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 (the "License").
6 * You may not use this file except in compliance with the License.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26#ifndef _THR_UBERDATA_H
27#define	_THR_UBERDATA_H
28
29#include <stdlib.h>
30#include <unistd.h>
31#include <sys/types.h>
32#include <fcntl.h>
33#include <string.h>
34#include <signal.h>
35#include <ucontext.h>
36#include <thread.h>
37#include <pthread.h>
38#include <atomic.h>
39#include <link.h>
40#include <sys/resource.h>
41#include <sys/lwp.h>
42#include <errno.h>
43#include <sys/asm_linkage.h>
44#include <sys/regset.h>
45#include <sys/fcntl.h>
46#include <sys/mman.h>
47#include <synch.h>
48#include <door.h>
49#include <limits.h>
50#include <sys/synch32.h>
51#include <schedctl.h>
52#include <sys/priocntl.h>
53#include <thread_db.h>
54#include <setjmp.h>
55#include "libc_int.h"
56#include "tdb_agent.h"
57#include "thr_debug.h"
58
59/*
60 * This is an implementation-specific include file for threading support.
61 * It is not to be seen by the clients of the library.
62 *
63 * This file also describes uberdata in libc.
64 *
65 * The term "uberdata" refers to data that is unique and visible across
66 * all link maps.  The name is meant to imply that such data is truly
67 * global, not just locally global to a particular link map.
68 *
69 * See the Linker and Libraries Guide for a full description of alternate
70 * link maps and how they are set up and used.
71 *
72 * Alternate link maps implement multiple global namespaces within a single
73 * process.  There may be multiple instances of identical dynamic libraries
74 * loaded in a process's address space at the same time, each on a different
75 * link map (as determined by the dynamic linker), each with its own set of
76 * global variables.  Which particular instance of a global variable is seen
77 * by a thread running in the process is determined by the link map on which
78 * the thread happens to be executing at the time.
79 *
80 * However, there are aspects of a process that are unique across all
81 * link maps, in particular the structures used to implement threads
82 * of control (in Sparc terminology, there is only one %g7 regardless
83 * of the link map on which the thread is executing).
84 *
85 * All uberdata is referenced from a base pointer in the thread's ulwp_t
86 * structure (which is also uberdata).  All allocations and deallocations
87 * of uberdata are made via the uberdata-aware lmalloc() and lfree()
88 * interfaces (malloc() and free() are simply locally-global).
89 */
90
91/*
92 * Special libc-private access to errno.
93 * We do this so that references to errno do not invoke the dynamic linker.
94 */
95#undef errno
96#define	errno (*curthread->ul_errnop)
97
98/*
99 * See <sys/synch32.h> for the reasons for these values
100 * and why they are different for sparc and intel.
101 */
102#if defined(__sparc)
103
104/* lock.lock64.pad[x]	   4 5 6 7 */
105#define	LOCKMASK	0xff000000
106#define	WAITERMASK	0x000000ff
107#define	SPINNERMASK	0x00ff0000
108#define	SPINNERSHIFT	16
109#define	WAITER		0x00000001
110#define	LOCKSET		0xff
111#define	LOCKCLEAR	0
112
113#define	PIDSHIFT	32
114#define	LOCKMASK64	0xffffffffff000000ULL
115#define	LOCKBYTE64	0x00000000ff000000ULL
116#define	WAITERMASK64	0x00000000000000ffULL
117#define	SPINNERMASK64	0x0000000000ff0000ULL
118
119#elif defined(__x86)
120
121/* lock.lock64.pad[x]	   7 6 5 4 */
122#define	LOCKMASK	0xff000000
123#define	WAITERMASK	0x00ff0000
124#define	SPINNERMASK	0x0000ff00
125#define	SPINNERSHIFT	8
126#define	WAITER		0x00010000
127#define	LOCKSET		0x01
128#define	LOCKCLEAR	0
129
130#define	PIDSHIFT	0
131#define	LOCKMASK64	0xff000000ffffffffULL
132#define	LOCKBYTE64	0x0100000000000000ULL
133#define	WAITERMASK64	0x00ff000000000000ULL
134#define	SPINNERMASK64	0x0000ff0000000000ULL
135
136#else
137#error "neither __sparc nor __x86 is defined"
138#endif
139
140/*
141 * Fetch the owner of a USYNC_THREAD mutex.
142 * Don't use this with process-shared mutexes;
143 * the owing thread may be in a different process.
144 */
145#define	MUTEX_OWNER(mp)	((ulwp_t *)(uintptr_t)(mp)->mutex_owner)
146
147/*
148 * Test if a thread owns a process-private (USYNC_THREAD) mutex.
149 * This is inappropriate for a process-shared (USYNC_PROCESS) mutex.
150 * The 'mp' argument must not have side-effects since it is evaluated twice.
151 */
152#define	MUTEX_OWNED(mp, thrp)	\
153	((mp)->mutex_lockw != 0 && MUTEX_OWNER(mp) == thrp)
154
155
156/*
157 * uberflags.uf_tdb_register_sync is an interface with libc_db to enable the
158 * collection of lock statistics by a debugger or other collecting tool.
159 *
160 * uberflags.uf_thread_error_detection is set by an environment variable:
161 *	_THREAD_ERROR_DETECTION
162 *		0 == no detection of locking primitive errors.
163 *		1 == detect errors and issue a warning message.
164 *		2 == detect errors, issue a warning message, and dump core.
165 *
166 * We bundle these together in uberflags.uf_trs_ted to make a test of either
167 * being non-zero a single memory reference (for speed of mutex_lock(), etc).
168 *
169 * uberflags.uf_mt is set non-zero when the first thread (in addition
170 * to the main thread) is created.
171 *
172 * We bundle all these flags together in uberflags.uf_all to make a test
173 * of any being non-zero a single memory reference (again, for speed).
174 */
175typedef union {
176	int	uf_all;			/* combined all flags */
177	struct {
178		short	h_pad;
179		short	h_trs_ted;	/* combined reg sync & error detect */
180	} uf_h;
181	struct {
182		char	x_mt;
183		char	x_pad;
184		char	x_tdb_register_sync;
185		char	x_thread_error_detection;
186	} uf_x;
187} uberflags_t;
188
189#define	uf_mt				uf_x.x_mt
190#define	uf_tdb_register_sync		uf_x.x_tdb_register_sync
191#define	uf_thread_error_detection	uf_x.x_thread_error_detection
192#define	uf_trs_ted			uf_h.h_trs_ted	/* both of the above */
193
194/*
195 * NOTE WELL:
196 * To enable further optimization, the "ul_schedctl_called" member
197 * of the ulwp_t structure (below) serves double-duty:
198 *	1. If NULL, it means that the thread must call __schedctl()
199 *	   to set up its schedctl mappings before acquiring a mutex.
200 *	   This is required by the implementation of adaptive mutex locking.
201 *	2. If non-NULL, it points to uberdata.uberflags, so that tests of
202 *	   uberflags can be made without additional memory references.
203 * This allows the common case of _mutex_lock() and _mutex_unlock() for
204 * USYNC_THREAD mutexes with no error detection and no lock statistics
205 * to be optimized for speed.
206 */
207
208/* double the default stack size for 64-bit processes */
209#ifdef _LP64
210#define	MINSTACK	(8 * 1024)
211#define	DEFAULTSTACK	(2 * 1024 * 1024)
212#else
213#define	MINSTACK	(4 * 1024)
214#define	DEFAULTSTACK	(1024 * 1024)
215#endif
216
217#define	MUTEX_TRY	0
218#define	MUTEX_LOCK	1
219#define	MUTEX_NOCEIL	0x40
220
221#if defined(__x86)
222
223typedef struct {	/* structure returned by fnstenv */
224	int	fctrl;		/* control word */
225	int	fstat;		/* status word (flags, etc) */
226	int	ftag;		/* tag of which regs busy */
227	int	misc[4];	/* other stuff, 28 bytes total */
228} fpuenv_t;
229
230#ifdef _SYSCALL32
231typedef fpuenv_t fpuenv32_t;
232#endif	/* _SYSCALL32 */
233
234#elif defined(__sparc)
235
236typedef struct {	/* fp state structure */
237	greg_t	fsr;
238	greg_t	fpu_en;
239} fpuenv_t;
240
241#ifdef _SYSCALL32
242typedef struct {
243	greg32_t	fsr;
244	greg32_t	fpu_en;
245} fpuenv32_t;
246#endif	/* _SYSCALL32 */
247
248#endif	/* __x86 */
249
250#if defined(__x86)
251extern	void	ht_pause(void);		/* "pause" instruction */
252#define	SMT_PAUSE()	ht_pause()
253#elif defined(SMT_PAUSE_FUNCTION)
254extern	void	SMT_PAUSE_FUNCTION(void);
255#define	SMT_PAUSE()	SMT_PAUSE_FUNCTION()
256#else
257#define	SMT_PAUSE()	smt_pause()
258#endif	/* __x86 */
259
260/*
261 * Cleanup handler related data.
262 * This structure is exported as _cleanup_t in pthread.h.
263 * pthread.h exports only the size of this structure, so check
264 * _cleanup_t in pthread.h before making any change here.
265 */
266typedef struct __cleanup {
267	struct __cleanup *next;		/* pointer to next handler */
268	caddr_t	fp;			/* current frame pointer */
269	void	(*func)(void *);	/* cleanup handler address */
270	void	*arg;			/* handler's argument */
271} __cleanup_t;
272
273/*
274 * Thread-Specific Data (TSD)
275 * TSD_NFAST includes the invalid key zero, so there
276 * are really only (TSD_NFAST - 1) fast key slots.
277 */
278typedef	void (*PFrV)(void *);
279#define	TSD_UNALLOCATED	((PFrV)1)
280#define	TSD_NFAST	9
281
282/*
283 * The tsd union is designed to burn a little memory (9 words) to make
284 * lookups blindingly fast.  Note that tsd_nalloc could be placed at the
285 * end of the pad region to increase the likelihood that it falls on the
286 * same cache line as the data.
287 */
288typedef union tsd {
289	uint_t tsd_nalloc;		/* Amount of allocated storage */
290	void *tsd_pad[TSD_NFAST];
291	void *tsd_data[1];
292} tsd_t;
293
294typedef struct {
295	mutex_t tsdm_lock;		/* Lock protecting the data */
296	uint_t tsdm_nkeys;		/* Number of allocated keys */
297	uint_t tsdm_nused;		/* Number of used keys */
298	PFrV *tsdm_destro;		/* Per-key destructors */
299	char tsdm_pad[64 -		/* pad to 64 bytes */
300		(sizeof (mutex_t) + 2 * sizeof (uint_t) + sizeof (PFrV *))];
301} tsd_metadata_t;
302
303#ifdef _SYSCALL32
304typedef union tsd32 {
305	uint_t tsd_nalloc;		/* Amount of allocated storage */
306	caddr32_t tsd_pad[TSD_NFAST];
307	caddr32_t tsd_data[1];
308} tsd32_t;
309
310typedef struct {
311	mutex_t tsdm_lock;		/* Lock protecting the data */
312	uint_t tsdm_nkeys;		/* Number of allocated keys */
313	uint_t tsdm_nused;		/* Number of used keys */
314	caddr32_t tsdm_destro;		/* Per-key destructors */
315	char tsdm_pad[64 -		/* pad to 64 bytes */
316		(sizeof (mutex_t) + 2 * sizeof (uint_t) + sizeof (caddr32_t))];
317} tsd_metadata32_t;
318#endif	/* _SYSCALL32 */
319
320
321/*
322 * Thread-Local Storage (TLS)
323 */
324typedef struct {
325	void		*tls_data;
326	size_t		tls_size;
327} tls_t;
328
329typedef struct {
330	mutex_t	tls_lock;		/* Lock protecting the data */
331	tls_t	tls_modinfo;		/* Root of all TLS_modinfo data */
332	tls_t	static_tls;		/* Template for static TLS */
333	char	tls_pad[64 -		/* pad to 64 bytes */
334		(sizeof (mutex_t) + 2 * sizeof (tls_t))];
335} tls_metadata_t;
336
337#ifdef _SYSCALL32
338typedef struct {
339	caddr32_t	tls_data;
340	size32_t	tls_size;
341} tls32_t;
342
343typedef struct {
344	mutex_t	tls_lock;		/* Lock protecting the data */
345	tls32_t	tls_modinfo;		/* Root of all TLS_modinfo data */
346	tls32_t	static_tls;		/* Template for static TLS */
347	char	tls_pad[64 -		/* pad to 64 bytes */
348		(sizeof (mutex_t) + 2 * sizeof (tls32_t))];
349} tls_metadata32_t;
350#endif	/* _SYSCALL32 */
351
352
353/*
354 * Sleep queue root for USYNC_THREAD condvars and mutexes.
355 * There is a default queue root for each queue head (see below).
356 * Also, each ulwp_t contains a queue root that can be used
357 * when the thread is enqueued on the queue, if necessary
358 * (when more than one wchan hashes to the same queue head).
359 */
360typedef struct queue_root {
361	struct queue_root	*qr_next;
362	struct queue_root	*qr_prev;
363	struct ulwp		*qr_head;
364	struct ulwp		*qr_tail;
365	void			*qr_wchan;
366	uint32_t		qr_rtcount;
367	uint32_t		qr_qlen;
368	uint32_t		qr_qmax;
369} queue_root_t;
370
371#ifdef _SYSCALL32
372typedef struct queue_root32 {
373	caddr32_t		qr_next;
374	caddr32_t		qr_prev;
375	caddr32_t		qr_head;
376	caddr32_t		qr_tail;
377	caddr32_t		qr_wchan;
378	uint32_t		qr_rtcount;
379	uint32_t		qr_qlen;
380	uint32_t		qr_qmax;
381} queue_root32_t;
382#endif
383
384/*
385 * Sleep queue heads for USYNC_THREAD condvars and mutexes.
386 * The size and alignment is 128 bytes to reduce cache conflicts.
387 * Each queue head points to a list of queue roots, defined above.
388 * Each queue head contains a default queue root for use when only one
389 * is needed.  It is always at the tail of the queue root hash chain.
390 */
391typedef union {
392	uint64_t		qh_64[16];
393	struct {
394		mutex_t		q_lock;
395		uint8_t		q_qcnt;
396		uint8_t		q_type;		/* MX or CV */
397		uint8_t		q_pad1[2];
398		uint32_t	q_lockcount;
399		uint32_t	q_qlen;
400		uint32_t	q_qmax;
401		void		*q_wchan;	/* valid only while locked */
402		struct queue_root *q_root;	/* valid only while locked */
403		struct queue_root *q_hlist;
404#if !defined(_LP64)
405		caddr_t		q_pad2[3];
406#endif
407		queue_root_t	q_def_root;
408		uint32_t	q_hlen;
409		uint32_t	q_hmax;
410	} qh_qh;
411} queue_head_t;
412
413#define	qh_lock		qh_qh.q_lock
414#define	qh_qcnt		qh_qh.q_qcnt
415#define	qh_type		qh_qh.q_type
416#if defined(THREAD_DEBUG)
417#define	qh_lockcount	qh_qh.q_lockcount
418#define	qh_qlen		qh_qh.q_qlen
419#define	qh_qmax		qh_qh.q_qmax
420#endif
421#define	qh_wchan	qh_qh.q_wchan
422#define	qh_root		qh_qh.q_root
423#define	qh_hlist	qh_qh.q_hlist
424#define	qh_def_root	qh_qh.q_def_root
425#define	qh_hlen		qh_qh.q_hlen
426#define	qh_hmax		qh_qh.q_hmax
427
428/* queue types passed to queue_lock() */
429#define	MX	0
430#define	CV	1
431#define	QHASHSHIFT	9			/* number of hashing bits */
432#define	QHASHSIZE	(1 << QHASHSHIFT)	/* power of 2 (1<<9 == 512) */
433#define	QUEUE_HASH(wchan, type)	((uint_t)			\
434	((((uintptr_t)(wchan) >> 3)				\
435	^ ((uintptr_t)(wchan) >> (QHASHSHIFT + 3)))		\
436	& (QHASHSIZE - 1)) + (((type) == MX)? 0 : QHASHSIZE))
437
438extern	queue_head_t	*queue_lock(void *, int);
439extern	void		queue_unlock(queue_head_t *);
440extern	void		enqueue(queue_head_t *, struct ulwp *, int);
441extern	struct ulwp	*dequeue(queue_head_t *, int *);
442extern	struct ulwp	**queue_slot(queue_head_t *, struct ulwp **, int *);
443extern	struct ulwp	*queue_waiter(queue_head_t *);
444extern	int		dequeue_self(queue_head_t *);
445extern	void		queue_unlink(queue_head_t *,
446				struct ulwp **, struct ulwp *);
447extern	void		unsleep_self(void);
448extern	void		spin_lock_set(mutex_t *);
449extern	void		spin_lock_clear(mutex_t *);
450
451/*
452 * Scheduling class information structure.
453 */
454typedef struct {
455	short		pcc_state;
456	short		pcc_policy;
457	pri_t		pcc_primin;
458	pri_t		pcc_primax;
459	pcinfo_t	pcc_info;
460} pcclass_t;
461
462/*
463 * Memory block for chain of owned ceiling mutexes.
464 */
465typedef struct mxchain {
466	struct mxchain	*mxchain_next;
467	mutex_t		*mxchain_mx;
468} mxchain_t;
469
470/*
471 * Pointer to an rwlock that is held for reading.
472 * Used in rw_rdlock() to allow a thread that already holds a read
473 * lock to acquire another read lock on the same rwlock even if
474 * there are writers waiting.  This to avoid deadlock when acquiring
475 * a read lock more than once in the presence of pending writers.
476 * POSIX mandates this behavior.
477 */
478typedef struct {
479	void	*rd_rwlock;	/* the rwlock held for reading */
480	size_t	rd_count;	/* count of read locks applied */
481} readlock_t;
482
483#ifdef _SYSCALL32
484typedef struct {
485	caddr32_t	rd_rwlock;
486	size32_t	rd_count;
487} readlock32_t;
488#endif	/* _SYSCALL32 */
489
490/*
491 * Maximum number of read locks allowed for one thread on one rwlock.
492 * This could be as large as INT_MAX, but the SUSV3 test suite would
493 * take an inordinately long time to complete.  This is big enough.
494 */
495#define	READ_LOCK_MAX	100000
496
497#define	ul_tlsent	ul_tls.tls_data	/* array of pointers to dynamic TLS */
498#define	ul_ntlsent	ul_tls.tls_size	/* number of entries in ul_tlsent */
499
500/*
501 * Round up an integral value to a multiple of 64
502 */
503#define	roundup64(x)	(-(-(x) & -64))
504
505/*
506 * NOTE:  Whatever changes are made to ulwp_t must be
507 * reflected in $SRC/cmd/mdb/common/modules/libc/libc.c
508 *
509 * NOTE: ul_self *must* be the first member of ulwp_t on x86
510 * Low-level x86 code relies on this.
511 */
512typedef struct ulwp {
513	/*
514	 * These members always need to come first on sparc.
515	 * For dtrace, a ulwp_t must be aligned on a 64-byte boundary.
516	 */
517#if defined(__sparc)
518	uint32_t	ul_dinstr;	/* scratch space for dtrace */
519	uint32_t	ul_padsparc0[15];
520	uint32_t	ul_dsave;	/* dtrace: save %g1, %g0, %sp */
521	uint32_t	ul_drestore;	/* dtrace: restore %g0, %g0, %g0 */
522	uint32_t	ul_dftret;	/* dtrace: return probe fasttrap */
523	uint32_t	ul_dreturn;	/* dtrace: return %o0 */
524#endif
525	struct ulwp	*ul_self;	/* pointer to self */
526#if defined(__i386)
527	uint8_t		ul_dinstr[40];	/* scratch space for dtrace */
528#elif defined(__amd64)
529	uint8_t		ul_dinstr[56];	/* scratch space for dtrace */
530#endif
531	struct uberdata *ul_uberdata;	/* uber (super-global) data */
532	tls_t		ul_tls;		/* dynamic thread-local storage base */
533	struct ulwp	*ul_forw;	/* forw, back all_lwps list, */
534	struct ulwp	*ul_back;	/* protected by link_lock */
535	struct ulwp	*ul_next;	/* list to keep track of stacks */
536	struct ulwp	*ul_hash;	/* hash chain linked list */
537	void		*ul_rval;	/* return value from thr_exit() */
538	caddr_t		ul_stk;		/* mapping base of the stack */
539	size_t		ul_mapsiz;	/* mapping size of the stack */
540	size_t		ul_guardsize;	/* normally _lpagesize */
541	uintptr_t	ul_stktop;	/* broken thr_stksegment() interface */
542	size_t		ul_stksiz;	/* broken thr_stksegment() interface */
543	stack_t		ul_ustack;	/* current stack boundaries */
544	int		ul_ix;		/* hash index */
545	lwpid_t		ul_lwpid;	/* thread id, aka the lwp id */
546	pri_t		ul_pri;		/* scheduling priority */
547	pri_t		ul_epri;	/* real-time ceiling priority */
548	char		ul_policy;	/* scheduling policy */
549	char		ul_cid;		/* scheduling class id */
550	union {
551		struct {
552			char	cursig;	/* deferred signal number */
553			char	pleasestop; /* lwp requested to stop itself */
554		} s;
555		short	curplease;	/* for testing both at once */
556	} ul_cp;
557	char		ul_stop;	/* reason for stopping */
558	char		ul_signalled;	/* this lwp was cond_signal()d */
559	char		ul_dead;	/* this lwp has called thr_exit */
560	char		ul_unwind;	/* posix: unwind C++ stack */
561	char		ul_detached;	/* THR_DETACHED at thread_create() */
562					/* or pthread_detach() was called */
563	char		ul_writer;	/* sleeping in rw_wrlock() */
564	char		ul_stopping;	/* set by curthread: stopping self */
565	char		ul_cancel_prologue;	/* for _cancel_prologue() */
566	short		ul_preempt;	/* no_preempt()/preempt() */
567	short		ul_savpreempt;	/* pre-existing preempt value */
568	char		ul_sigsuspend;	/* thread is in sigsuspend/pollsys */
569	char		ul_main;	/* thread is the main thread */
570	char		ul_fork;	/* thread is performing a fork */
571	char		ul_primarymap;	/* primary link-map is initialized */
572	/* per-thread copies of the corresponding global variables */
573	uint8_t		ul_max_spinners;	/* thread_max_spinners */
574	char		ul_door_noreserve;	/* thread_door_noreserve */
575	char		ul_queue_fifo;		/* thread_queue_fifo */
576	char		ul_cond_wait_defer;	/* thread_cond_wait_defer */
577	char		ul_error_detection;	/* thread_error_detection */
578	char		ul_async_safe;		/* thread_async_safe */
579	char		ul_rt;			/* found on an RT queue */
580	char		ul_rtqueued;		/* was RT when queued */
581	char		ul_misaligned;		/* thread_locks_misaligned */
582	char		ul_pad[3];
583	int		ul_adaptive_spin;	/* thread_adaptive_spin */
584	int		ul_queue_spin;		/* thread_queue_spin */
585	volatile int	ul_critical;	/* non-zero == in a critical region */
586	int		ul_sigdefer;	/* non-zero == defer signals */
587	int		ul_vfork;	/* thread is the child of vfork() */
588	int		ul_cancelable;	/* _cancelon()/_canceloff() */
589	char		ul_cancel_pending;  /* pthread_cancel() was called */
590	char		ul_cancel_disabled; /* PTHREAD_CANCEL_DISABLE */
591	char		ul_cancel_async;    /* PTHREAD_CANCEL_ASYNCHRONOUS */
592	char		ul_save_async;	/* saved copy of ul_cancel_async */
593	char		ul_mutator;	/* lwp is a mutator (java interface) */
594	char		ul_created;	/* created suspended */
595	char		ul_replace;	/* replacement; must be free()d */
596	uchar_t		ul_nocancel;	/* cancellation can't happen */
597	int		ul_errno;	/* per-thread errno */
598	int		*ul_errnop;	/* pointer to errno or self->ul_errno */
599	__cleanup_t	*ul_clnup_hdr;	/* head of cleanup handlers list */
600	uberflags_t	*ul_schedctl_called;	/* ul_schedctl is set up */
601	volatile sc_shared_t *ul_schedctl;	/* schedctl data */
602	int		ul_bindflags;	/* bind_guard() interface to ld.so.1 */
603	uint_t		ul_libc_locks;	/* count of cancel_safe_mutex_lock()s */
604	tsd_t		*ul_stsd;	/* slow TLS for keys >= TSD_NFAST */
605	void		*ul_ftsd[TSD_NFAST]; /* fast TLS for keys < TSD_NFAST */
606	td_evbuf_t	ul_td_evbuf;	/* event buffer */
607	char		ul_td_events_enable;	/* event mechanism enabled */
608	char		ul_sync_obj_reg;	/* tdb_sync_obj_register() */
609	char		ul_qtype;	/* MX or CV */
610	char		ul_cv_wake;	/* != 0: just wake up, don't requeue */
611	int		ul_rtld;	/* thread is running inside ld.so.1 */
612	int		ul_usropts;	/* flags given to thr_create() */
613	void		*(*ul_startpc)(void *); /* start func (thr_create()) */
614	void		*ul_startarg;	/* argument for start function */
615	void		*ul_wchan;	/* synch object when sleeping */
616	struct ulwp	*ul_link;	/* sleep queue link */
617	queue_head_t	*ul_sleepq;	/* sleep queue thread is waiting on */
618	mutex_t		*ul_cvmutex;	/* mutex dropped when waiting on a cv */
619	mxchain_t	*ul_mxchain;	/* chain of owned ceiling mutexes */
620	int		ul_save_state;	/* bind_guard() interface to ld.so.1 */
621	uint_t		ul_rdlockcnt;	/* # entries in ul_readlock array */
622				/* 0 means there is but a single entry */
623	union {				/* single entry or pointer to array */
624		readlock_t	single;
625		readlock_t	*array;
626	} ul_readlock;
627	uint_t		ul_heldlockcnt;	/* # entries in ul_heldlocks array */
628				/* 0 means there is but a single entry */
629	union {				/* single entry or pointer to array */
630		mutex_t		*single;
631		mutex_t		**array;
632	} ul_heldlocks;
633	/* PROBE_SUPPORT begin */
634	void		*ul_tpdp;
635	/* PROBE_SUPPORT end */
636	ucontext_t	*ul_siglink;	/* pointer to previous context */
637	uint_t		ul_spin_lock_spin;	/* spin lock statistics */
638	uint_t		ul_spin_lock_spin2;
639	uint_t		ul_spin_lock_sleep;
640	uint_t		ul_spin_lock_wakeup;
641	queue_root_t	ul_queue_root;	/* root of a sleep queue */
642	id_t		ul_rtclassid;	/* real-time class id */
643	uint_t		ul_pilocks;	/* count of PI locks held */
644		/* the following members *must* be last in the structure */
645		/* they are discarded when ulwp is replaced on thr_exit() */
646	sigset_t	ul_sigmask;	/* thread's current signal mask */
647	sigset_t	ul_tmpmask;	/* signal mask for sigsuspend/pollsys */
648	siginfo_t	ul_siginfo;	/* deferred siginfo */
649	mutex_t		ul_spinlock;	/* used when suspending/continuing */
650	fpuenv_t	ul_fpuenv;	/* floating point state */
651	uintptr_t	ul_sp;		/* stack pointer when blocked */
652	void		*ul_ex_unwind;	/* address of _ex_unwind() or -1 */
653#if defined(sparc)
654	void		*ul_unwind_ret;	/* used only by _ex_clnup_handler() */
655#endif
656} ulwp_t;
657
658#define	ul_cursig	ul_cp.s.cursig		/* deferred signal number */
659#define	ul_pleasestop	ul_cp.s.pleasestop	/* lwp requested to stop */
660#define	ul_curplease	ul_cp.curplease		/* for testing both at once */
661
662/*
663 * This is the size of a replacement ulwp, retained only for the benefit
664 * of thr_join().  The trailing members are unneeded for this purpose.
665 */
666#define	REPLACEMENT_SIZE	((size_t)&((ulwp_t *)NULL)->ul_sigmask)
667
668/*
669 * Definitions for static initialization of signal sets,
670 * plus some sneaky optimizations in various places.
671 */
672
673#define	SIGMASK(sig)	((uint32_t)1 << (((sig) - 1) & (32 - 1)))
674
675#if (MAXSIG > (2 * 32) && MAXSIG <= (3 * 32))
676#define	FILLSET0	0xffffffffu
677#define	FILLSET1	0xffffffffu
678#define	FILLSET2	((1u << (MAXSIG - 64)) - 1)
679#define	FILLSET3	0
680#else
681#error "fix me: MAXSIG out of bounds"
682#endif
683
684#define	CANTMASK0	(SIGMASK(SIGKILL) | SIGMASK(SIGSTOP))
685#define	CANTMASK1	0
686#define	CANTMASK2	0
687#define	CANTMASK3	0
688
689#define	MASKSET0	(FILLSET0 & ~CANTMASK0)
690#define	MASKSET1	(FILLSET1 & ~CANTMASK1)
691#define	MASKSET2	(FILLSET2 & ~CANTMASK2)
692#define	MASKSET3	(FILLSET3 & ~CANTMASK3)
693
694extern	const sigset_t maskset;		/* set of all maskable signals */
695
696extern	int	thread_adaptive_spin;
697extern	uint_t	thread_max_spinners;
698extern	int	thread_queue_spin;
699extern	int	thread_queue_fifo;
700extern	int	thread_queue_dump;
701extern	int	thread_cond_wait_defer;
702extern	int	thread_async_safe;
703extern	int	thread_queue_verify;
704
705/*
706 * pthread_atfork() related data, used to store atfork handlers.
707 */
708typedef struct atfork {
709	struct atfork *forw;		/* forward pointer */
710	struct atfork *back;		/* backward pointer */
711	void (*prepare)(void);		/* pre-fork handler */
712	void (*parent)(void);		/* post-fork parent handler */
713	void (*child)(void);		/* post-fork child handler */
714} atfork_t;
715
716/*
717 * Element in the table and in the list of registered process
718 * robust locks.  We keep track of these to make sure that we
719 * only call ___lwp_mutex_register() once for each such lock
720 * after it is first mapped in (or newly mapped in).
721 */
722typedef struct robust {
723	struct robust	*robust_next;	/* hash table list */
724	struct robust	*robust_list;	/* global list */
725	mutex_t		*robust_lock;
726} robust_t;
727
728/*
729 * Invalid address, used to mark an unused element in the hash table.
730 */
731#define	INVALID_ADDR	((void *)(uintptr_t)(-1L))
732
733/*
734 * Parameters of the lock registration hash table.
735 */
736#define	LOCKSHIFT	15			/* number of hashing bits */
737#define	LOCKHASHSZ	(1 << LOCKSHIFT)	/* power of 2 (1<<15 == 32K) */
738#define	LOCK_HASH(addr)	(uint_t)			\
739	((((uintptr_t)(addr) >> 3)			\
740	^ ((uintptr_t)(addr) >> (LOCKSHIFT + 3)))	\
741	& (LOCKHASHSZ - 1))
742
743/*
744 * Make our hot locks reside on private cache lines (64 bytes).
745 */
746typedef struct {
747	mutex_t	pad_lock;
748	char	pad_pad[64 - sizeof (mutex_t)];
749} pad_lock_t;
750
751/*
752 * Make our semi-hot locks reside on semi-private cache lines (32 bytes).
753 */
754typedef struct {
755	mutex_t	pad_lock;
756	char	pad_pad[32 - sizeof (mutex_t)];
757} pad32_lock_t;
758
759/*
760 * The threads hash table is used for fast lookup and locking of an active
761 * thread structure (ulwp_t) given a thread-id.  It is an N-element array of
762 * thr_hash_table_t structures, where N == 1 before the main thread creates
763 * the first additional thread and N == 1024 afterwards.  Each element of the
764 * table is 64 bytes in size and alignment to reduce cache conflicts.
765 */
766typedef struct {
767	mutex_t	hash_lock;	/* lock per bucket */
768	cond_t	hash_cond;	/* convar per bucket */
769	ulwp_t	*hash_bucket;	/* hash bucket points to the list of ulwps */
770	char	hash_pad[64 -	/* pad out to 64 bytes */
771		(sizeof (mutex_t) + sizeof (cond_t) + sizeof (ulwp_t *))];
772} thr_hash_table_t;
773
774#ifdef _SYSCALL32
775typedef struct {
776	mutex_t	hash_lock;
777	cond_t	hash_cond;
778	caddr32_t hash_bucket;
779	char	hash_pad[64 -
780		(sizeof (mutex_t) + sizeof (cond_t) + sizeof (caddr32_t))];
781} thr_hash_table32_t;
782#endif	/* _SYSCALL32 */
783
784
785/*
786 * siguaction members have 128-byte size and 64-byte alignment.
787 * We know that sizeof (struct sigaction) is 32 bytes for both
788 * _ILP32 and _LP64 and that sizeof (rwlock_t) is 64 bytes.
789 */
790typedef struct {
791	rwlock_t	sig_lock;
792	struct sigaction sig_uaction;
793	char	sig_pad[128 - sizeof (rwlock_t) - sizeof (struct sigaction)];
794} siguaction_t;
795
796#ifdef _SYSCALL32
797typedef struct {
798	rwlock_t	sig_lock;
799	struct sigaction32 sig_uaction;
800	char	sig_pad[128 - sizeof (rwlock_t) - sizeof (struct sigaction32)];
801} siguaction32_t;
802#endif	/* _SYSCALL32 */
803
804
805/*
806 * Bucket structures, used by lmalloc()/lfree().
807 * See port/threads/alloc.c for details.
808 * A bucket's size and alignment is 64 bytes.
809 */
810typedef struct {
811	mutex_t	bucket_lock;	/* protects the free list allocations */
812	void	*free_list;	/* LIFO list of blocks to allocate/free */
813	size_t	chunks;		/* number of 64K blocks mmap()ed last time */
814	char	pad64[64 -	/* pad out to 64 bytes */
815		(sizeof (mutex_t) + sizeof (void *) + sizeof (size_t))];
816} bucket_t;
817
818#ifdef _SYSCALL32
819typedef struct {
820	mutex_t		bucket_lock;
821	caddr32_t	free_list;
822	size32_t	chunks;
823	char	pad64[64 -	/* pad out to 64 bytes */
824		(sizeof (mutex_t) + sizeof (caddr32_t) + sizeof (size32_t))];
825} bucket32_t;
826#endif	/* _SYSCALL32 */
827
828#define	NBUCKETS	10	/* sizes ranging from 64 to 32768 */
829
830
831/*
832 * atexit() data structures.
833 * See port/gen/atexit.c for details.
834 */
835typedef void (*_exithdlr_func_t) (void);
836
837typedef struct _exthdlr {
838	struct _exthdlr 	*next;	/* next in handler list */
839	_exithdlr_func_t	hdlr;	/* handler itself */
840} _exthdlr_t;
841
842typedef struct {
843	mutex_t		exitfns_lock;
844	_exthdlr_t	*head;
845	void		*exit_frame_monitor;
846	char		exit_pad[64 -	/* pad out to 64 bytes */
847		(sizeof (mutex_t) + sizeof (_exthdlr_t *) + sizeof (void *))];
848} atexit_root_t;
849
850#ifdef _SYSCALL32
851typedef struct {
852	mutex_t		exitfns_lock;
853	caddr32_t	head;
854	caddr32_t	exit_frame_monitor;
855	char		exit_pad[64 -	/* pad out to 64 bytes */
856		(sizeof (mutex_t) + sizeof (caddr32_t) + sizeof (caddr32_t))];
857} atexit_root32_t;
858#endif	/* _SYSCALL32 */
859
860
861/*
862 * This is data that is global to all link maps (uberdata, aka super-global).
863 */
864typedef struct uberdata {
865	pad_lock_t	_link_lock;
866	pad_lock_t	_ld_lock;
867	pad_lock_t	_fork_lock;
868	pad_lock_t	_atfork_lock;
869	pad32_lock_t	_callout_lock;
870	pad32_lock_t	_tdb_hash_lock;
871	tdb_sync_stats_t tdb_hash_lock_stats;
872	siguaction_t	siguaction[NSIG];
873	bucket_t	bucket[NBUCKETS];
874	atexit_root_t	atexit_root;
875	tsd_metadata_t	tsd_metadata;
876	tls_metadata_t	tls_metadata;
877	/*
878	 * Every object before this point has size and alignment of 64 bytes.
879	 * Don't add any other type of data before this point.
880	 */
881	char	primary_map;	/* set when primary link map is initialized */
882	char	bucket_init;	/* set when bucket[NBUCKETS] is initialized */
883	char	pad[2];
884	uberflags_t	uberflags;
885	queue_head_t	*queue_head;
886	thr_hash_table_t *thr_hash_table;
887	uint_t		hash_size;	/* # of entries in thr_hash_table[] */
888	uint_t		hash_mask;	/* hash_size - 1 */
889	ulwp_t	*ulwp_one;	/* main thread */
890	ulwp_t	*all_lwps;	/* circular ul_forw/ul_back list of live lwps */
891	ulwp_t	*all_zombies;	/* circular ul_forw/ul_back list of zombies */
892	int	nthreads;	/* total number of live threads/lwps */
893	int	nzombies;	/* total number of zombie threads */
894	int	ndaemons;	/* total number of THR_DAEMON threads/lwps */
895	pid_t	pid;		/* the current process's pid */
896	void	(*sigacthandler)(int, siginfo_t *, void *);
897	ulwp_t	*lwp_stacks;
898	ulwp_t	*lwp_laststack;
899	int	nfreestack;
900	int	thread_stack_cache;
901	ulwp_t	*ulwp_freelist;
902	ulwp_t	*ulwp_lastfree;
903	ulwp_t	*ulwp_replace_free;
904	ulwp_t	*ulwp_replace_last;
905	atfork_t	*atforklist;	/* circular Q for fork handlers */
906	robust_t	**robustlocks;	/* table of registered robust locks */
907	robust_t	*robustlist;	/* list of registered robust locks */
908	char	*progname;	/* the basename of the program, from argv[0] */
909	struct uberdata **tdb_bootstrap;
910	tdb_t	tdb;		/* thread debug interfaces (for libc_db) */
911} uberdata_t;
912
913#define	link_lock	_link_lock.pad_lock
914#define	ld_lock		_ld_lock.pad_lock
915#define	fork_lock	_fork_lock.pad_lock
916#define	atfork_lock	_atfork_lock.pad_lock
917#define	callout_lock	_callout_lock.pad_lock
918#define	tdb_hash_lock	_tdb_hash_lock.pad_lock
919
920#pragma align 64(__uberdata)
921extern	uberdata_t	__uberdata;
922extern	uberdata_t	**__tdb_bootstrap;	/* known to libc_db and mdb */
923extern	int		primary_link_map;
924
925#define	ulwp_mutex(ulwp, udp)	\
926	(&(udp)->thr_hash_table[(ulwp)->ul_ix].hash_lock)
927#define	ulwp_condvar(ulwp, udp)	\
928	(&(udp)->thr_hash_table[(ulwp)->ul_ix].hash_cond)
929
930/*
931 * Grab and release the hash table lock for the specified lwp.
932 */
933#define	ulwp_lock(ulwp, udp)	lmutex_lock(ulwp_mutex(ulwp, udp))
934#define	ulwp_unlock(ulwp, udp)	lmutex_unlock(ulwp_mutex(ulwp, udp))
935
936#ifdef _SYSCALL32	/* needed by libc_db */
937
938typedef struct ulwp32 {
939#if defined(__sparc)
940	uint32_t	ul_dinstr;	/* scratch space for dtrace */
941	uint32_t	ul_padsparc0[15];
942	uint32_t	ul_dsave;	/* dtrace: save %g1, %g0, %sp */
943	uint32_t	ul_drestore;	/* dtrace: restore %g0, %g0, %g0 */
944	uint32_t	ul_dftret;	/* dtrace: return probe fasttrap */
945	uint32_t	ul_dreturn;	/* dtrace: return %o0 */
946#endif
947	caddr32_t	ul_self;	/* pointer to self */
948#if defined(__x86)
949	uint8_t		ul_dinstr[40];	/* scratch space for dtrace */
950#endif
951	caddr32_t	ul_uberdata;	/* uber (super-global) data */
952	tls32_t		ul_tls;		/* dynamic thread-local storage base */
953	caddr32_t	ul_forw;	/* forw, back all_lwps list, */
954	caddr32_t	ul_back;	/* protected by link_lock */
955	caddr32_t	ul_next;	/* list to keep track of stacks */
956	caddr32_t	ul_hash;	/* hash chain linked list */
957	caddr32_t	ul_rval;	/* return value from thr_exit() */
958	caddr32_t	ul_stk;		/* mapping base of the stack */
959	size32_t	ul_mapsiz;	/* mapping size of the stack */
960	size32_t	ul_guardsize;	/* normally _lpagesize */
961	caddr32_t	ul_stktop;	/* broken thr_stksegment() interface */
962	size32_t	ul_stksiz;	/* broken thr_stksegment() interface */
963	stack32_t	ul_ustack;	/* current stack boundaries */
964	int		ul_ix;		/* hash index */
965	lwpid_t		ul_lwpid;	/* thread id, aka the lwp id */
966	pri_t		ul_pri;		/* scheduling priority */
967	pri_t		ul_epri;	/* real-time ceiling priority */
968	char		ul_policy;	/* scheduling policy */
969	char		ul_cid;		/* scheduling class id */
970	union {
971		struct {
972			char	cursig;	/* deferred signal number */
973			char	pleasestop; /* lwp requested to stop itself */
974		} s;
975		short	curplease;	/* for testing both at once */
976	} ul_cp;
977	char		ul_stop;	/* reason for stopping */
978	char		ul_signalled;	/* this lwp was cond_signal()d */
979	char		ul_dead;	/* this lwp has called thr_exit */
980	char		ul_unwind;	/* posix: unwind C++ stack */
981	char		ul_detached;	/* THR_DETACHED at thread_create() */
982					/* or pthread_detach() was called */
983	char		ul_writer;	/* sleeping in rw_wrlock() */
984	char		ul_stopping;	/* set by curthread: stopping self */
985	char		ul_cancel_prologue;	/* for _cancel_prologue() */
986	short		ul_preempt;	/* no_preempt()/preempt() */
987	short		ul_savpreempt;	/* pre-existing preempt value */
988	char		ul_sigsuspend;	/* thread is in sigsuspend/pollsys */
989	char		ul_main;	/* thread is the main thread */
990	char		ul_fork;	/* thread is performing a fork */
991	char		ul_primarymap;	/* primary link-map is initialized */
992	/* per-thread copies of the corresponding global variables */
993	uint8_t		ul_max_spinners;	/* thread_max_spinners */
994	char		ul_door_noreserve;	/* thread_door_noreserve */
995	char		ul_queue_fifo;		/* thread_queue_fifo */
996	char		ul_cond_wait_defer;	/* thread_cond_wait_defer */
997	char		ul_error_detection;	/* thread_error_detection */
998	char		ul_async_safe;		/* thread_async_safe */
999	char		ul_rt;			/* found on an RT queue */
1000	char		ul_rtqueued;		/* was RT when queued */
1001	char		ul_misaligned;		/* thread_locks_misaligned */
1002	char		ul_pad[3];
1003	int		ul_adaptive_spin;	/* thread_adaptive_spin */
1004	int		ul_queue_spin;		/* thread_queue_spin */
1005	int		ul_critical;	/* non-zero == in a critical region */
1006	int		ul_sigdefer;	/* non-zero == defer signals */
1007	int		ul_vfork;	/* thread is the child of vfork() */
1008	int		ul_cancelable;	/* _cancelon()/_canceloff() */
1009	char		ul_cancel_pending;  /* pthread_cancel() was called */
1010	char		ul_cancel_disabled; /* PTHREAD_CANCEL_DISABLE */
1011	char		ul_cancel_async;    /* PTHREAD_CANCEL_ASYNCHRONOUS */
1012	char		ul_save_async;	/* saved copy of ul_cancel_async */
1013	char		ul_mutator;	/* lwp is a mutator (java interface) */
1014	char		ul_created;	/* created suspended */
1015	char		ul_replace;	/* replacement; must be free()d */
1016	uchar_t		ul_nocancel;	/* cancellation can't happen */
1017	int		ul_errno;	/* per-thread errno */
1018	caddr32_t	ul_errnop;	/* pointer to errno or self->ul_errno */
1019	caddr32_t	ul_clnup_hdr;	/* head of cleanup handlers list */
1020	caddr32_t	ul_schedctl_called; /* ul_schedctl is set up */
1021	caddr32_t	ul_schedctl;	/* schedctl data */
1022	int		ul_bindflags;	/* bind_guard() interface to ld.so.1 */
1023	uint_t		ul_libc_locks;	/* count of cancel_safe_mutex_lock()s */
1024	caddr32_t	ul_stsd;	/* slow TLS for keys >= TSD_NFAST */
1025	caddr32_t	ul_ftsd[TSD_NFAST]; /* fast TLS for keys < TSD_NFAST */
1026	td_evbuf32_t	ul_td_evbuf;	/* event buffer */
1027	char		ul_td_events_enable;	/* event mechanism enabled */
1028	char		ul_sync_obj_reg;	/* tdb_sync_obj_register() */
1029	char		ul_qtype;	/* MX or CV */
1030	char		ul_cv_wake;	/* != 0: just wake up, don't requeue */
1031	int		ul_rtld;	/* thread is running inside ld.so.1 */
1032	int		ul_usropts;	/* flags given to thr_create() */
1033	caddr32_t	ul_startpc;	/* start func (thr_create()) */
1034	caddr32_t	ul_startarg;	/* argument for start function */
1035	caddr32_t	ul_wchan;	/* synch object when sleeping */
1036	caddr32_t	ul_link;	/* sleep queue link */
1037	caddr32_t	ul_sleepq;	/* sleep queue thread is waiting on */
1038	caddr32_t	ul_cvmutex;	/* mutex dropped when waiting on a cv */
1039	caddr32_t	ul_mxchain;	/* chain of owned ceiling mutexes */
1040	int		ul_save_state;	/* bind_guard() interface to ld.so.1 */
1041	uint_t		ul_rdlockcnt;	/* # entries in ul_readlock array */
1042				/* 0 means there is but a single entry */
1043	union {				/* single entry or pointer to array */
1044		readlock32_t	single;
1045		caddr32_t	array;
1046	} ul_readlock;
1047	uint_t		ul_heldlockcnt;	/* # entries in ul_heldlocks array */
1048				/* 0 means there is but a single entry */
1049	union {				/* single entry or pointer to array */
1050		caddr32_t	single;
1051		caddr32_t	array;
1052	} ul_heldlocks;
1053	/* PROBE_SUPPORT begin */
1054	caddr32_t	ul_tpdp;
1055	/* PROBE_SUPPORT end */
1056	caddr32_t	ul_siglink;	/* pointer to previous context */
1057	uint_t		ul_spin_lock_spin;	/* spin lock statistics */
1058	uint_t		ul_spin_lock_spin2;
1059	uint_t		ul_spin_lock_sleep;
1060	uint_t		ul_spin_lock_wakeup;
1061	queue_root32_t	ul_queue_root;	/* root of a sleep queue */
1062	id_t		ul_rtclassid;	/* real-time class id */
1063	uint_t		ul_pilocks;	/* count of PI locks held */
1064		/* the following members *must* be last in the structure */
1065		/* they are discarded when ulwp is replaced on thr_exit() */
1066	sigset_t	ul_sigmask;	/* thread's current signal mask */
1067	sigset_t	ul_tmpmask;	/* signal mask for sigsuspend/pollsys */
1068	siginfo32_t	ul_siginfo;	/* deferred siginfo */
1069	mutex_t		ul_spinlock;	/* used when suspending/continuing */
1070	fpuenv32_t	ul_fpuenv;	/* floating point state */
1071	caddr32_t	ul_sp;		/* stack pointer when blocked */
1072#if defined(sparc)
1073	caddr32_t	ul_unwind_ret;	/* used only by _ex_clnup_handler() */
1074#endif
1075} ulwp32_t;
1076
1077#define	REPLACEMENT_SIZE32	((size_t)&((ulwp32_t *)NULL)->ul_sigmask)
1078
1079typedef struct uberdata32 {
1080	pad_lock_t	_link_lock;
1081	pad_lock_t	_ld_lock;
1082	pad_lock_t	_fork_lock;
1083	pad_lock_t	_atfork_lock;
1084	pad32_lock_t	_callout_lock;
1085	pad32_lock_t	_tdb_hash_lock;
1086	tdb_sync_stats_t tdb_hash_lock_stats;
1087	siguaction32_t	siguaction[NSIG];
1088	bucket32_t	bucket[NBUCKETS];
1089	atexit_root32_t	atexit_root;
1090	tsd_metadata32_t tsd_metadata;
1091	tls_metadata32_t tls_metadata;
1092	char		primary_map;
1093	char		bucket_init;
1094	char		pad[2];
1095	uberflags_t	uberflags;
1096	caddr32_t	queue_head;
1097	caddr32_t	thr_hash_table;
1098	uint_t		hash_size;
1099	uint_t		hash_mask;
1100	caddr32_t	ulwp_one;
1101	caddr32_t	all_lwps;
1102	caddr32_t	all_zombies;
1103	int		nthreads;
1104	int		nzombies;
1105	int		ndaemons;
1106	int		pid;
1107	caddr32_t	sigacthandler;
1108	caddr32_t	lwp_stacks;
1109	caddr32_t	lwp_laststack;
1110	int		nfreestack;
1111	int		thread_stack_cache;
1112	caddr32_t	ulwp_freelist;
1113	caddr32_t	ulwp_lastfree;
1114	caddr32_t	ulwp_replace_free;
1115	caddr32_t	ulwp_replace_last;
1116	caddr32_t	atforklist;
1117	caddr32_t	robustlocks;
1118	caddr32_t	robustlist;
1119	caddr32_t	tdb_bootstrap;
1120	tdb32_t		tdb;
1121} uberdata32_t;
1122
1123#endif	/* _SYSCALL32 */
1124
1125/* ul_stop values */
1126#define	TSTP_REGULAR	0x01	/* Stopped by thr_suspend() */
1127#define	TSTP_MUTATOR	0x08	/* stopped by thr_suspend_*mutator*() */
1128#define	TSTP_FORK	0x20	/* stopped by suspend_fork() */
1129
1130/*
1131 * Implementation-specific attribute types for pthread_mutexattr_init() etc.
1132 */
1133
1134typedef	struct	_cvattr {
1135	int	pshared;
1136	clockid_t clockid;
1137} cvattr_t;
1138
1139typedef	struct	_mattr {
1140	int	pshared;
1141	int	protocol;
1142	int	prioceiling;
1143	int	type;
1144	int	robustness;
1145} mattr_t;
1146
1147typedef	struct	_thrattr {
1148	size_t	stksize;
1149	void	*stkaddr;
1150	int	detachstate;
1151	int	daemonstate;
1152	int	scope;
1153	int	prio;
1154	int	policy;
1155	int	inherit;
1156	size_t	guardsize;
1157} thrattr_t;
1158
1159typedef	struct	_rwlattr {
1160	int	pshared;
1161} rwlattr_t;
1162
1163/* _curthread() is inline for speed */
1164extern	ulwp_t		*_curthread(void);
1165#define	curthread	(_curthread())
1166
1167/* this version (also inline) can be tested for NULL */
1168extern	ulwp_t		*__curthread(void);
1169
1170/* get the current stack pointer (also inline) */
1171extern	greg_t		stkptr(void);
1172
1173/*
1174 * Suppress __attribute__((...)) if we are not compiling with gcc
1175 */
1176#if !defined(__GNUC__)
1177#define	__attribute__(string)
1178#endif
1179
1180/* Fetch the dispatch (kernel) priority of a thread */
1181#define	real_priority(ulwp)	\
1182	((ulwp)->ul_schedctl? (ulwp)->ul_schedctl->sc_priority : 0)
1183
1184/*
1185 * Implementation functions.  Not visible outside of the library itself.
1186 */
1187extern	int	__nanosleep(const timespec_t *, timespec_t *);
1188extern	void	getgregs(ulwp_t *, gregset_t);
1189extern	void	setgregs(ulwp_t *, gregset_t);
1190extern	void	thr_panic(const char *);
1191#pragma rarely_called(thr_panic)
1192extern	ulwp_t	*find_lwp(thread_t);
1193extern	void	finish_init(void);
1194extern	void	update_sched(ulwp_t *);
1195extern	void	queue_alloc(void);
1196extern	void	tsd_exit(void);
1197extern	void	tsd_free(ulwp_t *);
1198extern	void	tls_setup(void);
1199extern	void	tls_exit(void);
1200extern	void	tls_free(ulwp_t *);
1201extern	void	rwl_free(ulwp_t *);
1202extern	void	heldlock_exit(void);
1203extern	void	heldlock_free(ulwp_t *);
1204extern	void	sigacthandler(int, siginfo_t *, void *);
1205extern	void	signal_init(void);
1206extern	int	sigequalset(const sigset_t *, const sigset_t *);
1207extern	void	mutex_setup(void);
1208extern	void	take_deferred_signal(int);
1209extern	void	*setup_top_frame(void *, size_t, ulwp_t *);
1210extern	int	setup_context(ucontext_t *, void *(*func)(ulwp_t *),
1211			ulwp_t *ulwp, caddr_t stk, size_t stksize);
1212extern	volatile sc_shared_t *setup_schedctl(void);
1213extern	void	*lmalloc(size_t);
1214extern	void	lfree(void *, size_t);
1215extern	void	*libc_malloc(size_t);
1216extern	void	*libc_realloc(void *, size_t);
1217extern	void	libc_free(void *);
1218extern	char	*libc_strdup(const char *);
1219extern	void	ultos(uint64_t, int, char *);
1220extern	void	lock_error(const mutex_t *, const char *, void *, const char *);
1221extern	void	rwlock_error(const rwlock_t *, const char *, const char *);
1222extern	void	thread_error(const char *);
1223extern	void	grab_assert_lock(void);
1224extern	void	dump_queue_statistics(void);
1225extern	void	collect_queue_statistics(void);
1226extern	void	record_spin_locks(ulwp_t *);
1227extern	void	remember_lock(mutex_t *);
1228extern	void	forget_lock(mutex_t *);
1229extern	void	register_lock(mutex_t *);
1230extern	void	unregister_locks(void);
1231#if defined(__sparc)
1232extern	void	_flush_windows(void);
1233#else
1234#define	_flush_windows()
1235#endif
1236extern	void	set_curthread(void *);
1237
1238/*
1239 * Utility function used when waking up many threads (more than MAXLWPS)
1240 * all at once.  See mutex_wakeup_all(), cond_broadcast(), and rw_unlock().
1241 */
1242#define	MAXLWPS	128	/* max remembered lwpids before overflow */
1243#define	NEWLWPS	2048	/* max remembered lwpids at first overflow */
1244extern	lwpid_t	*alloc_lwpids(lwpid_t *, int *, int *);
1245
1246/* enter a critical section */
1247#define	enter_critical(self)	(self->ul_critical++)
1248
1249/* exit a critical section, take deferred actions if necessary */
1250extern	void	do_exit_critical(void);
1251#define	exit_critical(self)					\
1252	(void) (self->ul_critical--,				\
1253	    ((self->ul_curplease && self->ul_critical == 0)?	\
1254	    (do_exit_critical(), 0) : 0))
1255
1256/*
1257 * Like enter_critical()/exit_critical() but just for deferring signals.
1258 * Unlike enter_critical()/exit_critical(), ul_sigdefer may be set while
1259 * calling application functions like constructors and destructors.
1260 * Care must be taken if the application function attempts to set
1261 * the signal mask while a deferred signal is present; the setting
1262 * of the signal mask must also be deferred.
1263 */
1264#define	sigoff(self)	(self->ul_sigdefer++)
1265#define	sigon(self)						\
1266	(void) ((--self->ul_sigdefer == 0 &&			\
1267	    self->ul_curplease && self->ul_critical == 0)?	\
1268	    (do_exit_critical(), 0) : 0)
1269
1270/* these are exported functions */
1271extern	void	_sigoff(void);
1272extern	void	_sigon(void);
1273
1274#define	sigorset(s1, s2)				\
1275	(((s1)->__sigbits[0] |= (s2)->__sigbits[0]),	\
1276	((s1)->__sigbits[1] |= (s2)->__sigbits[1]),	\
1277	((s1)->__sigbits[2] |= (s2)->__sigbits[2]),	\
1278	((s1)->__sigbits[3] |= (s2)->__sigbits[3]))
1279
1280#define	sigandset(s1, s2)				\
1281	(((s1)->__sigbits[0] &= (s2)->__sigbits[0]),	\
1282	((s1)->__sigbits[1] &= (s2)->__sigbits[1]),	\
1283	((s1)->__sigbits[2] &= (s2)->__sigbits[2]),	\
1284	((s1)->__sigbits[3] &= (s2)->__sigbits[3]))
1285
1286#define	sigdiffset(s1, s2)				\
1287	(((s1)->__sigbits[0] &= ~(s2)->__sigbits[0]),	\
1288	((s1)->__sigbits[1] &= ~(s2)->__sigbits[1]),	\
1289	((s1)->__sigbits[2] &= ~(s2)->__sigbits[2]),	\
1290	((s1)->__sigbits[3] &= ~(s2)->__sigbits[3]))
1291
1292#define	delete_reserved_signals(s)			\
1293	(((s)->__sigbits[0] &= MASKSET0),		\
1294	((s)->__sigbits[1] &= (MASKSET1 & ~SIGMASK(SIGCANCEL))),\
1295	((s)->__sigbits[2] &= MASKSET2),		\
1296	((s)->__sigbits[3] &= MASKSET3))
1297
1298extern	void	block_all_signals(ulwp_t *self);
1299
1300/*
1301 * When restoring the signal mask after having previously called
1302 * block_all_signals(), if we have a deferred signal present then
1303 * do nothing other than ASSERT() that we are in a critical region.
1304 * The signal mask will be set when we emerge from the critical region
1305 * and call take_deferred_signal().  There is no race condition here
1306 * because the kernel currently has all signals blocked for this thread.
1307 */
1308#define	restore_signals(self)						\
1309	((void) ((self)->ul_cursig?					\
1310	(ASSERT((self)->ul_critical + (self)->ul_sigdefer != 0), 0) :	\
1311	__lwp_sigmask(SIG_SETMASK, &(self)->ul_sigmask)))
1312
1313extern	void	set_cancel_pending_flag(ulwp_t *, int);
1314extern	void	set_cancel_eintr_flag(ulwp_t *);
1315extern	void	set_parking_flag(ulwp_t *, int);
1316extern	int	cancel_active(void);
1317
1318extern	void	*_thrp_setup(ulwp_t *);
1319extern	void	_fpinherit(ulwp_t *);
1320extern	void	_lwp_start(void);
1321extern	void	_lwp_terminate(void);
1322extern	void	lmutex_lock(mutex_t *);
1323extern	void	lmutex_unlock(mutex_t *);
1324extern	void	lrw_rdlock(rwlock_t *);
1325extern	void	lrw_wrlock(rwlock_t *);
1326extern	void	lrw_unlock(rwlock_t *);
1327extern	void	sig_mutex_lock(mutex_t *);
1328extern	void	sig_mutex_unlock(mutex_t *);
1329extern	int	sig_mutex_trylock(mutex_t *);
1330extern	int	sig_cond_wait(cond_t *, mutex_t *);
1331extern	int	sig_cond_reltimedwait(cond_t *, mutex_t *, const timespec_t *);
1332extern	void	cancel_safe_mutex_lock(mutex_t *);
1333extern	void	cancel_safe_mutex_unlock(mutex_t *);
1334extern	int	cancel_safe_mutex_trylock(mutex_t *);
1335extern	void	_prefork_handler(void);
1336extern	void	_postfork_parent_handler(void);
1337extern	void	_postfork_child_handler(void);
1338extern	void	postfork1_child(void);
1339extern	void	postfork1_child_aio(void);
1340extern	void	postfork1_child_sigev_aio(void);
1341extern	void	postfork1_child_sigev_mq(void);
1342extern	void	postfork1_child_sigev_timer(void);
1343extern	void	postfork1_child_tpool(void);
1344extern	void	fork_lock_enter(void);
1345extern	void	fork_lock_exit(void);
1346extern	void	suspend_fork(void);
1347extern	void	continue_fork(int);
1348extern	void	do_sigcancel(void);
1349extern	void	setup_cancelsig(int);
1350extern	void	init_sigev_thread(void);
1351extern	void	init_aio(void);
1352extern	void	init_progname(void);
1353extern	void	_cancelon(void);
1354extern	void	_canceloff(void);
1355extern	void	_canceloff_nocancel(void);
1356extern	void	_cancel_prologue(void);
1357extern	void	_cancel_epilogue(void);
1358extern	void	no_preempt(ulwp_t *);
1359extern	void	preempt(ulwp_t *);
1360extern	void	_thrp_unwind(void *);
1361
1362extern	pid_t	__forkx(int);
1363extern	pid_t	__forkallx(int);
1364extern	int	__open(const char *, int, mode_t);
1365extern	int	__open64(const char *, int, mode_t);
1366extern	int	__openat(int, const char *, int, mode_t);
1367extern	int	__openat64(int, const char *, int, mode_t);
1368extern	int	__close(int);
1369extern	ssize_t	__read(int, void *, size_t);
1370extern	ssize_t	__write(int, const void *, size_t);
1371extern	int	__fcntl(int, int, ...);
1372extern	int	__lwp_continue(lwpid_t);
1373extern	int	__lwp_create(ucontext_t *, uint_t, lwpid_t *);
1374extern	int	___lwp_suspend(lwpid_t);
1375extern	int	lwp_wait(lwpid_t, lwpid_t *);
1376extern	int	__lwp_wait(lwpid_t, lwpid_t *);
1377extern	int	__lwp_detach(lwpid_t);
1378extern	sc_shared_t *__schedctl(void);
1379
1380/* actual system call traps */
1381extern	int	__setcontext(const ucontext_t *);
1382extern	int	__getcontext(ucontext_t *);
1383extern	int	__clock_gettime(clockid_t, timespec_t *);
1384extern	void	abstime_to_reltime(clockid_t, const timespec_t *, timespec_t *);
1385extern	void	hrt2ts(hrtime_t, timespec_t *);
1386
1387extern	int	__sigaction(int, const struct sigaction *, struct sigaction *);
1388extern	int	__sigprocmask(int, const sigset_t *, sigset_t *);
1389extern	int	__lwp_sigmask(int, const sigset_t *);
1390extern	void	__sighndlr(int, siginfo_t *, ucontext_t *, void (*)());
1391extern	caddr_t	__sighndlrend;
1392#pragma unknown_control_flow(__sighndlr)
1393
1394/* belongs in <pthread.h> */
1395#define	PTHREAD_CREATE_DAEMON_NP	0x100	/* = THR_DAEMON */
1396#define	PTHREAD_CREATE_NONDAEMON_NP	0
1397extern	int	pthread_attr_setdaemonstate_np(pthread_attr_t *, int);
1398extern	int	pthread_attr_getdaemonstate_np(const pthread_attr_t *, int *);
1399
1400extern	int	mutex_held(mutex_t *);
1401extern	int	mutex_lock_internal(mutex_t *, timespec_t *, int);
1402extern	int	mutex_unlock_internal(mutex_t *, int);
1403
1404/* not cancellation points: */
1405extern	int	__cond_wait(cond_t *, mutex_t *);
1406extern	int	__cond_timedwait(cond_t *, mutex_t *, const timespec_t *);
1407extern	int	__cond_reltimedwait(cond_t *, mutex_t *, const timespec_t *);
1408
1409extern	int	rw_read_held(rwlock_t *);
1410extern	int	rw_write_held(rwlock_t *);
1411
1412extern	int	_thrp_create(void *, size_t, void *(*)(void *), void *, long,
1413			thread_t *, size_t);
1414extern	int	_thrp_suspend(thread_t, uchar_t);
1415extern	int	_thrp_continue(thread_t, uchar_t);
1416
1417extern	void	_thrp_terminate(void *);
1418extern	void	_thrp_exit(void);
1419
1420extern	const pcclass_t *get_info_by_class(id_t);
1421extern	const pcclass_t *get_info_by_policy(int);
1422extern	const thrattr_t *def_thrattr(void);
1423extern	id_t	setparam(idtype_t, id_t, int, int);
1424extern	id_t	setprio(idtype_t, id_t, int, int *);
1425extern	id_t	getparam(idtype_t, id_t, int *, struct sched_param *);
1426
1427/*
1428 * System call wrappers (direct interfaces to the kernel)
1429 */
1430extern	int	___lwp_mutex_register(mutex_t *, mutex_t **);
1431extern	int	___lwp_mutex_trylock(mutex_t *, ulwp_t *);
1432extern	int	___lwp_mutex_timedlock(mutex_t *, timespec_t *, ulwp_t *);
1433extern	int	___lwp_mutex_unlock(mutex_t *);
1434extern	int	___lwp_mutex_wakeup(mutex_t *, int);
1435extern	int	___lwp_cond_wait(cond_t *, mutex_t *, timespec_t *, int);
1436extern	int	___lwp_sema_timedwait(lwp_sema_t *, timespec_t *, int);
1437extern	int	__lwp_rwlock_rdlock(rwlock_t *, timespec_t *);
1438extern	int	__lwp_rwlock_wrlock(rwlock_t *, timespec_t *);
1439extern	int	__lwp_rwlock_tryrdlock(rwlock_t *);
1440extern	int	__lwp_rwlock_trywrlock(rwlock_t *);
1441extern	int	__lwp_rwlock_unlock(rwlock_t *);
1442extern	int	__lwp_park(timespec_t *, lwpid_t);
1443extern	int	__lwp_unpark(lwpid_t);
1444extern	int	__lwp_unpark_all(lwpid_t *, int);
1445#if defined(__x86)
1446extern	int	___lwp_private(int, int, void *);
1447#endif	/* __x86 */
1448
1449/*
1450 * inlines
1451 */
1452extern	int		set_lock_byte(volatile uint8_t *);
1453extern	uint32_t	atomic_swap_32(volatile uint32_t *, uint32_t);
1454extern	uint32_t	atomic_cas_32(volatile uint32_t *, uint32_t, uint32_t);
1455extern	void		atomic_inc_32(volatile uint32_t *);
1456extern	void		atomic_dec_32(volatile uint32_t *);
1457extern	void		atomic_and_32(volatile uint32_t *, uint32_t);
1458extern	void		atomic_or_32(volatile uint32_t *, uint32_t);
1459#if defined(__sparc)
1460extern	ulong_t		caller(void);
1461extern	ulong_t		getfp(void);
1462#endif	/* __sparc */
1463
1464#include "thr_inlines.h"
1465
1466#endif	/* _THR_UBERDATA_H */
1467