----> DIFO
168  *                     |         |          +--------------------+
169  *                     |         |
170  *            Next ECB |         |           Action
171  *            (if any) |         |       dtrace_action_t
172  *                     :         +--> +-------------------+
173  *                     :              | dtrace_actkind_t -+------> kind
174  *                     v              | dtrace_difo_t * --+------> DIFO (if any)
175  *                                    | dtrace_recdesc_t -+------> record descr.
176  *                                    | dtrace_action_t * +------+
177  *                                    +-------------------+      |
178  *                                                               | Next action
179  *                               +-------------------------------+  (if any)
180  *                               |
181  *                               |           Action
182  *                               |       dtrace_action_t
183  *                               +--> +-------------------+
184  *                                    | dtrace_actkind_t -+------> kind
185  *                                    | dtrace_difo_t * --+------> DIFO (if any)
186  *                                    | dtrace_action_t * +------+
187  *                                    +-------------------+      |
188  *                                                               | Next action
189  *                               +-------------------------------+  (if any)
190  *                               |
191  *                               :
192  *                               v
193  *
194  *
195  * dtrace_probe() iterates over the ECB list.  If the ECB needs less space
196  * than is available in the principal buffer, the ECB is processed:  if the
197  * predicate is non-NULL, the DIF object is executed.  If the result is
198  * non-zero, the action list is processed, with each action being executed
199  * accordingly.  When the action list has been completely executed, processing
200  * advances to the next ECB.  processing advances to the next ECB.  If the
201  * result is non-zero; For each ECB, it first determines the The ECB
202  * abstraction allows disjoint consumers to multiplex on single probes.
203  */
204 struct dtrace_ecb {
205 	dtrace_epid_t dte_epid;			/* enabled probe ID */
206 	uint32_t dte_alignment;			/* required alignment */
207 	size_t dte_needed;			/* bytes needed */
208 	size_t dte_size;			/* total size of payload */
209 	dtrace_predicate_t *dte_predicate;	/* predicate, if any */
210 	dtrace_action_t *dte_action;		/* actions, if any */
211 	dtrace_ecb_t *dte_next;			/* next ECB on probe */
212 	dtrace_state_t *dte_state;		/* pointer to state */
213 	uint32_t dte_cond;			/* security condition */
214 	dtrace_probe_t *dte_probe;		/* pointer to probe */
215 	dtrace_action_t *dte_action_last;	/* last action on ECB */
216 	uint64_t dte_uarg;			/* library argument */
217 };
218 
219 struct dtrace_predicate {
220 	dtrace_difo_t *dtp_difo;		/* DIF object */
221 	dtrace_cacheid_t dtp_cacheid;		/* cache identifier */
222 	int dtp_refcnt;				/* reference count */
223 };
224 
225 struct dtrace_action {
226 	dtrace_actkind_t dta_kind;		/* kind of action */
227 	uint16_t dta_intuple;			/* boolean:  in aggregation */
228 	uint32_t dta_refcnt;			/* reference count */
229 	dtrace_difo_t *dta_difo;		/* pointer to DIFO */
230 	dtrace_recdesc_t dta_rec;		/* record description */
231 	dtrace_action_t *dta_prev;		/* previous action */
232 	dtrace_action_t *dta_next;		/* next action */
233 };
234 
235 typedef struct dtrace_aggregation {
236 	dtrace_action_t dtag_action;		/* action; must be first */
237 	dtrace_aggid_t dtag_id;			/* identifier */
238 	dtrace_ecb_t *dtag_ecb;			/* corresponding ECB */
239 	dtrace_action_t *dtag_first;		/* first action in tuple */
240 	uint32_t dtag_base;			/* base of aggregation */
241 	uint64_t dtag_initial;			/* initial value */
242 	void (*dtag_aggregate)(uint64_t *, uint64_t);
243 } dtrace_aggregation_t;
244 
245 /*
246  * DTrace Buffers
247  *
248  * Principal buffers, aggregation buffers, and speculative buffers are all
249  * managed with the dtrace_buffer structure.  By default, this structure
250  * includes twin data buffers -- dtb_tomax and dtb_xamot -- that serve as the
251  * active and passive buffers, respectively.  For speculative buffers,
252  * dtb_xamot will be NULL; for "ring" and "fill" buffers, dtb_xamot will point
253  * to a scratch buffer.  For all buffer types, the dtrace_buffer structure is
254  * always allocated on a per-CPU basis; a single dtrace_buffer structure is
255  * never shared among CPUs.  (That is, there is never true sharing of the
256  * dtrace_buffer structure; to prevent false sharing of the structure, it must
257  * always be aligned to the coherence granularity -- generally 64 bytes.)
258  *
259  * One of the critical design decisions of DTrace is that a given ECB always
260  * stores the same quantity and type of data.  This is done to assure that the
261  * only metadata required for an ECB's traced data is the EPID.  That is, from
262  * the EPID, the consumer can determine the data layout.  (The data buffer
263  * layout is shown schematically below.)  By assuring that one can determine
264  * data layout from the EPID, the metadata stream can be separated from the
265  * data stream -- simplifying the data stream enormously.
266  *
267  *      base of data buffer --->  +------+--------------------+------+
268  *                                | EPID | data               | EPID |
269  *                                +------+--------+------+----+------+
270  *                                | data          | EPID | data      |
271  *                                +---------------+------+-----------+
272  *                                | data, cont.                      |
273  *                                +------+--------------------+------+
274  *                                | EPID | data               |      |
275  *                                +------+--------------------+      |
276  *                                |                ||                |
277  *                                |                ||                |
278  *                                |                \/                |
279  *                                :                                  :
280  *                                .                                  .
281  *                                .                                  .
282  *                                .                                  .
283  *                                :                                  :
284  *                                |                                  |
285  *     limit of data buffer --->  +----------------------------------+
286  *
287  * When evaluating an ECB, dtrace_probe() determines if the ECB's needs of the
288  * principal buffer (both scratch and payload) exceed the available space.  If
289  * the ECB's needs exceed available space (and if the principal buffer policy
290  * is the default "switch" policy), the ECB is dropped, the buffer's drop count
291  * is incremented, and processing advances to the next ECB.  If the ECB's needs
292  * can be met with the available space, the ECB is processed, but the offset in
293  * the principal buffer is only advanced if the ECB completes processing
294  * without error.
295  *
296  * When a buffer is to be switched (either because the buffer is the principal
297  * buffer with a "switch" policy or because it is an aggregation buffer), a
298  * cross call is issued to the CPU associated with the buffer.  In the cross
299  * call context, interrupts are disabled, and the active and the inactive
300  * buffers are atomically switched.  This involves switching the data pointers,
301  * copying the various state fields (offset, drops, errors, etc.) into their
302  * inactive equivalents, and clearing the state fields.  Because interrupts are
303  * disabled during this procedure, the switch is guaranteed to appear atomic to
304  * dtrace_probe().
305  *
306  * DTrace Ring Buffering
307  *
308  * To process a ring buffer correctly, one must know the oldest valid record.
309  * Processing starts at the oldest record in the buffer and continues until
310  * the end of the buffer is reached.  Processing then resumes starting with
311  * the record stored at offset 0 in the buffer, and continues until the
312  * youngest record is processed.  If trace records are of a fixed-length,
313  * determining the oldest record is trivial:
314  *
315  *   - If the ring buffer has not wrapped, the oldest record is the record
316  *     stored at offset 0.
317  *
318  *   - If the ring buffer has wrapped, the oldest record is the record stored
319  *     at the current offset.
320  *
321  * With variable length records, however, just knowing the current offset
322  * doesn't suffice for determining the oldest valid record:  assuming that one
323  * allows for arbitrary data, one has no way of searching forward from the
324  * current offset to find the oldest valid record.  (That is, one has no way
325  * of separating data from metadata.) It would be possible to simply refuse to
326  * process any data in the ring buffer between the current offset and the
327  * limit, but this leaves (potentially) an enormous amount of otherwise valid
328  * data unprocessed.
329  *
330  * To effect ring buffering, we track two offsets in the buffer:  the current
331  * offset and the _wrapped_ offset.  If a request is made to reserve some
332  * amount of data, and the buffer has wrapped, the wrapped offset is
333  * incremented until the wrapped offset minus the current offset is greater
334  * than or equal to the reserve request.  This is done by repeatedly looking
335  * up the ECB corresponding to the EPID at the current wrapped offset, and
336  * incrementing the wrapped offset by the size of the data payload
337  * corresponding to that ECB.  If this offset is greater than or equal to the
338  * limit of the data buffer, the wrapped offset is set to 0.  Thus, the
339  * current offset effectively "chases" the wrapped offset around the buffer.
340  * Schematically:
341  *
342  *      base of data buffer --->  +------+--------------------+------+
343  *                                | EPID | data               | EPID |
344  *                                +------+--------+------+----+------+
345  *                                | data          | EPID | data      |
346  *                                +---------------+------+-----------+
347  *                                | data, cont.                      |
348  *                                +------+---------------------------+
349  *                                | EPID | data                      |
350  *           current offset --->  +------+---------------------------+
351  *                                | invalid data                     |
352  *           wrapped offset --->  +------+--------------------+------+
353  *                                | EPID | data               | EPID |
354  *                                +------+--------+------+----+------+
355  *                                | data          | EPID | data      |
356  *                                +---------------+------+-----------+
357  *                                :                                  :
358  *                                .                                  .
359  *                                .        ... valid data ...        .
360  *                                .                                  .
361  *                                :                                  :
362  *                                +------+-------------+------+------+
363  *                                | EPID | data        | EPID | data |
364  *                                +------+------------++------+------+
365  *                                | data, cont.       | leftover     |
366  *     limit of data buffer --->  +-------------------+--------------+
367  *
368  * If the amount of requested buffer space exceeds the amount of space
369  * available between the current offset and the end of the buffer:
370  *
371  *  (1)  all words in the data buffer between the current offset and the limit
372  *       of the data buffer (marked "leftover", above) are set to
373  *       DTRACE_EPIDNONE
374  *
375  *  (2)  the wrapped offset is set to zero
376  *
377  *  (3)  the iteration process described above occurs until the wrapped offset
378  *       is greater than the amount of desired space.
379  *
380  * The wrapped offset is implemented by (re-)using the inactive offset.
381  * In a "switch" buffer policy, the inactive offset stores the offset in
382  * the inactive buffer; in a "ring" buffer policy, it stores the wrapped
383  * offset.
384  *
385  * DTrace Scratch Buffering
386  *
387  * Some ECBs may wish to allocate dynamically-sized temporary scratch memory.
388  * To accommodate such requests easily, scratch memory may be allocated in
389  * the buffer beyond the current offset plus the needed memory of the current
390  * ECB.  If there isn't sufficient room in the buffer for the requested amount
391  * of scratch space, the allocation fails and an error is generated.  Scratch
392  * memory is tracked in the dtrace_mstate_t and is automatically freed when
393  * the ECB ceases processing.  Note that ring buffers cannot allocate their
394  * scratch from the principal buffer -- lest they needlessly overwrite older,
395  * valid data.  Ring buffers therefore have their own dedicated scratch buffer
396  * from which scratch is allocated.
397  */
398 #define	DTRACEBUF_RING		0x0001		/* bufpolicy set to "ring" */
399 #define	DTRACEBUF_FILL		0x0002		/* bufpolicy set to "fill" */
400 #define	DTRACEBUF_NOSWITCH	0x0004		/* do not switch buffer */
401 #define	DTRACEBUF_WRAPPED	0x0008		/* ring buffer has wrapped */
402 #define	DTRACEBUF_DROPPED	0x0010		/* drops occurred */
403 #define	DTRACEBUF_ERROR		0x0020		/* errors occurred */
404 #define	DTRACEBUF_FULL		0x0040		/* "fill" buffer is full */
405 #define	DTRACEBUF_CONSUMED	0x0080		/* buffer has been consumed */
406 #define	DTRACEBUF_INACTIVE	0x0100		/* buffer is not yet active */
407 
408 typedef struct dtrace_buffer {
409 	uint64_t dtb_offset;			/* current offset in buffer */
410 	uint64_t dtb_size;			/* size of buffer */
411 	uint32_t dtb_flags;			/* flags */
412 	uint32_t dtb_drops;			/* number of drops */
413 	caddr_t dtb_tomax;			/* active buffer */
414 	caddr_t dtb_xamot;			/* inactive buffer */
415 	uint32_t dtb_xamot_flags;		/* inactive flags */
416 	uint32_t dtb_xamot_drops;		/* drops in inactive buffer */
417 	uint64_t dtb_xamot_offset;		/* offset in inactive buffer */
418 	uint32_t dtb_errors;			/* number of errors */
419 	uint32_t dtb_xamot_errors;		/* errors in inactive buffer */
420 #ifndef _LP64
421 	uint64_t dtb_pad1;
422 #endif
423 } dtrace_buffer_t;
424 
425 /*
426  * DTrace Aggregation Buffers
427  *
428  * Aggregation buffers use much of the same mechanism as described above
429  * ("DTrace Buffers").  However, because an aggregation is fundamentally a
430  * hash, there exists dynamic metadata associated with an aggregation buffer
431  * that is not associated with other kinds of buffers.  This aggregation
432  * metadata is _only_ relevant for the in-kernel implementation of
433  * aggregations; it is not actually relevant to user-level consumers.  To do
434  * this, we allocate dynamic aggregation data (hash keys and hash buckets)
435  * starting below the _limit_ of the buffer, and we allocate data from the
436  * _base_ of the buffer.  When the aggregation buffer is copied out, _only_ the
437  * data is copied out; the metadata is simply discarded.  Schematically,
438  * aggregation buffers look like:
439  *
440  *      base of data buffer --->  +-------+------+-----------+-------+
441  *                                | aggid | key  | value     | aggid |
442  *                                +-------+------+-----------+-------+
443  *                                | key                              |
444  *                                +-------+-------+-----+------------+
445  *                                | value | aggid | key | value      |
446  *                                +-------+------++-----+------+-----+
447  *                                | aggid | key  | value       |     |
448  *                                +-------+------+-------------+     |
449  *                                |                ||                |
450  *                                |                ||                |
451  *                                |                \/                |
452  *                                :                                  :
453  *                                .                                  .
454  *                                .                                  .
455  *                                .                                  .
456  *                                :                                  :
457  *                                |                /\                |
458  *                                |                ||   +------------+
459  *                                |                ||   |            |
460  *                                +---------------------+            |
461  *                                | hash keys                        |
462  *                                | (dtrace_aggkey structures)       |
463  *                                |                                  |
464  *                                +----------------------------------+
465  *                                | hash buckets                     |
466  *                                | (dtrace_aggbuffer structure)     |
467  *                                |                                  |
468  *     limit of data buffer --->  +----------------------------------+
469  *
470  *
471  * As implied above, just as we assure that ECBs always store a constant
472  * amount of data, we assure that a given aggregation -- identified by its
473  * aggregation ID -- always stores data of a constant quantity and type.
474  * As with EPIDs, this allows the aggregation ID to serve as the metadata for a
475  * given record.
476  *
477  * Note that the size of the dtrace_aggkey structure must be sizeof (uintptr_t)
478  * aligned.  (If this the structure changes such that this becomes false, an
479  * assertion will fail in dtrace_aggregate().)
480  */
481 typedef struct dtrace_aggkey {
482 	uint32_t dtak_hashval;			/* hash value */
483 	uint32_t dtak_action:4;			/* action -- 4 bits */
484 	uint32_t dtak_size:28;			/* size -- 28 bits */
485 	caddr_t dtak_data;			/* data pointer */
486 	struct dtrace_aggkey *dtak_next;	/* next in hash chain */
487 } dtrace_aggkey_t;
488 
489 typedef struct dtrace_aggbuffer {
490 	uintptr_t dtagb_hashsize;		/* number of buckets */
491 	uintptr_t dtagb_free;			/* free list of keys */
492 	dtrace_aggkey_t **dtagb_hash;		/* hash table */
493 } dtrace_aggbuffer_t;
494 
495 /*
496  * DTrace Speculations
497  *
498  * Speculations have a per-CPU buffer and a global state.  Once a speculation
499  * buffer has been comitted or discarded, it cannot be reused until all CPUs
500  * have taken the same action (commit or discard) on their respective
501  * speculative buffer.  However, because DTrace probes may execute in arbitrary
502  * context, other CPUs cannot simply be cross-called at probe firing time to
503  * perform the necessary commit or discard.  The speculation states thus
504  * optimize for the case that a speculative buffer is only active on one CPU at
505  * the time of a commit() or discard() -- for if this is the case, other CPUs
506  * need not take action, and the speculation is immediately available for
507  * reuse.  If the speculation is active on multiple CPUs, it must be
508  * asynchronously cleaned -- potentially leading to a higher rate of dirty
509  * speculative drops.  The speculation states are as follows:
510  *
511  *  DTRACESPEC_INACTIVE       <= Initial state; inactive speculation
512  *  DTRACESPEC_ACTIVE         <= Allocated, but not yet speculatively traced to
513  *  DTRACESPEC_ACTIVEONE      <= Speculatively traced to on one CPU
514  *  DTRACESPEC_ACTIVEMANY     <= Speculatively traced to on more than one CPU
515  *  DTRACESPEC_COMMITTING     <= Currently being commited on one CPU
516  *  DTRACESPEC_COMMITTINGMANY <= Currently being commited on many CPUs
517  *  DTRACESPEC_DISCARDING     <= Currently being discarded on many CPUs
518  *
519  * The state transition diagram is as follows:
520  *
521  *     +----------------------------------------------------------+
522  *     |                                                          |
523  *     |                      +------------+                      |
524  *     |  +-------------------| COMMITTING |<-----------------+   |
525  *     |  |                   +------------+                  |   |
526  *     |  | copied spec.            ^             commit() on |   | discard() on
527  *     |  | into principal          |              active CPU |   | active CPU
528  *     |  |                         | commit()                |   |
529  *     V  V                         |                         |   |
530  * +----------+                 +--------+                +-----------+
531  * | INACTIVE |---------------->| ACTIVE |--------------->| ACTIVEONE |
532  * +----------+  speculation()  +--------+  speculate()   +-----------+
533  *     ^  ^                         |                         |   |
534  *     |  |                         | discard()               |   |
535  *     |  | asynchronously          |            discard() on |   | speculate()
536  *     |  | cleaned                 V            inactive CPU |   | on inactive
537  *     |  |                   +------------+                  |   | CPU
538  *     |  +-------------------| DISCARDING |<-----------------+   |
539  *     |                      +------------+                      |
540  *     | asynchronously             ^                             |
541  *     | copied spec.               |       discard()             |
542  *     | into principal             +------------------------+    |
543  *     |                                                     |    V
544  *  +----------------+             commit()              +------------+
545  *  | COMMITTINGMANY |<----------------------------------| ACTIVEMANY |
546  *  +----------------+                                   +------------+
547  */
548 typedef enum dtrace_speculation_state {
549 	DTRACESPEC_INACTIVE = 0,
550 	DTRACESPEC_ACTIVE,
551 	DTRACESPEC_ACTIVEONE,
552 	DTRACESPEC_ACTIVEMANY,
553 	DTRACESPEC_COMMITTING,
554 	DTRACESPEC_COMMITTINGMANY,
555 	DTRACESPEC_DISCARDING
556 } dtrace_speculation_state_t;
557 
558 typedef struct dtrace_speculation {
559 	dtrace_speculation_state_t dtsp_state;	/* current speculation state */
560 	int dtsp_cleaning;			/* non-zero if being cleaned */
561 	dtrace_buffer_t *dtsp_buffer;		/* speculative buffer */
562 } dtrace_speculation_t;
563 
564 /*
565  * DTrace Dynamic Variables
566  *
567  * The dynamic variable problem is obviously decomposed into two subproblems:
568  * allocating new dynamic storage, and freeing old dynamic storage.  The
569  * presence of the second problem makes the first much more complicated -- or
570  * rather, the absence of the second renders the first trivial.  This is the
571  * case with aggregations, for which there is effectively no deallocation of
572  * dynamic storage.  (Or more accurately, all dynamic storage is deallocated
573  * when a snapshot is taken of the aggregation.)  As DTrace dynamic variables
574  * allow for both dynamic allocation and dynamic deallocation, the
575  * implementation of dynamic variables is quite a bit more complicated than
576  * that of their aggregation kin.
577  *
578  * We observe that allocating new dynamic storage is tricky only because the
579  * size can vary -- the allocation problem is much easier if allocation sizes
580  * are uniform.  We further observe that in D, the size of dynamic variables is
581  * actually _not_ dynamic -- dynamic variable sizes may be determined by static
582  * analysis of DIF text.  (This is true even of putatively dynamically-sized
583  * objects like strings and stacks, the sizes of which are dictated by the
584  * "stringsize" and "stackframes" variables, respectively.)  We exploit this by
585  * performing this analysis on all DIF before enabling any probes.  For each
586  * dynamic load or store, we calculate the dynamically-allocated size plus the
587  * size of the dtrace_dynvar structure plus the storage required to key the
588  * data.  For all DIF, we take the largest value and dub it the _chunksize_.
589  * We then divide dynamic memory into two parts:  a hash table that is wide
590  * enough to have every chunk in its own bucket, and a larger region of equal
591  * chunksize units.  Whenever we wish to dynamically allocate a variable, we
592  * always allocate a single chunk of memory.  Depending on the uniformity of
593  * allocation, this will waste some amount of memory -- but it eliminates the
594  * non-determinism inherent in traditional heap fragmentation.
595  *
596  * Dynamic objects are allocated by storing a non-zero value to them; they are
597  * deallocated by storing a zero value to them.  Dynamic variables are
598  * complicated enormously by being shared between CPUs.  In particular,
599  * consider the following scenario:
600  *
601  *                 CPU A                                 CPU B
602  *  +---------------------------------+   +---------------------------------+
603  *  |                                 |   |                                 |
604  *  | allocates dynamic object a[123] |   |                                 |
605  *  | by storing the value 345 to it  |   |                                 |
606  *  |                               --------->                              |
607  *  |                                 |   | wishing to load from object     |
608  *  |                                 |   | a[123], performs lookup in      |
609  *  |                                 |   | dynamic variable space          |
610  *  |                               <---------                              |
611  *  | deallocates object a[123] by    |   |                                 |
612  *  | storing 0 to it                 |   |                                 |
613  *  |                                 |   |                                 |
614  *  | allocates dynamic object b[567] |   | performs load from a[123]       |
615  *  | by storing the value 789 to it  |   |                                 |
616  *  :                                 :   :                                 :
617  *  .                                 .   .                                 .
618  *
619  * This is obviously a race in the D program, but there are nonetheless only
620  * two valid values for CPU B's load from a[123]:  345 or 0.  Most importantly,
621  * CPU B may _not_ see the value 789 for a[123].
622  *
623  * There are essentially two ways to deal with this:
624  *
625  *  (1)  Explicitly spin-lock variables.  That is, if CPU B wishes to load
626  *       from a[123], it needs to lock a[123] and hold the lock for the
627  *       duration that it wishes to manipulate it.
628  *
629  *  (2)  Avoid reusing freed chunks until it is known that no CPU is referring
630  *       to them.
631  *
632  * The implementation of (1) is rife with complexity, because it requires the
633  * user of a dynamic variable to explicitly decree when they are done using it.
634  * Were all variables by value, this perhaps wouldn't be debilitating -- but
635  * dynamic variables of non-scalar types are tracked by reference.  That is, if
636  * a dynamic variable is, say, a string, and that variable is to be traced to,
637  * say, the principal buffer, the DIF emulation code returns to the main
638  * dtrace_probe() loop a pointer to the underlying storage, not the contents of
639  * the storage.  Further, code calling on DIF emulation would have to be aware
640  * that the DIF emulation has returned a reference to a dynamic variable that
641  * has been potentially locked.  The variable would have to be unlocked after
642  * the main dtrace_probe() loop is finished with the variable, and the main
643  * dtrace_probe() loop would have to be careful to not call any further DIF
644  * emulation while the variable is locked to avoid deadlock.  More generally,
645  * if one were to implement (1), DIF emulation code dealing with dynamic
646  * variables could only deal with one dynamic variable at a time (lest deadlock
647  * result).  To sum, (1) exports too much subtlety to the users of dynamic
648  * variables -- increasing maintenance burden and imposing serious constraints
649  * on future DTrace development.
650  *
651  * The implementation of (2) is also complex, but the complexity is more
652  * manageable.  We need to be sure that when a variable is deallocated, it is
653  * not placed on a traditional free list, but rather on a _dirty_ list.  Once a
654  * variable is on a dirty list, it cannot be found by CPUs performing a
655  * subsequent lookup of the variable -- but it may still be in use by other
656  * CPUs.  To assure that all CPUs that may be seeing the old variable have
657  * cleared out of probe context, a dtrace_sync() can be issued.  Once the
658  * dtrace_sync() has completed, it can be known that all CPUs are done
659  * manipulating the dynamic variable -- the dirty list can be atomically
660  * appended to the free list.  Unfortunately, there's a slight hiccup in this
661  * mechanism:  dtrace_sync() may not be issued from probe context.  The
662  * dtrace_sync() must be therefore issued asynchronously from non-probe
663  * context.  For this we rely on the DTrace cleaner, a cyclic that runs at the
664  * "cleanrate" frequency.  To ease this implementation, we define several chunk
665  * lists:
666  *
667  *   - Dirty.  Deallocated chunks, not yet cleaned.  Not available.
668  *
669  *   - Rinsing.  Formerly dirty chunks that are currently being asynchronously
670  *     cleaned.  Not available, but will be shortly.  Dynamic variable
671  *     allocation may not spin or block for availability, however.
672  *
673  *   - Clean.  Clean chunks, ready for allocation -- but not on the free list.
674  *
675  *   - Free.  Available for allocation.
676  *
677  * Moreover, to avoid absurd contention, _each_ of these lists is implemented
678  * on a per-CPU basis.  This is only for performance, not correctness; chunks
679  * may be allocated from another CPU's free list.  The algorithm for allocation
680  * then is this:
681  *
682  *   (1)  Attempt to atomically allocate from current CPU's free list.  If list
683  *        is non-empty and allocation is successful, allocation is complete.
684  *
685  *   (2)  If the clean list is non-empty, atomically move it to the free list,
686  *        and reattempt (1).
687  *
688  *   (3)  If the dynamic variable space is in the CLEAN state, look for free
689  *        and clean lists on other CPUs by setting the current CPU to the next
690  *        CPU, and reattempting (1).  If the next CPU is the current CPU (that
691  *        is, if all CPUs have been checked), atomically switch the state of
692  *        the dynamic variable space based on the following:
693  *
694  *        - If no free chunks were found and no dirty chunks were found,
695  *          atomically set the state to EMPTY.
696  *
697  *        - If dirty chunks were found, atomically set the state to DIRTY.
698  *
699  *        - If rinsing chunks were found, atomically set the state to RINSING.
700  *
701  *   (4)  Based on state of dynamic variable space state, increment appropriate
702  *        counter to indicate dynamic drops (if in EMPTY state) vs. dynamic
703  *        dirty drops (if in DIRTY state) vs. dynamic rinsing drops (if in
704  *        RINSING state).  Fail the allocation.
705  *
706  * The cleaning cyclic operates with the following algorithm:  for all CPUs
707  * with a non-empty dirty list, atomically move the dirty list to the rinsing
708  * list.  Perform a dtrace_sync().  For all CPUs with a non-empty rinsing list,
709  * atomically move the rinsing list to the clean list.  Perform another
710  * dtrace_sync().  By this point, all CPUs have seen the new clean list; the
711  * state of the dynamic variable space can be restored to CLEAN.
712  *
713  * There exist two final races that merit explanation.  The first is a simple
714  * allocation race:
715  *
716  *                 CPU A                                 CPU B
717  *  +---------------------------------+   +---------------------------------+
718  *  |                                 |   |                                 |
719  *  | allocates dynamic object a[123] |   | allocates dynamic object a[123] |
720  *  | by storing the value 345 to it  |   | by storing the value 567 to it  |
721  *  |                                 |   |                                 |
722  *  :                                 :   :                                 :
723  *  .                                 .   .                                 .
724  *
725  * Again, this is a race in the D program.  It can be resolved by having a[123]
726  * hold the value 345 or a[123] hold the value 567 -- but it must be true that
727  * a[123] have only _one_ of these values.  (That is, the racing CPUs may not
728  * put the same element twice on the same hash chain.)  This is resolved
729  * simply:  before the allocation is undertaken, the start of the new chunk's
730  * hash chain is noted.  Later, after the allocation is complete, the hash
731  * chain is atomically switched to point to the new element.  If this fails
732  * (because of either concurrent allocations or an allocation concurrent with a
733  * deletion), the newly allocated chunk is deallocated to the dirty list, and
734  * the whole process of looking up (and potentially allocating) the dynamic
735  * variable is reattempted.
736  *
737  * The final race is a simple deallocation race:
738  *
739  *                 CPU A                                 CPU B
740  *  +---------------------------------+   +---------------------------------+
741  *  |                                 |   |                                 |
742  *  | deallocates dynamic object      |   | deallocates dynamic object      |
743  *  | a[123] by storing the value 0   |   | a[123] by storing the value 0   |
744  *  | to it                           |   | to it                           |
745  *  |                                 |   |                                 |
746  *  :                                 :   :                                 :
747  *  .                                 .   .                                 .
748  *
749  * Once again, this is a race in the D program, but it is one that we must
750  * handle without corrupting the underlying data structures.  Because
751  * deallocations require the deletion of a chunk from the middle of a hash
752  * chain, we cannot use a single-word atomic operation to remove it.  For this,
753  * we add a spin lock to the hash buckets that is _only_ used for deallocations
754  * (allocation races are handled as above).  Further, this spin lock is _only_
755  * held for the duration of the delete; before control is returned to the DIF
756  * emulation code, the hash bucket is unlocked.
757  */
758 typedef struct dtrace_key {
759 	uint64_t dttk_value;			/* data value or data pointer */
760 	uint64_t dttk_size;			/* 0 if by-val, >0 if by-ref */
761 } dtrace_key_t;
762 
763 typedef struct dtrace_tuple {
764 	uint32_t dtt_nkeys;			/* number of keys in tuple */
765 	uint32_t dtt_pad;			/* padding */
766 	dtrace_key_t dtt_key[1];		/* array of tuple keys */
767 } dtrace_tuple_t;
768 
769 typedef struct dtrace_dynvar {
770 	uint64_t dtdv_hashval;			/* hash value -- 0 if free */
771 	struct dtrace_dynvar *dtdv_next;	/* next on list or hash chain */
772 	void *dtdv_data;			/* pointer to data */
773 	dtrace_tuple_t dtdv_tuple;		/* tuple key */
774 } dtrace_dynvar_t;
775 
776 typedef enum dtrace_dynvar_op {
777 	DTRACE_DYNVAR_ALLOC,
778 	DTRACE_DYNVAR_NOALLOC,
779 	DTRACE_DYNVAR_DEALLOC
780 } dtrace_dynvar_op_t;
781 
782 typedef struct dtrace_dynhash {
783 	dtrace_dynvar_t *dtdh_chain;		/* hash chain for this bucket */
784 	uintptr_t dtdh_lock;			/* deallocation lock */
785 #ifdef _LP64
786 	uintptr_t dtdh_pad[6];			/* pad to avoid false sharing */
787 #else
788 	uintptr_t dtdh_pad[14];			/* pad to avoid false sharing */
789 #endif
790 } dtrace_dynhash_t;
791 
792 typedef struct dtrace_dstate_percpu {
793 	dtrace_dynvar_t *dtdsc_free;		/* free list for this CPU */
794 	dtrace_dynvar_t *dtdsc_dirty;		/* dirty list for this CPU */
795 	dtrace_dynvar_t *dtdsc_rinsing;		/* rinsing list for this CPU */
796 	dtrace_dynvar_t *dtdsc_clean;		/* clean list for this CPU */
797 	uint64_t dtdsc_drops;			/* number of capacity drops */
798 	uint64_t dtdsc_dirty_drops;		/* number of dirty drops */
799 	uint64_t dtdsc_rinsing_drops;		/* number of rinsing drops */
800 #ifdef _LP64
801 	uint64_t dtdsc_pad;			/* pad to avoid false sharing */
802 #else
803 	uint64_t dtdsc_pad[2];			/* pad to avoid false sharing */
804 #endif
805 } dtrace_dstate_percpu_t;
806 
807 typedef enum dtrace_dstate_state {
808 	DTRACE_DSTATE_CLEAN = 0,
809 	DTRACE_DSTATE_EMPTY,
810 	DTRACE_DSTATE_DIRTY,
811 	DTRACE_DSTATE_RINSING
812 } dtrace_dstate_state_t;
813 
814 typedef struct dtrace_dstate {
815 	void *dtds_base;			/* base of dynamic var. space */
816 	size_t dtds_size;			/* size of dynamic var. space */
817 	size_t dtds_hashsize;			/* number of buckets in hash */
818 	size_t dtds_chunksize;			/* size of each chunk */
819 	dtrace_dynhash_t *dtds_hash;		/* pointer to hash table */
820 	dtrace_dstate_state_t dtds_state;	/* current dynamic var. state */
821 	dtrace_dstate_percpu_t *dtds_percpu;	/* per-CPU dyn. var. state */
822 } dtrace_dstate_t;
823 
824 /*
825  * DTrace Variable State
826  *
827  * The DTrace variable state tracks user-defined variables in its dtrace_vstate
828  * structure.  Each DTrace consumer has exactly one dtrace_vstate structure,
829  * but some dtrace_vstate structures may exist without a corresponding DTrace
830  * consumer (see "DTrace Helpers", below).  As described in <sys/dtrace.h>,
831  * user-defined variables can have one of three scopes:
832  *
833  *  DIFV_SCOPE_GLOBAL  =>  global scope
834  *  DIFV_SCOPE_THREAD  =>  thread-local scope (i.e. "self->" variables)
835  *  DIFV_SCOPE_LOCAL   =>  clause-local scope (i.e. "this->" variables)
836  *
837  * The variable state tracks variables by both their scope and their allocation
838  * type:
839  *
840  *  - The dtvs_globals and dtvs_locals members each point to an array of
841  *    dtrace_statvar structures.  These structures contain both the variable
842  *    metadata (dtrace_difv structures) and the underlying storage for all
843  *    statically allocated variables, including statically allocated
844  *    DIFV_SCOPE_GLOBAL variables and all DIFV_SCOPE_LOCAL variables.
845  *
846  *  - The dtvs_tlocals member points to an array of dtrace_difv structures for
847  *    DIFV_SCOPE_THREAD variables.  As such, this array tracks _only_ the
848  *    variable metadata for DIFV_SCOPE_THREAD variables; the underlying storage
849  *    is allocated out of the dynamic variable space.
850  *
851  *  - The dtvs_dynvars member is the dynamic variable state associated with the
852  *    variable state.  The dynamic variable state (described in "DTrace Dynamic
853  *    Variables", above) tracks all DIFV_SCOPE_THREAD variables and all
854  *    dynamically-allocated DIFV_SCOPE_GLOBAL variables.
855  */
856 typedef struct dtrace_statvar {
857 	uint64_t dtsv_data;			/* data or pointer to it */
858 	size_t dtsv_size;			/* size of pointed-to data */
859 	int dtsv_refcnt;			/* reference count */
860 	dtrace_difv_t dtsv_var;			/* variable metadata */
861 } dtrace_statvar_t;
862 
863 typedef struct dtrace_vstate {
864 	dtrace_state_t *dtvs_state;		/* back pointer to state */
865 	dtrace_statvar_t **dtvs_globals;	/* statically-allocated glbls */
866 	int dtvs_nglobals;			/* number of globals */
867 	dtrace_difv_t *dtvs_tlocals;		/* thread-local metadata */
868 	int dtvs_ntlocals;			/* number of thread-locals */
869 	dtrace_statvar_t **dtvs_locals;		/* clause-local data */
870 	int dtvs_nlocals;			/* number of clause-locals */
871 	dtrace_dstate_t dtvs_dynvars;		/* dynamic variable state */
872 } dtrace_vstate_t;
873 
874 /*
875  * DTrace Machine State
876  *
877  * In the process of processing a fired probe, DTrace needs to track and/or
878  * cache some per-CPU state associated with that particular firing.  This is
879  * state that is always discarded after the probe firing has completed, and
880  * much of it is not specific to any DTrace consumer, remaining valid across
881  * all ECBs.  This state is tracked in the dtrace_mstate structure.
882  */
883 #define	DTRACE_MSTATE_ARGS		0x00000001
884 #define	DTRACE_MSTATE_PROBE		0x00000002
885 #define	DTRACE_MSTATE_EPID		0x00000004
886 #define	DTRACE_MSTATE_TIMESTAMP		0x00000008
887 #define	DTRACE_MSTATE_STACKDEPTH	0x00000010
888 #define	DTRACE_MSTATE_CALLER		0x00000020
889 #define	DTRACE_MSTATE_IPL		0x00000040
890 #define	DTRACE_MSTATE_FLTOFFS		0x00000080
891 #define	DTRACE_MSTATE_WALLTIMESTAMP	0x00000100
892 #define	DTRACE_MSTATE_USTACKDEPTH	0x00000200
893 
894 typedef struct dtrace_mstate {
895 	uintptr_t dtms_scratch_base;		/* base of scratch space */
896 	uintptr_t dtms_scratch_ptr;		/* current scratch pointer */
897 	size_t dtms_scratch_size;		/* scratch size */
898 	uint32_t dtms_present;			/* variables that are present */
899 	uint64_t dtms_arg[5];			/* cached arguments */
900 	dtrace_epid_t dtms_epid;		/* current EPID */
901 	uint64_t dtms_timestamp;		/* cached timestamp */
902 	hrtime_t dtms_walltimestamp;		/* cached wall timestamp */
903 	int dtms_stackdepth;			/* cached stackdepth */
904 	int dtms_ustackdepth;			/* cached ustackdepth */
905 	struct dtrace_probe *dtms_probe;	/* current probe */
906 	uintptr_t dtms_caller;			/* cached caller */
907 	int dtms_ipl;				/* cached interrupt pri lev */
908 	int dtms_fltoffs;			/* faulting DIFO offset */
909 	uintptr_t dtms_strtok;			/* saved strtok() pointer */
910 } dtrace_mstate_t;
911 
912 #define	DTRACE_COND_OWNER	0x1
913 #define	DTRACE_COND_USERMODE	0x2
914 
915 #define	DTRACE_PROBEKEY_MAXDEPTH	8	/* max glob recursion depth */
916 
917 /*
918  * DTrace Activity
919  *
920  * Each DTrace consumer is in one of several states, which (for purposes of
921  * avoiding yet-another overloading of the noun "state") we call the current
922  * _activity_.  The activity transitions on dtrace_go() (from DTRACIOCGO), on
923  * dtrace_stop() (from DTRACIOCSTOP) and on the exit() action.  Activities may
924  * only transition in one direction; the activity transition diagram is a
925  * directed acyclic graph.  The activity transition diagram is as follows:
926  *
927  *
928  * +----------+                   +--------+                   +--------+
929  * | INACTIVE |------------------>| WARMUP |------------------>| ACTIVE |
930  * +----------+   dtrace_go(),    +--------+   dtrace_go(),    +--------+
931  *                before BEGIN        |        after BEGIN       |  |  |
932  *                                    |                          |  |  |
933  *                      exit() action |                          |  |  |
934  *                     from BEGIN ECB |                          |  |  |
935  *                                    |                          |  |  |
936  *                                    v                          |  |  |
937  *                               +----------+     exit() action  |  |  |
938  *                               | DRAINING |<-------------------+  |  |
939  *                               +----------+                       |  |
940  *                                    |                             |  |
941  *                     dtrace_stop(), |                             |  |
942  *                       before END   |                             |  |
943  *                                    |                             |  |
944  *                                    v                             |  |
945  * +---------+                   +----------+                       |  |
946  * | STOPPED |<------------------| COOLDOWN |<----------------------+  |
947  * +---------+   dtrace_stop(),  +----------+     dtrace_stop(),       |
948  *                 after END                       before END          |
949  *                                                                     |
950  *                                +--------+                           |
951  *                                | KILLED |<--------------------------+
952  *                                +--------+     deadman timeout
953  *
954  * Note that once a DTrace consumer has stopped tracing, there is no way to
955  * restart it; if a DTrace consumer wishes to restart tracing, it must reopen
956  * the DTrace pseudodevice.
957  */
958 typedef enum dtrace_activity {
959 	DTRACE_ACTIVITY_INACTIVE = 0,		/* not yet running */
960 	DTRACE_ACTIVITY_WARMUP,			/* while starting */
961 	DTRACE_ACTIVITY_ACTIVE,			/* running */
962 	DTRACE_ACTIVITY_DRAINING,		/* before stopping */
963 	DTRACE_ACTIVITY_COOLDOWN,		/* while stopping */
964 	DTRACE_ACTIVITY_STOPPED,		/* after stopping */
965 	DTRACE_ACTIVITY_KILLED			/* killed due to deadman */
966 } dtrace_activity_t;
967 
968 /*
969  * DTrace Helper Implementation
970  *
971  * A description of the helper architecture may be found in <sys/dtrace.h>.
972  * Each process contains a pointer to its helpers in its p_dtrace_helpers
973  * member.  This is a pointer to a dtrace_helpers structure, which contains an
974  * array of pointers to dtrace_helper structures, helper variable state (shared
975  * among a process's helpers) and a generation count.  (The generation count is
976  * used to provide an identifier when a helper is added so that it may be
977  * subsequently removed.)  The dtrace_helper structure is self-explanatory,
978  * containing pointers to the objects needed to execute the helper.  Note that
979  * helpers are _duplicated_ across fork(2), and destroyed on exec(2).  No more
980  * than dtrace_helpers_max are allowed per-process.
981  */
982 #define	DTRACE_HELPER_ACTION_USTACK	0
983 #define	DTRACE_NHELPER_ACTIONS		1
984 
985 typedef struct dtrace_helper_action {
986 	dtrace_difo_t *dthp_predicate;		/* helper action predicate */
987 	int dthp_nactions;			/* number of actions */
988 	dtrace_difo_t **dthp_actions;		/* array of actions */
989 	int dthp_generation;			/* helper action generation */
990 	struct dtrace_helper_action *dthp_next;	/* next helper action */
991 } dtrace_helper_action_t;
992 
993 typedef struct dtrace_helper_provider {
994 	dof_helper_t dthp_prov;			/* DOF w/ provider and probes */
995 	uint32_t dthp_ref;			/* reference count */
996 } dtrace_helper_provider_t;
997 
998 typedef struct dtrace_helpers {
999 	dtrace_helper_action_t **dthps_actions;	/* array of helper actions */
1000 	dtrace_vstate_t dthps_vstate;		/* helper action var. state */
1001 	dtrace_helper_provider_t **dthps_provs;	/* array of providers */
1002 	uint_t dthps_nprovs;			/* count of providers */
1003 	int dthps_generation;			/* current generation */
1004 	pid_t dthps_pid;			/* pid of associated proc */
1005 	struct dtrace_helpers *dthps_next;	/* next pointer */
1006 	struct dtrace_helpers *dthps_prev;	/* prev pointer */
1007 } dtrace_helpers_t;
1008 
1009 /*
1010  * DTrace Helper Action Tracing
1011  *
1012  * Debugging helper actions can be arduous.  To ease the development and
1013  * debugging of helpers, DTrace contains a tracing-framework-within-a-tracing-
1014  * framework: helper tracing.  If dtrace_helptrace_enabled is non-zero (which
1015  * it is by default on DEBUG kernels), all helper activity will be traced to a
1016  * global, in-kernel ring buffer.  Each entry includes a pointer to the specific
1017  * helper, the location within the helper, and a trace of all local variables.
1018  * The ring buffer may be displayed in a human-readable format with the
1019  * ::dtrace_helptrace mdb(1) dcmd.
1020  */
1021 #define	DTRACE_HELPTRACE_NEXT	(-1)
1022 #define	DTRACE_HELPTRACE_DONE	(-2)
1023 #define	DTRACE_HELPTRACE_ERR	(-3)
1024 
1025 typedef struct dtrace_helptrace {
1026 	dtrace_helper_action_t	*dtht_helper;	/* helper action */
1027 	int dtht_where;				/* where in helper action */
1028 	int dtht_nlocals;			/* number of locals */
1029 	uint64_t dtht_locals[1];		/* local variables */
1030 } dtrace_helptrace_t;
1031 
1032 /*
1033  * DTrace Credentials
1034  *
1035  * In probe context, we don't have the flexibility to examine the credentials
1036  * of the DTrace consumer that created a particular enabling.  Instead, we use
1037  * the Least Privilege interfaces to cache the consumer's credentials in a
1038  * dtrace_cred_t structure. That structure contains two important sets of
1039  * credentials that limit the consumer's breadth of visibility and what
1040  * actions the consumer may take.
1041  */
1042 #define	DTRACE_CRV_ALLPROC		0x01
1043 #define	DTRACE_CRV_KERNEL		0x02
1044 
1045 #define	DTRACE_CRV_ALL		(DTRACE_CRV_ALLPROC | DTRACE_CRV_KERNEL)
1046 
1047 #define	DTRACE_CRA_PROC			0x0001
1048 #define	DTRACE_CRA_PROC_DESTRUCTIVE	0x0002
1049 #define	DTRACE_CRA_PROC_CONTROL		0x0004
1050 #define	DTRACE_CRA_KERNEL		0x0008
1051 #define	DTRACE_CRA_KERNEL_DESTRUCTIVE	0x0010
1052 
1053 #define	DTRACE_CRA_ALL		(DTRACE_CRA_PROC | \
1054 	DTRACE_CRA_PROC_DESTRUCTIVE | DTRACE_CRA_PROC_CONTROL | \
1055 	DTRACE_CRA_KERNEL | DTRACE_CRA_KERNEL_DESTRUCTIVE)
1056 
1057 typedef struct dtrace_cred {
1058 	uid_t			dcr_uid;
1059 	gid_t			dcr_gid;
1060 	uint8_t			dcr_destructive;
1061 	uint8_t			dcr_visible;
1062 	uint16_t		dcr_action;
1063 } dtrace_cred_t;
1064 
1065 /*
1066  * DTrace Consumer State
1067  *
1068  * Each DTrace consumer has an associated dtrace_state structure that contains
1069  * its in-kernel DTrace state -- including options, credentials, statistics and
1070  * pointers to ECBs, buffers, speculations and formats.  A dtrace_state
1071  * structure is also allocated for anonymous enablings.  When anonymous state
1072  * is grabbed, the grabbing consumers dts_anon pointer is set to the grabbed
1073  * dtrace_state structure.
1074  */
1075 struct dtrace_state {
1076 	dev_t dts_dev;				/* device */
1077 	int dts_necbs;				/* total number of ECBs */
1078 	dtrace_ecb_t **dts_ecbs;		/* array of ECBs */
1079 	dtrace_epid_t dts_epid;			/* next EPID to allocate */
1080 	size_t dts_needed;			/* greatest needed space */
1081 	struct dtrace_state *dts_anon;		/* anon. state, if grabbed */
1082 	dtrace_activity_t dts_activity;		/* current activity */
1083 	dtrace_vstate_t dts_vstate;		/* variable state */
1084 	dtrace_buffer_t *dts_buffer;		/* principal buffer */
1085 	dtrace_buffer_t *dts_aggbuffer;		/* aggregation buffer */
1086 	dtrace_speculation_t *dts_speculations;	/* speculation array */
1087 	int dts_nspeculations;			/* number of speculations */
1088 	int dts_naggregations;			/* number of aggregations */
1089 	dtrace_aggregation_t **dts_aggregations; /* aggregation array */
1090 	vmem_t *dts_aggid_arena;		/* arena for aggregation IDs */
1091 	uint32_t dts_speculations_busy;		/* number of spec. busy */
1092 	uint32_t dts_speculations_unavail;	/* number of spec unavail */
1093 	uint64_t dts_errors;			/* total number of errors */
1094 	uint32_t dts_reserve;			/* space reserved for END */
1095 	hrtime_t dts_laststatus;		/* time of last status */
1096 	cyclic_id_t dts_cleaner;		/* cleaning cyclic */
1097 	cyclic_id_t dts_deadman;		/* deadman cyclic */
1098 	hrtime_t dts_alive;			/* time last alive */
1099 	char dts_speculates;			/* boolean: has speculations */
1100 	char dts_destructive;			/* boolean: has dest. actions */
1101 	int dts_nformats;			/* number of formats */
1102 	char **dts_formats;			/* format string array */
1103 	dtrace_optval_t dts_options[DTRACEOPT_MAX]; /* options */
1104 	dtrace_cred_t dts_cred;			/* credentials */
1105 	size_t dts_nretained;			/* number of retained enabs */
1106 };
1107 
1108 struct dtrace_provider {
1109 	dtrace_pattr_t dtpv_attr;		/* provider attributes */
1110 	dtrace_ppriv_t dtpv_priv;		/* provider privileges */
1111 	dtrace_pops_t dtpv_pops;		/* provider operations */
1112 	char *dtpv_name;			/* provider name */
1113 	void *dtpv_arg;				/* provider argument */
1114 	uint_t dtpv_defunct;			/* boolean: defunct provider */
1115 	struct dtrace_provider *dtpv_next;	/* next provider */
1116 };
1117 
1118 struct dtrace_meta {
1119 	dtrace_mops_t dtm_mops;			/* meta provider operations */
1120 	char *dtm_name;				/* meta provider name */
1121 	void *dtm_arg;				/* meta provider user arg */
1122 	uint64_t dtm_count;			/* no. of associated provs. */
1123 };
1124 
1125 /*
1126  * DTrace Enablings
1127  *
1128  * A dtrace_enabling structure is used to track a collection of ECB
1129  * descriptions -- before they have been turned into actual ECBs.  This is
1130  * created as a result of DOF processing, and is generally used to generate
1131  * ECBs immediately thereafter.  However, enablings are also generally
1132  * retained should the probes they describe be created at a later time; as
1133  * each new module or provider registers with the framework, the retained
1134  * enablings are reevaluated, with any new match resulting in new ECBs.  To
1135  * prevent probes from being matched more than once, the enabling tracks the
1136  * last probe generation matched, and only matches probes from subsequent
1137  * generations.
1138  */
1139 typedef struct dtrace_enabling {
1140 	dtrace_ecbdesc_t **dten_desc;		/* all ECB descriptions */
1141 	int dten_ndesc;				/* number of ECB descriptions */
1142 	int dten_maxdesc;			/* size of ECB array */
1143 	dtrace_vstate_t *dten_vstate;		/* associated variable state */
1144 	dtrace_genid_t dten_probegen;		/* matched probe generation */
1145 	dtrace_ecbdesc_t *dten_current;		/* current ECB description */
1146 	int dten_error;				/* current error value */
1147 	int dten_primed;			/* boolean: set if primed */
1148 	struct dtrace_enabling *dten_prev;	/* previous enabling */
1149 	struct dtrace_enabling *dten_next;	/* next enabling */
1150 } dtrace_enabling_t;
1151 
1152 /*
1153  * DTrace Anonymous Enablings
1154  *
1155  * Anonymous enablings are DTrace enablings that are not associated with a
1156  * controlling process, but rather derive their enabling from DOF stored as
1157  * properties in the dtrace.conf file.  If there is an anonymous enabling, a
1158  * DTrace consumer state and enabling are created on attach.  The state may be
1159  * subsequently grabbed by the first consumer specifying the "grabanon"
1160  * option.  As long as an anonymous DTrace enabling exists, dtrace(7D) will
1161  * refuse to unload.
1162  */
1163 typedef struct dtrace_anon {
1164 	dtrace_state_t *dta_state;		/* DTrace consumer state */
1165 	dtrace_enabling_t *dta_enabling;	/* pointer to enabling */
1166 	processorid_t dta_beganon;		/* which CPU BEGIN ran on */
1167 } dtrace_anon_t;
1168 
1169 /*
1170  * DTrace Error Debugging
1171  */
1172 #ifdef DEBUG
1173 #define	DTRACE_ERRDEBUG
1174 #endif
1175 
1176 #ifdef DTRACE_ERRDEBUG
1177 
1178 typedef struct dtrace_errhash {
1179 	const char	*dter_msg;	/* error message */
1180 	int		dter_count;	/* number of times seen */
1181 } dtrace_errhash_t;
1182 
1183 #define	DTRACE_ERRHASHSZ	256	/* must be > number of err msgs */
1184 
1185 #endif	/* DTRACE_ERRDEBUG */
1186 
1187 /*
1188  * DTrace Toxic Ranges
1189  *
1190  * DTrace supports safe loads from probe context; if the address turns out to
1191  * be invalid, a bit will be set by the kernel indicating that DTrace
1192  * encountered a memory error, and DTrace will propagate the error to the user
1193  * accordingly.  However, there may exist some regions of memory in which an
1194  * arbitrary load can change system state, and from which it is impossible to
1195  * recover from such a load after it has been attempted.  Examples of this may
1196  * include memory in which programmable I/O registers are mapped (for which a
1197  * read may have some implications for the device) or (in the specific case of
1198  * UltraSPARC-I and -II) the virtual address hole.  The platform is required
1199  * to make DTrace aware of these toxic ranges; DTrace will then check that
1200  * target addresses are not in a toxic range before attempting to issue a
1201  * safe load.
1202  */
1203 typedef struct dtrace_toxrange {
1204 	uintptr_t	dtt_base;		/* base of toxic range */
1205 	uintptr_t	dtt_limit;		/* limit of toxic range */
1206 } dtrace_toxrange_t;
1207 
1208 extern uint64_t dtrace_getarg(int, int);
1209 extern greg_t dtrace_getfp(void);
1210 extern int dtrace_getipl(void);
1211 extern uintptr_t dtrace_caller(int);
1212 extern uint32_t dtrace_cas32(uint32_t *, uint32_t, uint32_t);
1213 extern void *dtrace_casptr(void *, void *, void *);
1214 extern void dtrace_copyin(uintptr_t, uintptr_t, size_t);
1215 extern void dtrace_copyinstr(uintptr_t, uintptr_t, size_t);
1216 extern void dtrace_copyout(uintptr_t, uintptr_t, size_t);
1217 extern void dtrace_copyoutstr(uintptr_t, uintptr_t, size_t);
1218 extern void dtrace_getpcstack(pc_t *, int, int, uint32_t *);
1219 extern ulong_t dtrace_getreg(struct regs *, uint_t);
1220 extern int dtrace_getstackdepth(int);
1221 extern void dtrace_getupcstack(uint64_t *, int);
1222 extern void dtrace_getufpstack(uint64_t *, uint64_t *, int);
1223 extern int dtrace_getustackdepth(void);
1224 extern uintptr_t dtrace_fulword(void *);
1225 extern uint8_t dtrace_fuword8(void *);
1226 extern uint16_t dtrace_fuword16(void *);
1227 extern uint32_t dtrace_fuword32(void *);
1228 extern uint64_t dtrace_fuword64(void *);
1229 extern void dtrace_probe_error(dtrace_state_t *, dtrace_epid_t, int, int,
1230     int, uintptr_t);
1231 extern int dtrace_assfail(const char *, const char *, int);
1232 extern int dtrace_attached(void);
1233 extern hrtime_t dtrace_gethrestime();
1234 
1235 #ifdef __sparc
1236 extern void dtrace_flush_windows(void);
1237 extern void dtrace_flush_user_windows(void);
1238 extern uint_t dtrace_getotherwin(void);
1239 extern uint_t dtrace_getfprs(void);
1240 #else
1241 extern void dtrace_copy(uintptr_t, uintptr_t, size_t);
1242 extern void dtrace_copystr(uintptr_t, uintptr_t, size_t);
1243 #endif
1244 
1245 /*
1246  * DTrace Assertions
1247  *
1248  * DTrace calls ASSERT from probe context.  To assure that a failed ASSERT
1249  * does not induce a markedly more catastrophic failure (e.g., one from which
1250  * a dump cannot be gleaned), DTrace must define its own ASSERT to be one that
1251  * may safely be called from probe context.  This header file must thus be
1252  * included by any DTrace component that calls ASSERT from probe context, and
1253  * _only_ by those components.  (The only exception to this is kernel
1254  * debugging infrastructure at user-level that doesn't depend on calling
1255  * ASSERT.)
1256  */
1257 #undef ASSERT
1258 #ifdef DEBUG
1259 #define	ASSERT(EX)	((void)((EX) || \
1260 			dtrace_assfail(#EX, __FILE__, __LINE__)))
1261 #else
1262 #define	ASSERT(X)	((void)0)
1263 #endif
1264 
1265 #ifdef	__cplusplus
1266 }
1267 #endif
1268 
1269 #endif /* _SYS_DTRACE_IMPL_H */
1270