Home | History | Annotate | Download | only in os
      1      0     stevel /*
      2      0     stevel  * CDDL HEADER START
      3      0     stevel  *
      4      0     stevel  * The contents of this file are subject to the terms of the
      5   2712    nn35248  * Common Development and Distribution License (the "License").
      6   2712    nn35248  * You may not use this file except in compliance with the License.
      7      0     stevel  *
      8      0     stevel  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
      9      0     stevel  * or http://www.opensolaris.org/os/licensing.
     10      0     stevel  * See the License for the specific language governing permissions
     11      0     stevel  * and limitations under the License.
     12      0     stevel  *
     13      0     stevel  * When distributing Covered Code, include this CDDL HEADER in each
     14      0     stevel  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
     15      0     stevel  * If applicable, add the following below this CDDL HEADER, with the
     16      0     stevel  * fields enclosed by brackets "[]" replaced with your own identifying
     17      0     stevel  * information: Portions Copyright [yyyy] [name of copyright owner]
     18      0     stevel  *
     19      0     stevel  * CDDL HEADER END
     20      0     stevel  */
     21    390        raf 
     22      0     stevel /*
     23   9393      Roger  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
     24      0     stevel  * Use is subject to license terms.
     25      0     stevel  */
     26      0     stevel 
     27      0     stevel #include <sys/param.h>
     28      0     stevel #include <sys/types.h>
     29      0     stevel #include <sys/sysmacros.h>
     30      0     stevel #include <sys/systm.h>
     31      0     stevel #include <sys/thread.h>
     32      0     stevel #include <sys/proc.h>
     33      0     stevel #include <sys/task.h>
     34      0     stevel #include <sys/project.h>
     35      0     stevel #include <sys/signal.h>
     36      0     stevel #include <sys/errno.h>
     37      0     stevel #include <sys/vmparam.h>
     38      0     stevel #include <sys/stack.h>
     39      0     stevel #include <sys/procfs.h>
     40      0     stevel #include <sys/prsystm.h>
     41      0     stevel #include <sys/cpuvar.h>
     42      0     stevel #include <sys/kmem.h>
     43      0     stevel #include <sys/vtrace.h>
     44      0     stevel #include <sys/door.h>
     45      0     stevel #include <vm/seg_kp.h>
     46      0     stevel #include <sys/debug.h>
     47      0     stevel #include <sys/tnf.h>
     48      0     stevel #include <sys/schedctl.h>
     49      0     stevel #include <sys/poll.h>
     50      0     stevel #include <sys/copyops.h>
     51      0     stevel #include <sys/lwp_upimutex_impl.h>
     52      0     stevel #include <sys/cpupart.h>
     53      0     stevel #include <sys/lgrp.h>
     54      0     stevel #include <sys/rctl.h>
     55      0     stevel #include <sys/contract_impl.h>
     56      0     stevel #include <sys/cpc_impl.h>
     57      0     stevel #include <sys/sdt.h>
     58      0     stevel #include <sys/cmn_err.h>
     59   2712    nn35248 #include <sys/brand.h>
     60   9870      Roger #include <sys/cyclic.h>
     61  10541  Gangadhar #include <sys/pool.h>
     62      0     stevel 
     63   9393      Roger /* hash function for the lwpid hash table, p->p_tidhash[] */
     64   9393      Roger #define	TIDHASH(tid, hash_sz)	((tid) & ((hash_sz) - 1))
     65   9393      Roger 
     66      0     stevel void *segkp_lwp;		/* cookie for pool of segkp resources */
     67   5834   pt157919 extern void reapq_move_lq_to_tq(kthread_t *);
     68   5834   pt157919 extern void freectx_ctx(struct ctxop *);
     69      0     stevel 
     70      0     stevel /*
     71  11173   Jonathan  * Create a kernel thread associated with a particular system process.  Give
     72  11173   Jonathan  * it an LWP so that microstate accounting will be available for it.
     73  11173   Jonathan  */
     74  11173   Jonathan kthread_t *
     75  11173   Jonathan lwp_kernel_create(proc_t *p, void (*proc)(), void *arg, int state, pri_t pri)
     76  11173   Jonathan {
     77  11173   Jonathan 	klwp_t *lwp;
     78  11173   Jonathan 
     79  11173   Jonathan 	VERIFY((p->p_flag & SSYS) != 0);
     80  11173   Jonathan 
     81  11173   Jonathan 	lwp = lwp_create(proc, arg, 0, p, state, pri, &t0.t_hold, syscid, 0);
     82  11173   Jonathan 
     83  11173   Jonathan 	VERIFY(lwp != NULL);
     84  11173   Jonathan 
     85  11173   Jonathan 	return (lwptot(lwp));
     86  11173   Jonathan }
     87  11173   Jonathan 
     88  11173   Jonathan /*
     89      0     stevel  * Create a thread that appears to be stopped at sys_rtt.
     90      0     stevel  */
     91      0     stevel klwp_t *
     92      0     stevel lwp_create(void (*proc)(), caddr_t arg, size_t len, proc_t *p,
     93      0     stevel     int state, int pri, const k_sigset_t *smask, int cid, id_t lwpid)
     94      0     stevel {
     95      0     stevel 	klwp_t *lwp = NULL;
     96      0     stevel 	kthread_t *t;
     97      0     stevel 	kthread_t *tx;
     98      0     stevel 	cpupart_t *oldpart = NULL;
     99      0     stevel 	size_t	stksize;
    100      0     stevel 	caddr_t lwpdata = NULL;
    101      0     stevel 	processorid_t	binding;
    102      0     stevel 	int err = 0;
    103      0     stevel 	kproject_t *oldkpj, *newkpj;
    104      0     stevel 	void *bufp = NULL;
    105  11173   Jonathan 	klwp_t *curlwp;
    106      0     stevel 	lwpent_t *lep;
    107      0     stevel 	lwpdir_t *old_dir = NULL;
    108      0     stevel 	uint_t old_dirsz = 0;
    109   9393      Roger 	tidhash_t *old_hash = NULL;
    110      0     stevel 	uint_t old_hashsz = 0;
    111   9393      Roger 	ret_tidhash_t *ret_tidhash = NULL;
    112      0     stevel 	int i;
    113      0     stevel 	int rctlfail = 0;
    114   2712    nn35248 	boolean_t branded = 0;
    115   5834   pt157919 	struct ctxop *ctx = NULL;
    116      0     stevel 
    117  11173   Jonathan 	ASSERT(cid != sysdccid);	/* system threads must start in SYS */
    118  11173   Jonathan 
    119  11173   Jonathan 	ASSERT(p != &p0);		/* No new LWPs in p0. */
    120  11173   Jonathan 
    121      0     stevel 	mutex_enter(&p->p_lock);
    122      0     stevel 	mutex_enter(&p->p_zone->zone_nlwps_lock);
    123      0     stevel 	/*
    124      0     stevel 	 * don't enforce rctl limits on system processes
    125      0     stevel 	 */
    126  11173   Jonathan 	if (!CLASS_KERNEL(cid)) {
    127      0     stevel 		if (p->p_task->tk_nlwps >= p->p_task->tk_nlwps_ctl)
    128      0     stevel 			if (rctl_test(rc_task_lwps, p->p_task->tk_rctls, p,
    129      0     stevel 			    1, 0) & RCT_DENY)
    130      0     stevel 				rctlfail = 1;
    131      0     stevel 		if (p->p_task->tk_proj->kpj_nlwps >=
    132      0     stevel 		    p->p_task->tk_proj->kpj_nlwps_ctl)
    133      0     stevel 			if (rctl_test(rc_project_nlwps,
    134      0     stevel 			    p->p_task->tk_proj->kpj_rctls, p, 1, 0)
    135      0     stevel 			    & RCT_DENY)
    136      0     stevel 				rctlfail = 1;
    137      0     stevel 		if (p->p_zone->zone_nlwps >= p->p_zone->zone_nlwps_ctl)
    138      0     stevel 			if (rctl_test(rc_zone_nlwps, p->p_zone->zone_rctls, p,
    139      0     stevel 			    1, 0) & RCT_DENY)
    140      0     stevel 				rctlfail = 1;
    141      0     stevel 	}
    142      0     stevel 	if (rctlfail) {
    143      0     stevel 		mutex_exit(&p->p_zone->zone_nlwps_lock);
    144      0     stevel 		mutex_exit(&p->p_lock);
    145      0     stevel 		return (NULL);
    146      0     stevel 	}
    147      0     stevel 	p->p_task->tk_nlwps++;
    148      0     stevel 	p->p_task->tk_proj->kpj_nlwps++;
    149      0     stevel 	p->p_zone->zone_nlwps++;
    150      0     stevel 	mutex_exit(&p->p_zone->zone_nlwps_lock);
    151      0     stevel 	mutex_exit(&p->p_lock);
    152      0     stevel 
    153  11173   Jonathan 	if (CLASS_KERNEL(cid)) {
    154  11173   Jonathan 		curlwp = NULL;		/* don't inherit from curlwp */
    155      0     stevel 		stksize = lwp_default_stksize;
    156  11173   Jonathan 	} else {
    157  11173   Jonathan 		curlwp = ttolwp(curthread);
    158  11173   Jonathan 		if (curlwp == NULL || (stksize = curlwp->lwp_childstksz) == 0)
    159  11173   Jonathan 			stksize = lwp_default_stksize;
    160  11173   Jonathan 	}
    161      0     stevel 
    162      0     stevel 	/*
    163  11173   Jonathan 	 * For system threads, we sleep for our swap reservation, and the
    164  11173   Jonathan 	 * thread stack can't be swapped.
    165  11173   Jonathan 	 *
    166  11173   Jonathan 	 * Otherwise, try to reclaim a <lwp,stack> from 'deathrow'
    167      0     stevel 	 */
    168  11173   Jonathan 	if (CLASS_KERNEL(cid)) {
    169  11173   Jonathan 		lwpdata = (caddr_t)segkp_get(segkp, stksize,
    170  11173   Jonathan 		    (KPD_NO_ANON | KPD_HASREDZONE | KPD_LOCKED));
    171  11173   Jonathan 
    172  11173   Jonathan 	} else if (stksize == lwp_default_stksize) {
    173      0     stevel 		if (lwp_reapcnt > 0) {
    174      0     stevel 			mutex_enter(&reaplock);
    175      0     stevel 			if ((t = lwp_deathrow) != NULL) {
    176      0     stevel 				ASSERT(t->t_swap);
    177      0     stevel 				lwp_deathrow = t->t_forw;
    178      0     stevel 				lwp_reapcnt--;
    179      0     stevel 				lwpdata = t->t_swap;
    180      0     stevel 				lwp = t->t_lwp;
    181   5834   pt157919 				ctx = t->t_ctx;
    182   5834   pt157919 				t->t_swap = NULL;
    183   5834   pt157919 				t->t_lwp = NULL;
    184   5834   pt157919 				t->t_ctx = NULL;
    185   5834   pt157919 				reapq_move_lq_to_tq(t);
    186      0     stevel 			}
    187      0     stevel 			mutex_exit(&reaplock);
    188   5834   pt157919 			if (lwp != NULL) {
    189   5834   pt157919 				lwp_stk_fini(lwp);
    190   5834   pt157919 			}
    191   5834   pt157919 			if (ctx != NULL) {
    192   5834   pt157919 				freectx_ctx(ctx);
    193      0     stevel 			}
    194      0     stevel 		}
    195      0     stevel 		if (lwpdata == NULL &&
    196      0     stevel 		    (lwpdata = (caddr_t)segkp_cache_get(segkp_lwp)) == NULL) {
    197      0     stevel 			mutex_enter(&p->p_lock);
    198      0     stevel 			mutex_enter(&p->p_zone->zone_nlwps_lock);
    199      0     stevel 			p->p_task->tk_nlwps--;
    200      0     stevel 			p->p_task->tk_proj->kpj_nlwps--;
    201      0     stevel 			p->p_zone->zone_nlwps--;
    202      0     stevel 			mutex_exit(&p->p_zone->zone_nlwps_lock);
    203      0     stevel 			mutex_exit(&p->p_lock);
    204      0     stevel 			return (NULL);
    205      0     stevel 		}
    206      0     stevel 	} else {
    207      0     stevel 		stksize = roundup(stksize, PAGESIZE);
    208      0     stevel 		if ((lwpdata = (caddr_t)segkp_get(segkp, stksize,
    209      0     stevel 		    (KPD_NOWAIT | KPD_HASREDZONE | KPD_LOCKED))) == NULL) {
    210      0     stevel 			mutex_enter(&p->p_lock);
    211      0     stevel 			mutex_enter(&p->p_zone->zone_nlwps_lock);
    212      0     stevel 			p->p_task->tk_nlwps--;
    213      0     stevel 			p->p_task->tk_proj->kpj_nlwps--;
    214      0     stevel 			p->p_zone->zone_nlwps--;
    215      0     stevel 			mutex_exit(&p->p_zone->zone_nlwps_lock);
    216      0     stevel 			mutex_exit(&p->p_lock);
    217      0     stevel 			return (NULL);
    218      0     stevel 		}
    219      0     stevel 	}
    220      0     stevel 
    221      0     stevel 	/*
    222      0     stevel 	 * Create a thread, initializing the stack pointer
    223      0     stevel 	 */
    224      0     stevel 	t = thread_create(lwpdata, stksize, NULL, NULL, 0, p, TS_STOPPED, pri);
    225      0     stevel 
    226      0     stevel 	t->t_swap = lwpdata;	/* Start of page-able data */
    227      0     stevel 	if (lwp == NULL)
    228      0     stevel 		lwp = kmem_cache_alloc(lwp_cache, KM_SLEEP);
    229      0     stevel 	bzero(lwp, sizeof (*lwp));
    230      0     stevel 	t->t_lwp = lwp;
    231      0     stevel 
    232      0     stevel 	t->t_hold = *smask;
    233      0     stevel 	lwp->lwp_thread = t;
    234      0     stevel 	lwp->lwp_procp = p;
    235      0     stevel 	lwp->lwp_sigaltstack.ss_flags = SS_DISABLE;
    236      0     stevel 	if (curlwp != NULL && curlwp->lwp_childstksz != 0)
    237      0     stevel 		lwp->lwp_childstksz = curlwp->lwp_childstksz;
    238      0     stevel 
    239      0     stevel 	t->t_stk = lwp_stk_init(lwp, t->t_stk);
    240      0     stevel 	thread_load(t, proc, arg, len);
    241      0     stevel 
    242      0     stevel 	/*
    243      0     stevel 	 * Allocate the SIGPROF buffer if ITIMER_REALPROF is in effect.
    244      0     stevel 	 */
    245   9870      Roger 	if (p->p_rprof_cyclic != CYCLIC_NONE)
    246      0     stevel 		t->t_rprof = kmem_zalloc(sizeof (struct rprof), KM_SLEEP);
    247      0     stevel 
    248      0     stevel 	if (cid != NOCLASS)
    249      0     stevel 		(void) CL_ALLOC(&bufp, cid, KM_SLEEP);
    250      0     stevel 
    251      0     stevel 	/*
    252      0     stevel 	 * Allocate an lwp directory entry for the new lwp.
    253      0     stevel 	 */
    254      0     stevel 	lep = kmem_zalloc(sizeof (*lep), KM_SLEEP);
    255      0     stevel 
    256      0     stevel 	mutex_enter(&p->p_lock);
    257      0     stevel grow:
    258      0     stevel 	/*
    259      0     stevel 	 * Grow the lwp (thread) directory and lwpid hash table if necessary.
    260      0     stevel 	 * A note on the growth algorithm:
    261      0     stevel 	 *	The new lwp directory size is computed as:
    262      0     stevel 	 *		new = 2 * old + 2
    263      0     stevel 	 *	Starting with an initial size of 2 (see exec_common()),
    264      0     stevel 	 *	this yields numbers that are a power of two minus 2:
    265      0     stevel 	 *		2, 6, 14, 30, 62, 126, 254, 510, 1022, ...
    266      0     stevel 	 *	The size of the lwpid hash table must be a power of two
    267      0     stevel 	 *	and must be commensurate in size with the lwp directory
    268      0     stevel 	 *	so that hash bucket chains remain short.  Therefore,
    269      0     stevel 	 *	the lwpid hash table size is computed as:
    270      0     stevel 	 *		hashsz = (dirsz + 2) / 2
    271      0     stevel 	 *	which leads to these hash table sizes corresponding to
    272      0     stevel 	 *	the above directory sizes:
    273      0     stevel 	 *		2, 4, 8, 16, 32, 64, 128, 256, 512, ...
    274   9393      Roger 	 * A note on growing the hash table:
    275   9393      Roger 	 *	For performance reasons, code in lwp_unpark() does not
    276   9393      Roger 	 *	acquire curproc->p_lock when searching the hash table.
    277   9393      Roger 	 *	Rather, it calls lwp_hash_lookup_and_lock() which
    278   9393      Roger 	 *	acquires only the individual hash bucket lock, taking
    279   9393      Roger 	 *	care to deal with reallocation of the hash table
    280   9393      Roger 	 *	during the time it takes to acquire the lock.
    281   9393      Roger 	 *
    282   9393      Roger 	 *	This is sufficient to protect the integrity of the
    283   9393      Roger 	 *	hash table, but it requires us to acquire all of the
    284   9393      Roger 	 *	old hash bucket locks before growing the hash table
    285   9393      Roger 	 *	and to release them afterwards.  It also requires us
    286   9393      Roger 	 *	not to free the old hash table because some thread
    287   9393      Roger 	 *	in lwp_hash_lookup_and_lock() might still be trying
    288   9393      Roger 	 *	to acquire the old bucket lock.
    289   9393      Roger 	 *
    290   9393      Roger 	 *	So we adopt the tactic of keeping all of the retired
    291   9393      Roger 	 *	hash tables on a linked list, so they can be safely
    292   9393      Roger 	 *	freed when the process exits or execs.
    293   9393      Roger 	 *
    294   9393      Roger 	 *	Because the hash table grows in powers of two, the
    295   9393      Roger 	 *	total size of all of the hash tables will be slightly
    296   9393      Roger 	 *	less than twice the size of the largest hash table.
    297      0     stevel 	 */
    298      0     stevel 	while (p->p_lwpfree == NULL) {
    299      0     stevel 		uint_t dirsz = p->p_lwpdir_sz;
    300   9393      Roger 		lwpdir_t *new_dir;
    301      0     stevel 		uint_t new_dirsz;
    302   9393      Roger 		lwpdir_t *ldp;
    303   9393      Roger 		tidhash_t *new_hash;
    304      0     stevel 		uint_t new_hashsz;
    305      0     stevel 
    306      0     stevel 		mutex_exit(&p->p_lock);
    307      0     stevel 
    308   9393      Roger 		/*
    309   9393      Roger 		 * Prepare to remember the old p_tidhash for later
    310   9393      Roger 		 * kmem_free()ing when the process exits or execs.
    311   9393      Roger 		 */
    312   9393      Roger 		if (ret_tidhash == NULL)
    313   9393      Roger 			ret_tidhash = kmem_zalloc(sizeof (ret_tidhash_t),
    314   9393      Roger 			    KM_SLEEP);
    315   9393      Roger 		if (old_dir != NULL)
    316      0     stevel 			kmem_free(old_dir, old_dirsz * sizeof (*old_dir));
    317   9393      Roger 		if (old_hash != NULL)
    318      0     stevel 			kmem_free(old_hash, old_hashsz * sizeof (*old_hash));
    319   9393      Roger 
    320      0     stevel 		new_dirsz = 2 * dirsz + 2;
    321      0     stevel 		new_dir = kmem_zalloc(new_dirsz * sizeof (lwpdir_t), KM_SLEEP);
    322      0     stevel 		for (ldp = new_dir, i = 1; i < new_dirsz; i++, ldp++)
    323      0     stevel 			ldp->ld_next = ldp + 1;
    324      0     stevel 		new_hashsz = (new_dirsz + 2) / 2;
    325   9393      Roger 		new_hash = kmem_zalloc(new_hashsz * sizeof (tidhash_t),
    326   5834   pt157919 		    KM_SLEEP);
    327      0     stevel 
    328      0     stevel 		mutex_enter(&p->p_lock);
    329      0     stevel 		if (p == curproc)
    330      0     stevel 			prbarrier(p);
    331      0     stevel 
    332      0     stevel 		if (dirsz != p->p_lwpdir_sz || p->p_lwpfree != NULL) {
    333      0     stevel 			/*
    334      0     stevel 			 * Someone else beat us to it or some lwp exited.
    335      0     stevel 			 * Set up to free our memory and take a lap.
    336      0     stevel 			 */
    337      0     stevel 			old_dir = new_dir;
    338      0     stevel 			old_dirsz = new_dirsz;
    339      0     stevel 			old_hash = new_hash;
    340      0     stevel 			old_hashsz = new_hashsz;
    341      0     stevel 		} else {
    342   9393      Roger 			/*
    343   9393      Roger 			 * For the benefit of lwp_hash_lookup_and_lock(),
    344   9393      Roger 			 * called from lwp_unpark(), which searches the
    345   9393      Roger 			 * tid hash table without acquiring p->p_lock,
    346   9393      Roger 			 * we must acquire all of the tid hash table
    347   9393      Roger 			 * locks before replacing p->p_tidhash.
    348   9393      Roger 			 */
    349      0     stevel 			old_hash = p->p_tidhash;
    350      0     stevel 			old_hashsz = p->p_tidhash_sz;
    351   9393      Roger 			for (i = 0; i < old_hashsz; i++) {
    352   9393      Roger 				mutex_enter(&old_hash[i].th_lock);
    353   9393      Roger 				mutex_enter(&new_hash[i].th_lock);
    354   9393      Roger 			}
    355   9393      Roger 
    356      0     stevel 			/*
    357      0     stevel 			 * We simply hash in all of the old directory entries.
    358      0     stevel 			 * This works because the old directory has no empty
    359      0     stevel 			 * slots and the new hash table starts out empty.
    360      0     stevel 			 * This reproduces the original directory ordering
    361      0     stevel 			 * (required for /proc directory semantics).
    362      0     stevel 			 */
    363   9393      Roger 			old_dir = p->p_lwpdir;
    364   9393      Roger 			old_dirsz = p->p_lwpdir_sz;
    365   9393      Roger 			p->p_lwpdir = new_dir;
    366   9393      Roger 			p->p_lwpfree = new_dir;
    367   9393      Roger 			p->p_lwpdir_sz = new_dirsz;
    368   9393      Roger 			for (ldp = old_dir, i = 0; i < old_dirsz; i++, ldp++)
    369   9393      Roger 				lwp_hash_in(p, ldp->ld_entry,
    370   9393      Roger 				    new_hash, new_hashsz, 0);
    371   9393      Roger 
    372      0     stevel 			/*
    373   9393      Roger 			 * Remember the old hash table along with all
    374   9393      Roger 			 * of the previously-remembered hash tables.
    375   9393      Roger 			 * We will free them at process exit or exec.
    376      0     stevel 			 */
    377   9393      Roger 			ret_tidhash->rth_tidhash = old_hash;
    378   9393      Roger 			ret_tidhash->rth_tidhash_sz = old_hashsz;
    379   9393      Roger 			ret_tidhash->rth_next = p->p_ret_tidhash;
    380   9393      Roger 			p->p_ret_tidhash = ret_tidhash;
    381   9393      Roger 
    382   9393      Roger 			/*
    383   9393      Roger 			 * Now establish the new tid hash table.
    384   9393      Roger 			 * As soon as we assign p->p_tidhash,
    385   9393      Roger 			 * code in lwp_unpark() can start using it.
    386   9393      Roger 			 */
    387   9393      Roger 			membar_producer();
    388   9393      Roger 			p->p_tidhash = new_hash;
    389   9393      Roger 
    390   9393      Roger 			/*
    391   9393      Roger 			 * It is necessary that p_tidhash reach global
    392   9393      Roger 			 * visibility before p_tidhash_sz.  Otherwise,
    393   9393      Roger 			 * code in lwp_hash_lookup_and_lock() could
    394   9393      Roger 			 * index into the old p_tidhash using the new
    395   9393      Roger 			 * p_tidhash_sz and thereby access invalid data.
    396   9393      Roger 			 */
    397   9393      Roger 			membar_producer();
    398   9393      Roger 			p->p_tidhash_sz = new_hashsz;
    399   9393      Roger 
    400   9393      Roger 			/*
    401   9393      Roger 			 * Release the locks; allow lwp_unpark() to carry on.
    402   9393      Roger 			 */
    403   9393      Roger 			for (i = 0; i < old_hashsz; i++) {
    404   9393      Roger 				mutex_exit(&old_hash[i].th_lock);
    405   9393      Roger 				mutex_exit(&new_hash[i].th_lock);
    406   9393      Roger 			}
    407   9393      Roger 
    408   9393      Roger 			/*
    409   9393      Roger 			 * Avoid freeing these objects below.
    410   9393      Roger 			 */
    411   9393      Roger 			ret_tidhash = NULL;
    412   9393      Roger 			old_hash = NULL;
    413   9393      Roger 			old_hashsz = 0;
    414      0     stevel 		}
    415      0     stevel 	}
    416      0     stevel 
    417      0     stevel 	/*
    418      0     stevel 	 * Block the process against /proc while we manipulate p->p_tlist,
    419      0     stevel 	 * unless lwp_create() was called by /proc for the PCAGENT operation.
    420      0     stevel 	 * We want to do this early enough so that we don't drop p->p_lock
    421      0     stevel 	 * until the thread is put on the p->p_tlist.
    422      0     stevel 	 */
    423      0     stevel 	if (p == curproc) {
    424      0     stevel 		prbarrier(p);
    425      0     stevel 		/*
    426      0     stevel 		 * If the current lwp has been requested to stop, do so now.
    427      0     stevel 		 * Otherwise we have a race condition between /proc attempting
    428      0     stevel 		 * to stop the process and this thread creating a new lwp
    429      0     stevel 		 * that was not seen when the /proc PCSTOP request was issued.
    430      0     stevel 		 * We rely on stop() to call prbarrier(p) before returning.
    431      0     stevel 		 */
    432      0     stevel 		while ((curthread->t_proc_flag & TP_PRSTOP) &&
    433  10541  Gangadhar 		    !ttolwp(curthread)->lwp_nostop) {
    434  10541  Gangadhar 			/*
    435  10541  Gangadhar 			 * We called pool_barrier_enter() before calling
    436  10541  Gangadhar 			 * here to lwp_create(). We have to call
    437  10541  Gangadhar 			 * pool_barrier_exit() before stopping.
    438  10541  Gangadhar 			 */
    439  10541  Gangadhar 			pool_barrier_exit();
    440  10541  Gangadhar 			prbarrier(p);
    441      0     stevel 			stop(PR_REQUESTED, 0);
    442  10541  Gangadhar 			/*
    443  10541  Gangadhar 			 * And we have to repeat the call to
    444  10541  Gangadhar 			 * pool_barrier_enter after stopping.
    445  10541  Gangadhar 			 */
    446  10541  Gangadhar 			pool_barrier_enter();
    447  10541  Gangadhar 			prbarrier(p);
    448  10541  Gangadhar 		}
    449      0     stevel 
    450      0     stevel 		/*
    451      0     stevel 		 * If process is exiting, there could be a race between
    452      0     stevel 		 * the agent lwp creation and the new lwp currently being
    453      0     stevel 		 * created. So to prevent this race lwp creation is failed
    454      0     stevel 		 * if the process is exiting.
    455      0     stevel 		 */
    456      0     stevel 		if (p->p_flag & (SEXITLWPS|SKILLED)) {
    457      0     stevel 			err = 1;
    458      0     stevel 			goto error;
    459      0     stevel 		}
    460      0     stevel 
    461      0     stevel 		/*
    462      0     stevel 		 * Since we might have dropped p->p_lock, the
    463      0     stevel 		 * lwp directory free list might have changed.
    464      0     stevel 		 */
    465      0     stevel 		if (p->p_lwpfree == NULL)
    466      0     stevel 			goto grow;
    467      0     stevel 	}
    468      0     stevel 
    469      0     stevel 	kpreempt_disable();	/* can't grab cpu_lock here */
    470      0     stevel 
    471      0     stevel 	/*
    472  11173   Jonathan 	 * Inherit processor and processor set bindings from curthread.
    473  11173   Jonathan 	 *
    474  11173   Jonathan 	 * For kernel LWPs, we do not inherit processor set bindings at
    475  11173   Jonathan 	 * process creation time (i.e. when p != curproc).  After the
    476  11173   Jonathan 	 * kernel process is created, any subsequent LWPs must be created
    477  11173   Jonathan 	 * by threads in the kernel process, at which point we *will*
    478  11173   Jonathan 	 * inherit processor set bindings.
    479      0     stevel 	 */
    480  11173   Jonathan 	if (CLASS_KERNEL(cid) && p != curproc) {
    481      0     stevel 		t->t_bind_cpu = binding = PBIND_NONE;
    482      0     stevel 		t->t_cpupart = oldpart = &cp_default;
    483      0     stevel 		t->t_bind_pset = PS_NONE;
    484   6298      akolb 		t->t_bindflag = (uchar_t)default_binding_mode;
    485      0     stevel 	} else {
    486      0     stevel 		binding = curthread->t_bind_cpu;
    487      0     stevel 		t->t_bind_cpu = binding;
    488      0     stevel 		oldpart = t->t_cpupart;
    489      0     stevel 		t->t_cpupart = curthread->t_cpupart;
    490      0     stevel 		t->t_bind_pset = curthread->t_bind_pset;
    491   6298      akolb 		t->t_bindflag = curthread->t_bindflag |
    492   6298      akolb 		    (uchar_t)default_binding_mode;
    493      0     stevel 	}
    494      0     stevel 
    495      0     stevel 	/*
    496      0     stevel 	 * thread_create() initializes this thread's home lgroup to the root.
    497      0     stevel 	 * Choose a more suitable lgroup, since this thread is associated
    498      0     stevel 	 * with an lwp.
    499      0     stevel 	 */
    500      0     stevel 	ASSERT(oldpart != NULL);
    501      0     stevel 	if (binding != PBIND_NONE && t->t_affinitycnt == 0) {
    502      0     stevel 		t->t_bound_cpu = cpu[binding];
    503      0     stevel 		if (t->t_lpl != t->t_bound_cpu->cpu_lpl)
    504      0     stevel 			lgrp_move_thread(t, t->t_bound_cpu->cpu_lpl, 1);
    505      0     stevel 	} else {
    506      0     stevel 		lgrp_move_thread(t, lgrp_choose(t, t->t_cpupart), 1);
    507      0     stevel 	}
    508      0     stevel 
    509      0     stevel 	kpreempt_enable();
    510      0     stevel 
    511      0     stevel 	/*
    512      0     stevel 	 * make sure lpl points to our own partition
    513      0     stevel 	 */
    514      0     stevel 	ASSERT(t->t_lpl >= t->t_cpupart->cp_lgrploads);
    515      0     stevel 	ASSERT(t->t_lpl < t->t_cpupart->cp_lgrploads +
    516      0     stevel 	    t->t_cpupart->cp_nlgrploads);
    517      0     stevel 
    518      0     stevel 	/*
    519      0     stevel 	 * If we're creating a new process, then inherit the project from our
    520      0     stevel 	 * parent. If we're only creating an additional lwp then use the
    521      0     stevel 	 * project pointer of the target process.
    522      0     stevel 	 */
    523      0     stevel 	if (p->p_task == NULL)
    524      0     stevel 		newkpj = ttoproj(curthread);
    525      0     stevel 	else
    526      0     stevel 		newkpj = p->p_task->tk_proj;
    527      0     stevel 
    528      0     stevel 	/*
    529      0     stevel 	 * It is safe to point the thread to the new project without holding it
    530      0     stevel 	 * since we're holding the target process' p_lock here and therefore
    531      0     stevel 	 * we're guaranteed that it will not move to another project.
    532      0     stevel 	 */
    533      0     stevel 	oldkpj = ttoproj(t);
    534      0     stevel 	if (newkpj != oldkpj) {
    535      0     stevel 		t->t_proj = newkpj;
    536      0     stevel 		(void) project_hold(newkpj);
    537      0     stevel 		project_rele(oldkpj);
    538      0     stevel 	}
    539      0     stevel 
    540      0     stevel 	if (cid != NOCLASS) {
    541      0     stevel 		/*
    542      0     stevel 		 * If the lwp is being created in the current process
    543      0     stevel 		 * and matches the current thread's scheduling class,
    544      0     stevel 		 * we should propagate the current thread's scheduling
    545      0     stevel 		 * parameters by calling CL_FORK.  Otherwise just use
    546      0     stevel 		 * the defaults by calling CL_ENTERCLASS.
    547      0     stevel 		 */
    548      0     stevel 		if (p != curproc || curthread->t_cid != cid) {
    549      0     stevel 			err = CL_ENTERCLASS(t, cid, NULL, NULL, bufp);
    550      0     stevel 			t->t_pri = pri;	/* CL_ENTERCLASS may have changed it */
    551   6247        raf 			/*
    552   6247        raf 			 * We don't call schedctl_set_cidpri(t) here
    553   6247        raf 			 * because the schedctl data is not yet set
    554   6247        raf 			 * up for the newly-created lwp.
    555   6247        raf 			 */
    556      0     stevel 		} else {
    557      0     stevel 			t->t_clfuncs = &(sclass[cid].cl_funcs->thread);
    558      0     stevel 			err = CL_FORK(curthread, t, bufp);
    559      0     stevel 			t->t_cid = cid;
    560      0     stevel 		}
    561      0     stevel 		if (err)
    562      0     stevel 			goto error;
    563      0     stevel 		else
    564      0     stevel 			bufp = NULL;
    565      0     stevel 	}
    566      0     stevel 
    567      0     stevel 	/*
    568      0     stevel 	 * If we were given an lwpid then use it, else allocate one.
    569      0     stevel 	 */
    570      0     stevel 	if (lwpid != 0)
    571      0     stevel 		t->t_tid = lwpid;
    572      0     stevel 	else {
    573      0     stevel 		/*
    574      0     stevel 		 * lwp/thread id 0 is never valid; reserved for special checks.
    575      0     stevel 		 * lwp/thread id 1 is reserved for the main thread.
    576      0     stevel 		 * Start again at 2 when INT_MAX has been reached
    577      0     stevel 		 * (id_t is a signed 32-bit integer).
    578      0     stevel 		 */
    579      0     stevel 		id_t prev_id = p->p_lwpid;	/* last allocated tid */
    580      0     stevel 
    581      0     stevel 		do {			/* avoid lwpid duplication */
    582      0     stevel 			if (p->p_lwpid == INT_MAX) {
    583      0     stevel 				p->p_flag |= SLWPWRAP;
    584      0     stevel 				p->p_lwpid = 1;
    585      0     stevel 			}
    586      0     stevel 			if ((t->t_tid = ++p->p_lwpid) == prev_id) {
    587      0     stevel 				/*
    588      0     stevel 				 * All lwpids are allocated; fail the request.
    589      0     stevel 				 */
    590      0     stevel 				err = 1;
    591      0     stevel 				goto error;
    592      0     stevel 			}
    593      0     stevel 			/*
    594      0     stevel 			 * We only need to worry about colliding with an id
    595      0     stevel 			 * that's already in use if this process has
    596      0     stevel 			 * cycled through all available lwp ids.
    597      0     stevel 			 */
    598      0     stevel 			if ((p->p_flag & SLWPWRAP) == 0)
    599      0     stevel 				break;
    600      0     stevel 		} while (lwp_hash_lookup(p, t->t_tid) != NULL);
    601      0     stevel 	}
    602   2712    nn35248 
    603   2712    nn35248 	/*
    604   2712    nn35248 	 * If this is a branded process, let the brand do any necessary lwp
    605   2712    nn35248 	 * initialization.
    606   2712    nn35248 	 */
    607   2712    nn35248 	if (PROC_IS_BRANDED(p)) {
    608   2712    nn35248 		if (BROP(p)->b_initlwp(lwp)) {
    609   2712    nn35248 			err = 1;
    610   2712    nn35248 			goto error;
    611   2712    nn35248 		}
    612   2712    nn35248 		branded = 1;
    613   2712    nn35248 	}
    614   2712    nn35248 
    615   4426   aguzovsk 	if (t->t_tid == 1) {
    616   4426   aguzovsk 		kpreempt_disable();
    617   4426   aguzovsk 		ASSERT(t->t_lpl != NULL);
    618   4426   aguzovsk 		p->p_t1_lgrpid = t->t_lpl->lpl_lgrpid;
    619   4426   aguzovsk 		kpreempt_enable();
    620   4426   aguzovsk 		if (p->p_tr_lgrpid != LGRP_NONE &&
    621   4426   aguzovsk 		    p->p_tr_lgrpid != p->p_t1_lgrpid) {
    622   4426   aguzovsk 			lgrp_update_trthr_migrations(1);
    623   4426   aguzovsk 		}
    624   4426   aguzovsk 	}
    625   4426   aguzovsk 
    626      0     stevel 	p->p_lwpcnt++;
    627      0     stevel 	t->t_waitfor = -1;
    628      0     stevel 
    629      0     stevel 	/*
    630      0     stevel 	 * Turn microstate accounting on for thread if on for process.
    631      0     stevel 	 */
    632      0     stevel 	if (p->p_flag & SMSACCT)
    633      0     stevel 		t->t_proc_flag |= TP_MSACCT;
    634      0     stevel 
    635      0     stevel 	/*
    636      0     stevel 	 * If the process has watchpoints, mark the new thread as such.
    637      0     stevel 	 */
    638      0     stevel 	if (pr_watch_active(p))
    639      0     stevel 		watch_enable(t);
    640      0     stevel 
    641      0     stevel 	/*
    642      0     stevel 	 * The lwp is being created in the stopped state.
    643      0     stevel 	 * We set all the necessary flags to indicate that fact here.
    644      0     stevel 	 * We omit the TS_CREATE flag from t_schedflag so that the lwp
    645      0     stevel 	 * cannot be set running until the caller is finished with it,
    646      0     stevel 	 * even if lwp_continue() is called on it after we drop p->p_lock.
    647      0     stevel 	 * When the caller is finished with the newly-created lwp,
    648      0     stevel 	 * the caller must call lwp_create_done() to allow the lwp
    649      0     stevel 	 * to be set running.  If the TP_HOLDLWP is left set, the
    650      0     stevel 	 * lwp will suspend itself after reaching system call exit.
    651      0     stevel 	 */
    652      0     stevel 	init_mstate(t, LMS_STOPPED);
    653      0     stevel 	t->t_proc_flag |= TP_HOLDLWP;
    654      0     stevel 	t->t_schedflag |= (TS_ALLSTART & ~(TS_CSTART | TS_CREATE));
    655      0     stevel 	t->t_whystop = PR_SUSPENDED;
    656      0     stevel 	t->t_whatstop = SUSPEND_NORMAL;
    657      0     stevel 	t->t_sig_check = 1;	/* ensure that TP_HOLDLWP is honored */
    658      0     stevel 
    659      0     stevel 	/*
    660      0     stevel 	 * Set system call processing flags in case tracing or profiling
    661      0     stevel 	 * is set.  The first system call will evaluate these and turn
    662      0     stevel 	 * them off if they aren't needed.
    663      0     stevel 	 */
    664      0     stevel 	t->t_pre_sys = 1;
    665      0     stevel 	t->t_post_sys = 1;
    666      0     stevel 
    667      0     stevel 	/*
    668      0     stevel 	 * Insert the new thread into the list of all threads.
    669      0     stevel 	 */
    670      0     stevel 	if ((tx = p->p_tlist) == NULL) {
    671      0     stevel 		t->t_back = t;
    672      0     stevel 		t->t_forw = t;
    673      0     stevel 		p->p_tlist = t;
    674      0     stevel 	} else {
    675      0     stevel 		t->t_forw = tx;
    676      0     stevel 		t->t_back = tx->t_back;
    677      0     stevel 		tx->t_back->t_forw = t;
    678      0     stevel 		tx->t_back = t;
    679      0     stevel 	}
    680      0     stevel 
    681      0     stevel 	/*
    682      0     stevel 	 * Insert the new lwp into an lwp directory slot position
    683      0     stevel 	 * and into the lwpid hash table.
    684      0     stevel 	 */
    685      0     stevel 	lep->le_thread = t;
    686      0     stevel 	lep->le_lwpid = t->t_tid;
    687      0     stevel 	lep->le_start = t->t_start;
    688   9393      Roger 	lwp_hash_in(p, lep, p->p_tidhash, p->p_tidhash_sz, 1);
    689      0     stevel 
    690      0     stevel 	if (state == TS_RUN) {
    691      0     stevel 		/*
    692      0     stevel 		 * We set the new lwp running immediately.
    693      0     stevel 		 */
    694      0     stevel 		t->t_proc_flag &= ~TP_HOLDLWP;
    695      0     stevel 		lwp_create_done(t);
    696      0     stevel 	}
    697      0     stevel 
    698      0     stevel error:
    699      0     stevel 	if (err) {
    700  11173   Jonathan 		if (CLASS_KERNEL(cid)) {
    701  11173   Jonathan 			/*
    702  11173   Jonathan 			 * This should only happen if a system process runs
    703  11173   Jonathan 			 * out of lwpids, which shouldn't occur.
    704  11173   Jonathan 			 */
    705  11173   Jonathan 			panic("Failed to create a system LWP");
    706  11173   Jonathan 		}
    707      0     stevel 		/*
    708      0     stevel 		 * We have failed to create an lwp, so decrement the number
    709      0     stevel 		 * of lwps in the task and let the lgroup load averages know
    710      0     stevel 		 * that this thread isn't going to show up.
    711      0     stevel 		 */
    712      0     stevel 		kpreempt_disable();
    713      0     stevel 		lgrp_move_thread(t, NULL, 1);
    714      0     stevel 		kpreempt_enable();
    715      0     stevel 
    716      0     stevel 		ASSERT(MUTEX_HELD(&p->p_lock));
    717      0     stevel 		mutex_enter(&p->p_zone->zone_nlwps_lock);
    718      0     stevel 		p->p_task->tk_nlwps--;
    719      0     stevel 		p->p_task->tk_proj->kpj_nlwps--;
    720      0     stevel 		p->p_zone->zone_nlwps--;
    721      0     stevel 		mutex_exit(&p->p_zone->zone_nlwps_lock);
    722      0     stevel 		if (cid != NOCLASS && bufp != NULL)
    723      0     stevel 			CL_FREE(cid, bufp);
    724   2712    nn35248 
    725   2712    nn35248 		if (branded)
    726   2712    nn35248 			BROP(p)->b_freelwp(lwp);
    727      0     stevel 
    728      0     stevel 		mutex_exit(&p->p_lock);
    729      0     stevel 		t->t_state = TS_FREE;
    730      0     stevel 		thread_rele(t);
    731      0     stevel 
    732      0     stevel 		/*
    733      0     stevel 		 * We need to remove t from the list of all threads
    734      0     stevel 		 * because thread_exit()/lwp_exit() isn't called on t.
    735      0     stevel 		 */
    736      0     stevel 		mutex_enter(&pidlock);
    737      0     stevel 		ASSERT(t != t->t_next);		/* t0 never exits */
    738      0     stevel 		t->t_next->t_prev = t->t_prev;
    739      0     stevel 		t->t_prev->t_next = t->t_next;
    740      0     stevel 		mutex_exit(&pidlock);
    741      0     stevel 
    742      0     stevel 		thread_free(t);
    743      0     stevel 		kmem_free(lep, sizeof (*lep));
    744      0     stevel 		lwp = NULL;
    745      0     stevel 	} else {
    746      0     stevel 		mutex_exit(&p->p_lock);
    747      0     stevel 	}
    748      0     stevel 
    749   9393      Roger 	if (old_dir != NULL)
    750      0     stevel 		kmem_free(old_dir, old_dirsz * sizeof (*old_dir));
    751   9393      Roger 	if (old_hash != NULL)
    752      0     stevel 		kmem_free(old_hash, old_hashsz * sizeof (*old_hash));
    753   9393      Roger 	if (ret_tidhash != NULL)
    754   9393      Roger 		kmem_free(ret_tidhash, sizeof (ret_tidhash_t));
    755      0     stevel 
    756      0     stevel 	DTRACE_PROC1(lwp__create, kthread_t *, t);
    757      0     stevel 	return (lwp);
    758      0     stevel }
    759      0     stevel 
    760      0     stevel /*
    761      0     stevel  * lwp_create_done() is called by the caller of lwp_create() to set the
    762      0     stevel  * newly-created lwp running after the caller has finished manipulating it.
    763      0     stevel  */
    764      0     stevel void
    765      0     stevel lwp_create_done(kthread_t *t)
    766      0     stevel {
    767      0     stevel 	proc_t *p = ttoproc(t);
    768      0     stevel 
    769      0     stevel 	ASSERT(MUTEX_HELD(&p->p_lock));
    770      0     stevel 
    771      0     stevel 	/*
    772      0     stevel 	 * We set the TS_CREATE and TS_CSTART flags and call setrun_locked().
    773      0     stevel 	 * (The absence of the TS_CREATE flag prevents the lwp from running
    774      0     stevel 	 * until we are finished with it, even if lwp_continue() is called on
    775      0     stevel 	 * it by some other lwp in the process or elsewhere in the kernel.)
    776      0     stevel 	 */
    777      0     stevel 	thread_lock(t);
    778      0     stevel 	ASSERT(t->t_state == TS_STOPPED && !(t->t_schedflag & TS_CREATE));
    779      0     stevel 	/*
    780      0     stevel 	 * If TS_CSTART is set, lwp_continue(t) has been called and
    781      0     stevel 	 * has already incremented p_lwprcnt; avoid doing this twice.
    782      0     stevel 	 */
    783      0     stevel 	if (!(t->t_schedflag & TS_CSTART))
    784      0     stevel 		p->p_lwprcnt++;
    785      0     stevel 	t->t_schedflag |= (TS_CSTART | TS_CREATE);
    786      0     stevel 	setrun_locked(t);
    787      0     stevel 	thread_unlock(t);
    788      0     stevel }
    789      0     stevel 
    790      0     stevel /*
    791      0     stevel  * Copy an LWP's active templates, and clear the latest contracts.
    792      0     stevel  */
    793      0     stevel void
    794      0     stevel lwp_ctmpl_copy(klwp_t *dst, klwp_t *src)
    795      0     stevel {
    796      0     stevel 	int i;
    797      0     stevel 
    798      0     stevel 	for (i = 0; i < ct_ntypes; i++) {
    799      0     stevel 		dst->lwp_ct_active[i] = ctmpl_dup(src->lwp_ct_active[i]);
    800      0     stevel 		dst->lwp_ct_latest[i] = NULL;
    801      0     stevel 	}
    802      0     stevel }
    803      0     stevel 
    804      0     stevel /*
    805      0     stevel  * Clear an LWP's contract template state.
    806      0     stevel  */
    807      0     stevel void
    808      0     stevel lwp_ctmpl_clear(klwp_t *lwp)
    809      0     stevel {
    810      0     stevel 	ct_template_t *tmpl;
    811      0     stevel 	int i;
    812      0     stevel 
    813      0     stevel 	for (i = 0; i < ct_ntypes; i++) {
    814      0     stevel 		if ((tmpl = lwp->lwp_ct_active[i]) != NULL) {
    815      0     stevel 			ctmpl_free(tmpl);
    816      0     stevel 			lwp->lwp_ct_active[i] = NULL;
    817      0     stevel 		}
    818      0     stevel 
    819      0     stevel 		if (lwp->lwp_ct_latest[i] != NULL) {
    820      0     stevel 			contract_rele(lwp->lwp_ct_latest[i]);
    821      0     stevel 			lwp->lwp_ct_latest[i] = NULL;
    822      0     stevel 		}
    823      0     stevel 	}
    824      0     stevel }
    825      0     stevel 
    826      0     stevel /*
    827      0     stevel  * Individual lwp exit.
    828      0     stevel  * If this is the last lwp, exit the whole process.
    829      0     stevel  */
    830      0     stevel void
    831      0     stevel lwp_exit(void)
    832      0     stevel {
    833      0     stevel 	kthread_t *t = curthread;
    834      0     stevel 	klwp_t *lwp = ttolwp(t);
    835      0     stevel 	proc_t *p = ttoproc(t);
    836      0     stevel 
    837      0     stevel 	ASSERT(MUTEX_HELD(&p->p_lock));
    838      0     stevel 
    839      0     stevel 	mutex_exit(&p->p_lock);
    840      0     stevel 
    841      0     stevel #if defined(__sparc)
    842      0     stevel 	/*
    843      0     stevel 	 * Ensure that the user stack is fully abandoned..
    844      0     stevel 	 */
    845      0     stevel 	trash_user_windows();
    846      0     stevel #endif
    847      0     stevel 
    848      0     stevel 	tsd_exit();			/* free thread specific data */
    849      0     stevel 
    850      0     stevel 	kcpc_passivate();		/* Clean up performance counter state */
    851      0     stevel 
    852      0     stevel 	pollcleanup();
    853      0     stevel 
    854      0     stevel 	if (t->t_door)
    855      0     stevel 		door_slam();
    856      0     stevel 
    857      0     stevel 	if (t->t_schedctl != NULL)
    858      0     stevel 		schedctl_lwp_cleanup(t);
    859      0     stevel 
    860      0     stevel 	if (t->t_upimutex != NULL)
    861      0     stevel 		upimutex_cleanup();
    862   2712    nn35248 
    863   2712    nn35248 	/*
    864   2712    nn35248 	 * Perform any brand specific exit processing, then release any
    865   2712    nn35248 	 * brand data associated with the lwp
    866   2712    nn35248 	 */
    867   2712    nn35248 	if (PROC_IS_BRANDED(p))
    868   2712    nn35248 		BROP(p)->b_lwpexit(lwp);
    869      0     stevel 
    870      0     stevel 	mutex_enter(&p->p_lock);
    871      0     stevel 	lwp_cleanup();
    872      0     stevel 
    873      0     stevel 	/*
    874      0     stevel 	 * When this process is dumping core, its lwps are held here
    875      0     stevel 	 * until the core dump is finished. Then exitlwps() is called
    876      0     stevel 	 * again to release these lwps so that they can finish exiting.
    877      0     stevel 	 */
    878      0     stevel 	if (p->p_flag & SCOREDUMP)
    879      0     stevel 		stop(PR_SUSPENDED, SUSPEND_NORMAL);
    880      0     stevel 
    881      0     stevel 	/*
    882      0     stevel 	 * Block the process against /proc now that we have really acquired
    883      0     stevel 	 * p->p_lock (to decrement p_lwpcnt and manipulate p_tlist at least).
    884      0     stevel 	 */
    885      0     stevel 	prbarrier(p);
    886      0     stevel 
    887      0     stevel 	/*
    888      0     stevel 	 * Call proc_exit() if this is the last non-daemon lwp in the process.
    889      0     stevel 	 */
    890      0     stevel 	if (!(t->t_proc_flag & TP_DAEMON) &&
    891      0     stevel 	    p->p_lwpcnt == p->p_lwpdaemon + 1) {
    892      0     stevel 		mutex_exit(&p->p_lock);
    893    390        raf 		if (proc_exit(CLD_EXITED, 0) == 0) {
    894      0     stevel 			/* Restarting init. */
    895      0     stevel 			return;
    896      0     stevel 		}
    897      0     stevel 
    898      0     stevel 		/*
    899      0     stevel 		 * proc_exit() returns a non-zero value when some other
    900      0     stevel 		 * lwp got there first.  We just have to continue in
    901      0     stevel 		 * lwp_exit().
    902      0     stevel 		 */
    903      0     stevel 		mutex_enter(&p->p_lock);
    904      0     stevel 		ASSERT(curproc->p_flag & SEXITLWPS);
    905      0     stevel 		prbarrier(p);
    906      0     stevel 	}
    907      0     stevel 
    908      0     stevel 	DTRACE_PROC(lwp__exit);
    909      0     stevel 
    910      0     stevel 	/*
    911      0     stevel 	 * If the lwp is a detached lwp or if the process is exiting,
    912      0     stevel 	 * remove (lwp_hash_out()) the lwp from the lwp directory.
    913      0     stevel 	 * Otherwise null out the lwp's le_thread pointer in the lwp
    914      0     stevel 	 * directory so that other threads will see it as a zombie lwp.
    915      0     stevel 	 */
    916      0     stevel 	prlwpexit(t);		/* notify /proc */
    917      0     stevel 	if (!(t->t_proc_flag & TP_TWAIT) || (p->p_flag & SEXITLWPS))
    918      0     stevel 		lwp_hash_out(p, t->t_tid);
    919      0     stevel 	else {
    920      0     stevel 		ASSERT(!(t->t_proc_flag & TP_DAEMON));
    921      0     stevel 		p->p_lwpdir[t->t_dslot].ld_entry->le_thread = NULL;
    922      0     stevel 		p->p_zombcnt++;
    923      0     stevel 		cv_broadcast(&p->p_lwpexit);
    924      0     stevel 	}
    925      0     stevel 	if (t->t_proc_flag & TP_DAEMON) {
    926      0     stevel 		p->p_lwpdaemon--;
    927      0     stevel 		t->t_proc_flag &= ~TP_DAEMON;
    928      0     stevel 	}
    929      0     stevel 	t->t_proc_flag &= ~TP_TWAIT;
    930      0     stevel 
    931      0     stevel 	/*
    932      0     stevel 	 * Maintain accurate lwp count for task.max-lwps resource control.
    933      0     stevel 	 */
    934      0     stevel 	mutex_enter(&p->p_zone->zone_nlwps_lock);
    935      0     stevel 	p->p_task->tk_nlwps--;
    936      0     stevel 	p->p_task->tk_proj->kpj_nlwps--;
    937      0     stevel 	p->p_zone->zone_nlwps--;
    938      0     stevel 	mutex_exit(&p->p_zone->zone_nlwps_lock);
    939      0     stevel 
    940      0     stevel 	CL_EXIT(t);		/* tell the scheduler that t is exiting */
    941      0     stevel 	ASSERT(p->p_lwpcnt != 0);
    942      0     stevel 	p->p_lwpcnt--;
    943      0     stevel 
    944      0     stevel 	/*
    945      0     stevel 	 * If all remaining non-daemon lwps are waiting in lwp_wait(),
    946      0     stevel 	 * wake them up so someone can return EDEADLK.
    947      0     stevel 	 * (See the block comment preceeding lwp_wait().)
    948      0     stevel 	 */
    949      0     stevel 	if (p->p_lwpcnt == p->p_lwpdaemon + (p->p_lwpwait - p->p_lwpdwait))
    950      0     stevel 		cv_broadcast(&p->p_lwpexit);
    951      0     stevel 
    952      0     stevel 	t->t_proc_flag |= TP_LWPEXIT;
    953      0     stevel 	term_mstate(t);
    954   3792      akolb 
    955      0     stevel #ifndef NPROBE
    956      0     stevel 	/* Kernel probe */
    957      0     stevel 	if (t->t_tnf_tpdp)
    958      0     stevel 		tnf_thread_exit();
    959      0     stevel #endif /* NPROBE */
    960      0     stevel 
    961      0     stevel 	t->t_forw->t_back = t->t_back;
    962      0     stevel 	t->t_back->t_forw = t->t_forw;
    963      0     stevel 	if (t == p->p_tlist)
    964      0     stevel 		p->p_tlist = t->t_forw;
    965      0     stevel 
    966      0     stevel 	/*
    967      0     stevel 	 * Clean up the signal state.
    968      0     stevel 	 */
    969      0     stevel 	if (t->t_sigqueue != NULL)
    970      0     stevel 		sigdelq(p, t, 0);
    971      0     stevel 	if (lwp->lwp_curinfo != NULL) {
    972      0     stevel 		siginfofree(lwp->lwp_curinfo);
    973      0     stevel 		lwp->lwp_curinfo = NULL;
    974      0     stevel 	}
    975      0     stevel 
    976      0     stevel 	thread_rele(t);
    977      0     stevel 
    978      0     stevel 	/*
    979      0     stevel 	 * Terminated lwps are associated with process zero and are put onto
    980      0     stevel 	 * death-row by resume().  Avoid preemption after resetting t->t_procp.
    981      0     stevel 	 */
    982      0     stevel 	t->t_preempt++;
    983   1217        rab 
    984   1217        rab 	if (t->t_ctx != NULL)
    985   1217        rab 		exitctx(t);
    986   1217        rab 	if (p->p_pctx != NULL)
    987   1217        rab 		exitpctx(p);
    988   1217        rab 
    989      0     stevel 	t->t_procp = &p0;
    990      0     stevel 
    991      0     stevel 	/*
    992      0     stevel 	 * Notify the HAT about the change of address space
    993      0     stevel 	 */
    994      0     stevel 	hat_thread_exit(t);
    995      0     stevel 	/*
    996      0     stevel 	 * When this is the last running lwp in this process and some lwp is
    997      0     stevel 	 * waiting for this condition to become true, or this thread was being
    998      0     stevel 	 * suspended, then the waiting lwp is awakened.
    999      0     stevel 	 *
   1000      0     stevel 	 * Also, if the process is exiting, we may have a thread waiting in
   1001      0     stevel 	 * exitlwps() that needs to be notified.
   1002      0     stevel 	 */
   1003      0     stevel 	if (--p->p_lwprcnt == 0 || (t->t_proc_flag & TP_HOLDLWP) ||
   1004      0     stevel 	    (p->p_flag & SEXITLWPS))
   1005      0     stevel 		cv_broadcast(&p->p_holdlwps);
   1006      0     stevel 
   1007      0     stevel 	/*
   1008      0     stevel 	 * Need to drop p_lock so we can reacquire pidlock.
   1009      0     stevel 	 */
   1010      0     stevel 	mutex_exit(&p->p_lock);
   1011      0     stevel 	mutex_enter(&pidlock);
   1012      0     stevel 
   1013      0     stevel 	ASSERT(t != t->t_next);		/* t0 never exits */
   1014      0     stevel 	t->t_next->t_prev = t->t_prev;
   1015      0     stevel 	t->t_prev->t_next = t->t_next;
   1016      0     stevel 	cv_broadcast(&t->t_joincv);	/* wake up anyone in thread_join */
   1017      0     stevel 	mutex_exit(&pidlock);
   1018      0     stevel 
   1019      0     stevel 	lwp_pcb_exit();
   1020      0     stevel 
   1021      0     stevel 	t->t_state = TS_ZOMB;
   1022      0     stevel 	swtch_from_zombie();
   1023      0     stevel 	/* never returns */
   1024      0     stevel }
   1025      0     stevel 
   1026      0     stevel 
   1027      0     stevel /*
   1028      0     stevel  * Cleanup function for an exiting lwp.
   1029      0     stevel  * Called both from lwp_exit() and from proc_exit().
   1030      0     stevel  * p->p_lock is repeatedly released and grabbed in this function.
   1031      0     stevel  */
   1032      0     stevel void
   1033      0     stevel lwp_cleanup(void)
   1034      0     stevel {
   1035      0     stevel 	kthread_t *t = curthread;
   1036      0     stevel 	proc_t *p = ttoproc(t);
   1037      0     stevel 
   1038      0     stevel 	ASSERT(MUTEX_HELD(&p->p_lock));
   1039      0     stevel 
   1040      0     stevel 	/* untimeout any lwp-bound realtime timers */
   1041      0     stevel 	if (p->p_itimer != NULL)
   1042      0     stevel 		timer_lwpexit();
   1043      0     stevel 
   1044      0     stevel 	/*
   1045      0     stevel 	 * If this is the /proc agent lwp that is exiting, readjust p_lwpid
   1046      0     stevel 	 * so it appears that the agent never existed, and clear p_agenttp.
   1047      0     stevel 	 */
   1048      0     stevel 	if (t == p->p_agenttp) {
   1049      0     stevel 		ASSERT(t->t_tid == p->p_lwpid);
   1050      0     stevel 		p->p_lwpid--;
   1051      0     stevel 		p->p_agenttp = NULL;
   1052      0     stevel 	}
   1053      0     stevel 
   1054      0     stevel 	/*
   1055      0     stevel 	 * Do lgroup bookkeeping to account for thread exiting.
   1056      0     stevel 	 */
   1057      0     stevel 	kpreempt_disable();
   1058      0     stevel 	lgrp_move_thread(t, NULL, 1);
   1059   4426   aguzovsk 	if (t->t_tid == 1) {
   1060   4426   aguzovsk 		p->p_t1_lgrpid = LGRP_NONE;
   1061   4426   aguzovsk 	}
   1062      0     stevel 	kpreempt_enable();
   1063      0     stevel 
   1064      0     stevel 	lwp_ctmpl_clear(ttolwp(t));
   1065      0     stevel }
   1066      0     stevel 
   1067      0     stevel int
   1068      0     stevel lwp_suspend(kthread_t *t)
   1069      0     stevel {
   1070      0     stevel 	int tid;
   1071      0     stevel 	proc_t *p = ttoproc(t);
   1072      0     stevel 
   1073      0     stevel 	ASSERT(MUTEX_HELD(&p->p_lock));
   1074      0     stevel 
   1075      0     stevel 	/*
   1076      0     stevel 	 * Set the thread's TP_HOLDLWP flag so it will stop in holdlwp().
   1077      0     stevel 	 * If an lwp is stopping itself, there is no need to wait.
   1078      0     stevel 	 */
   1079    875        raf top:
   1080      0     stevel 	t->t_proc_flag |= TP_HOLDLWP;
   1081      0     stevel 	if (t == curthread) {
   1082      0     stevel 		t->t_sig_check = 1;
   1083      0     stevel 	} else {
   1084      0     stevel 		/*
   1085      0     stevel 		 * Make sure the lwp stops promptly.
   1086      0     stevel 		 */
   1087      0     stevel 		thread_lock(t);
   1088      0     stevel 		t->t_sig_check = 1;
   1089      0     stevel 		/*
   1090      0     stevel 		 * XXX Should use virtual stop like /proc does instead of
   1091      0     stevel 		 * XXX waking the thread to get it to stop.
   1092      0     stevel 		 */
   1093   3792      akolb 		if (ISWAKEABLE(t) || ISWAITING(t)) {
   1094      0     stevel 			setrun_locked(t);
   1095   3792      akolb 		} else if (t->t_state == TS_ONPROC && t->t_cpu != CPU) {
   1096      0     stevel 			poke_cpu(t->t_cpu->cpu_id);
   1097   3792      akolb 		}
   1098   3792      akolb 
   1099      0     stevel 		tid = t->t_tid;	 /* remember thread ID */
   1100      0     stevel 		/*
   1101      0     stevel 		 * Wait for lwp to stop
   1102      0     stevel 		 */
   1103      0     stevel 		while (!SUSPENDED(t)) {
   1104      0     stevel 			/*
   1105      0     stevel 			 * Drop the thread lock before waiting and reacquire it
   1106      0     stevel 			 * afterwards, so the thread can change its t_state
   1107      0     stevel 			 * field.
   1108      0     stevel 			 */
   1109      0     stevel 			thread_unlock(t);
   1110      0     stevel 
   1111      0     stevel 			/*
   1112      0     stevel 			 * Check if aborted by exitlwps().
   1113      0     stevel 			 */
   1114      0     stevel 			if (p->p_flag & SEXITLWPS)
   1115      0     stevel 				lwp_exit();
   1116      0     stevel 
   1117      0     stevel 			/*
   1118      0     stevel 			 * Cooperate with jobcontrol signals and /proc stopping
   1119      0     stevel 			 * by calling cv_wait_sig() to wait for the target
   1120      0     stevel 			 * lwp to stop.  Just using cv_wait() can lead to
   1121      0     stevel 			 * deadlock because, if some other lwp has stopped
   1122      0     stevel 			 * by either of these mechanisms, then p_lwprcnt will
   1123      0     stevel 			 * never become zero if we do a cv_wait().
   1124      0     stevel 			 */
   1125      0     stevel 			if (!cv_wait_sig(&p->p_holdlwps, &p->p_lock))
   1126      0     stevel 				return (EINTR);
   1127      0     stevel 
   1128      0     stevel 			/*
   1129      0     stevel 			 * Check to see if thread died while we were
   1130      0     stevel 			 * waiting for it to suspend.
   1131      0     stevel 			 */
   1132      0     stevel 			if (idtot(p, tid) == NULL)
   1133      0     stevel 				return (ESRCH);
   1134      0     stevel 
   1135      0     stevel 			thread_lock(t);
   1136      0     stevel 			/*
   1137    875        raf 			 * If the TP_HOLDLWP flag went away, lwp_continue()
   1138    875        raf 			 * or vfork() must have been called while we were
   1139    875        raf 			 * waiting, so start over again.
   1140      0     stevel 			 */
   1141      0     stevel 			if ((t->t_proc_flag & TP_HOLDLWP) == 0) {
   1142      0     stevel 				thread_unlock(t);
   1143    875        raf 				goto top;
   1144      0     stevel 			}
   1145      0     stevel 		}
   1146      0     stevel 		thread_unlock(t);
   1147      0     stevel 	}
   1148      0     stevel 	return (0);
   1149      0     stevel }
   1150      0     stevel 
   1151      0     stevel /*
   1152      0     stevel  * continue a lwp that's been stopped by lwp_suspend().
   1153      0     stevel  */
   1154      0     stevel void
   1155      0     stevel lwp_continue(kthread_t *t)
   1156      0     stevel {
   1157      0     stevel 	proc_t *p = ttoproc(t);
   1158      0     stevel 	int was_suspended = t->t_proc_flag & TP_HOLDLWP;
   1159      0     stevel 
   1160      0     stevel 	ASSERT(MUTEX_HELD(&p->p_lock));
   1161      0     stevel 
   1162      0     stevel 	t->t_proc_flag &= ~TP_HOLDLWP;
   1163      0     stevel 	thread_lock(t);
   1164      0     stevel 	if (SUSPENDED(t) &&
   1165      0     stevel 	    !(p->p_flag & (SHOLDFORK | SHOLDFORK1 | SHOLDWATCH))) {
   1166      0     stevel 		p->p_lwprcnt++;
   1167      0     stevel 		t->t_schedflag |= TS_CSTART;
   1168      0     stevel 		setrun_locked(t);
   1169      0     stevel 	}
   1170      0     stevel 	thread_unlock(t);
   1171      0     stevel 	/*
   1172      0     stevel 	 * Wakeup anyone waiting for this thread to be suspended
   1173      0     stevel 	 */
   1174      0     stevel 	if (was_suspended)
   1175      0     stevel 		cv_broadcast(&p->p_holdlwps);
   1176      0     stevel }
   1177      0     stevel 
   1178      0     stevel /*
   1179      0     stevel  * ********************************
   1180      0     stevel  *  Miscellaneous lwp routines	  *
   1181      0     stevel  * ********************************
   1182      0     stevel  */
   1183      0     stevel /*
   1184      0     stevel  * When a process is undergoing a forkall(), its p_flag is set to SHOLDFORK.
   1185      0     stevel  * This will cause the process's lwps to stop at a hold point.  A hold
   1186      0     stevel  * point is where a kernel thread has a flat stack.  This is at the
   1187      0     stevel  * return from a system call and at the return from a user level trap.
   1188      0     stevel  *
   1189      0     stevel  * When a process is undergoing a fork1() or vfork(), its p_flag is set to
   1190      0     stevel  * SHOLDFORK1.  This will cause the process's lwps to stop at a modified
   1191      0     stevel  * hold point.  The lwps in the process are not being cloned, so they
   1192      0     stevel  * are held at the usual hold points and also within issig_forreal().
   1193      0     stevel  * This has the side-effect that their system calls do not return
   1194      0     stevel  * showing EINTR.
   1195      0     stevel  *
   1196      0     stevel  * An lwp can also be held.  This is identified by the TP_HOLDLWP flag on
   1197      0     stevel  * the thread.  The TP_HOLDLWP flag is set in lwp_suspend(), where the active
   1198      0     stevel  * lwp is waiting for the target lwp to be stopped.
   1199      0     stevel  */
   1200      0     stevel void
   1201      0     stevel holdlwp(void)
   1202      0     stevel {
   1203      0     stevel 	proc_t *p = curproc;
   1204      0     stevel 	kthread_t *t = curthread;
   1205      0     stevel 
   1206      0     stevel 	mutex_enter(&p->p_lock);
   1207      0     stevel 	/*
   1208      0     stevel 	 * Don't terminate immediately if the process is dumping core.
   1209      0     stevel 	 * Once the process has dumped core, all lwps are terminated.
   1210      0     stevel 	 */
   1211      0     stevel 	if (!(p->p_flag & SCOREDUMP)) {
   1212      0     stevel 		if ((p->p_flag & SEXITLWPS) || (t->t_proc_flag & TP_EXITLWP))
   1213      0     stevel 			lwp_exit();
   1214      0     stevel 	}
   1215      0     stevel 	if (!(ISHOLD(p)) && !(p->p_flag & (SHOLDFORK1 | SHOLDWATCH))) {
   1216      0     stevel 		mutex_exit(&p->p_lock);
   1217      0     stevel 		return;
   1218      0     stevel 	}
   1219      0     stevel 	/*
   1220      0     stevel 	 * stop() decrements p->p_lwprcnt and cv_signal()s &p->p_holdlwps
   1221      0     stevel 	 * when p->p_lwprcnt becomes zero.
   1222      0     stevel 	 */
   1223      0     stevel 	stop(PR_SUSPENDED, SUSPEND_NORMAL);
   1224      0     stevel 	if (p->p_flag & SEXITLWPS)
   1225      0     stevel 		lwp_exit();
   1226      0     stevel 	mutex_exit(&p->p_lock);
   1227      0     stevel }
   1228      0     stevel 
   1229      0     stevel /*
   1230      0     stevel  * Have all lwps within the process hold at a point where they are
   1231      0     stevel  * cloneable (SHOLDFORK) or just safe w.r.t. fork1 (SHOLDFORK1).
   1232      0     stevel  */
   1233      0     stevel int
   1234      0     stevel holdlwps(int holdflag)
   1235      0     stevel {
   1236      0     stevel 	proc_t *p = curproc;
   1237      0     stevel 
   1238      0     stevel 	ASSERT(holdflag == SHOLDFORK || holdflag == SHOLDFORK1);
   1239      0     stevel 	mutex_enter(&p->p_lock);
   1240      0     stevel 	schedctl_finish_sigblock(curthread);
   1241      0     stevel again:
   1242      0     stevel 	while (p->p_flag & (SEXITLWPS | SHOLDFORK | SHOLDFORK1 | SHOLDWATCH)) {
   1243      0     stevel 		/*
   1244      0     stevel 		 * If another lwp is doing a forkall() or proc_exit(), bail out.
   1245      0     stevel 		 */
   1246      0     stevel 		if (p->p_flag & (SEXITLWPS | SHOLDFORK)) {
   1247      0     stevel 			mutex_exit(&p->p_lock);
   1248      0     stevel 			return (0);
   1249      0     stevel 		}
   1250      0     stevel 		/*
   1251      0     stevel 		 * Another lwp is doing a fork1() or is undergoing
   1252      0     stevel 		 * watchpoint activity.  We hold here for it to complete.
   1253      0     stevel 		 */
   1254      0     stevel 		stop(PR_SUSPENDED, SUSPEND_NORMAL);
   1255      0     stevel 	}
   1256      0     stevel 	p->p_flag |= holdflag;
   1257      0     stevel 	pokelwps(p);
   1258      0     stevel 	--p->p_lwprcnt;
   1259      0     stevel 	/*
   1260      0     stevel 	 * Wait for the process to become quiescent (p->p_lwprcnt == 0).
   1261      0     stevel 	 */
   1262      0     stevel 	while (p->p_lwprcnt > 0) {
   1263      0     stevel 		/*
   1264      0     stevel 		 * Check if aborted by exitlwps().
   1265      0     stevel 		 * Also check if SHOLDWATCH is set; it takes precedence.
   1266      0     stevel 		 */
   1267      0     stevel 		if (p->p_flag & (SEXITLWPS | SHOLDWATCH)) {
   1268      0     stevel 			p->p_lwprcnt++;
   1269      0     stevel 			p->p_flag &= ~holdflag;
   1270      0     stevel 			cv_broadcast(&p->p_holdlwps);
   1271      0     stevel 			goto again;
   1272      0     stevel 		}
   1273      0     stevel 		/*
   1274      0     stevel 		 * Cooperate with jobcontrol signals and /proc stopping.
   1275      0     stevel 		 * If some other lwp has stopped by either of these
   1276      0     stevel 		 * mechanisms, then p_lwprcnt will never become zero
   1277      0     stevel 		 * and the process will appear deadlocked unless we
   1278      0     stevel 		 * stop here in sympathy with the other lwp before
   1279      0     stevel 		 * doing the cv_wait() below.
   1280      0     stevel 		 *
   1281      0     stevel 		 * If the other lwp stops after we do the cv_wait(), it
   1282      0     stevel 		 * will wake us up to loop around and do the sympathy stop.
   1283      0     stevel 		 *
   1284      0     stevel 		 * Since stop() drops p->p_lock, we must start from
   1285      0     stevel 		 * the top again on returning from stop().
   1286      0     stevel 		 */
   1287      0     stevel 		if (p->p_stopsig | (curthread->t_proc_flag & TP_PRSTOP)) {
   1288      0     stevel 			int whystop = p->p_stopsig? PR_JOBCONTROL :
   1289      0     stevel 			    PR_REQUESTED;
   1290      0     stevel 			p->p_lwprcnt++;
   1291      0     stevel 			p->p_flag &= ~holdflag;
   1292      0     stevel 			stop(whystop, p->p_stopsig);
   1293      0     stevel 			goto again;
   1294      0     stevel 		}
   1295      0     stevel 		cv_wait(&p->p_holdlwps, &p->p_lock);
   1296      0     stevel 	}
   1297      0     stevel 	p->p_lwprcnt++;
   1298      0     stevel 	p->p_flag &= ~holdflag;
   1299      0     stevel 	mutex_exit(&p->p_lock);
   1300      0     stevel 	return (1);
   1301      0     stevel }
   1302      0     stevel 
   1303      0     stevel /*
   1304      0     stevel  * See comments for holdwatch(), below.
   1305      0     stevel  */
   1306      0     stevel static int
   1307      0     stevel holdcheck(int clearflags)
   1308      0     stevel {
   1309      0     stevel 	proc_t *p = curproc;
   1310      0     stevel 
   1311      0     stevel 	/*
   1312      0     stevel 	 * If we are trying to exit, that takes precedence over anything else.
   1313      0     stevel 	 */
   1314      0     stevel 	if (p->p_flag & SEXITLWPS) {
   1315      0     stevel 		p->p_lwprcnt++;
   1316      0     stevel 		p->p_flag &= ~clearflags;
   1317      0     stevel 		lwp_exit();
   1318      0     stevel 	}
   1319      0     stevel 
   1320      0     stevel 	/*
   1321      0     stevel 	 * If another thread is calling fork1(), stop the current thread so the
   1322      0     stevel 	 * other can complete.
   1323      0     stevel 	 */
   1324      0     stevel 	if (p->p_flag & SHOLDFORK1) {
   1325      0     stevel 		p->p_lwprcnt++;
   1326      0     stevel 		stop(PR_SUSPENDED, SUSPEND_NORMAL);
   1327      0     stevel 		if (p->p_flag & SEXITLWPS) {
   1328      0     stevel 			p->p_flag &= ~clearflags;
   1329      0     stevel 			lwp_exit();
   1330      0     stevel 		}
   1331      0     stevel 		return (-1);
   1332      0     stevel 	}
   1333      0     stevel 
   1334      0     stevel 	/*
   1335      0     stevel 	 * If another thread is calling fork(), then indicate we are doing
   1336      0     stevel 	 * watchpoint activity.  This will cause holdlwps() above to stop the
   1337      0     stevel 	 * forking thread, at which point we can continue with watchpoint
   1338      0     stevel 	 * activity.
   1339      0     stevel 	 */
   1340      0     stevel 	if (p->p_flag & SHOLDFORK) {
   1341      0     stevel 		p->p_lwprcnt++;
   1342      0     stevel 		while (p->p_flag & SHOLDFORK) {
   1343      0     stevel 			p->p_flag |= SHOLDWATCH;
   1344      0     stevel 			cv_broadcast(&p->p_holdlwps);
   1345      0     stevel 			cv_wait(&p->p_holdlwps, &p->p_lock);
   1346      0     stevel 			p->p_flag &= ~SHOLDWATCH;
   1347      0     stevel 		}
   1348      0     stevel 		return (-1);
   1349      0     stevel 	}
   1350      0     stevel 
   1351      0     stevel 	return (0);
   1352      0     stevel }
   1353      0     stevel 
   1354      0     stevel /*
   1355      0     stevel  * Stop all lwps within the process, holding themselves in the kernel while the
   1356      0     stevel  * active lwp undergoes watchpoint activity.  This is more complicated than
   1357      0     stevel  * expected because stop() relies on calling holdwatch() in order to copyin data
   1358      0     stevel  * from the user's address space.  A double barrier is used to prevent an
   1359      0     stevel  * infinite loop.
   1360      0     stevel  *
   1361      0     stevel  * 	o The first thread into holdwatch() is the 'master' thread and does
   1362      0     stevel  *        the following:
   1363      0     stevel  *
   1364      0     stevel  *              - Sets SHOLDWATCH on the current process
   1365      0     stevel  *              - Sets TP_WATCHSTOP on the current thread
   1366      0     stevel  *              - Waits for all threads to be either stopped or have
   1367      0     stevel  *                TP_WATCHSTOP set.
   1368      0     stevel  *              - Sets the SWATCHOK flag on the process
   1369      0     stevel  *              - Unsets TP_WATCHSTOP
   1370      0     stevel  *              - Waits for the other threads to completely stop
   1371      0     stevel  *              - Unsets SWATCHOK
   1372      0     stevel  *
   1373      0     stevel  * 	o If SHOLDWATCH is already set when we enter this function, then another
   1374      0     stevel  *        thread is already trying to stop this thread.  This 'slave' thread
   1375      0     stevel  *        does the following:
   1376      0     stevel  *
   1377      0     stevel  *              - Sets TP_WATCHSTOP on the current thread
   1378      0     stevel  *              - Waits for SWATCHOK flag to be set
   1379      0     stevel  *              - Calls stop()
   1380      0     stevel  *
   1381      0     stevel  * 	o If SWATCHOK is set on the process, then this function immediately
   1382      0     stevel  *        returns, as we must have been called via stop().
   1383      0     stevel  *
   1384      0     stevel  * In addition, there are other flags that take precedence over SHOLDWATCH:
   1385      0     stevel  *
   1386      0     stevel  * 	o If SEXITLWPS is set, exit immediately.
   1387      0     stevel  *
   1388      0     stevel  * 	o If SHOLDFORK1 is set, wait for fork1() to complete.
   1389      0     stevel  *
   1390      0     stevel  * 	o If SHOLDFORK is set, then watchpoint activity takes precedence In this
   1391      0     stevel  *        case, set SHOLDWATCH, signalling the forking thread to stop first.
   1392      0     stevel  *
   1393      0     stevel  * 	o If the process is being stopped via /proc (TP_PRSTOP is set), then we
   1394      0     stevel  *        stop the current thread.
   1395      0     stevel  *
   1396      0     stevel  * Returns 0 if all threads have been quiesced.  Returns non-zero if not all
   1397      0     stevel  * threads were stopped, or the list of watched pages has changed.
   1398      0     stevel  */
   1399      0     stevel int
   1400      0     stevel holdwatch(void)
   1401      0     stevel {
   1402      0     stevel 	proc_t *p = curproc;
   1403      0     stevel 	kthread_t *t = curthread;
   1404      0     stevel 	int ret = 0;
   1405      0     stevel 
   1406      0     stevel 	mutex_enter(&p->p_lock);
   1407      0     stevel 
   1408      0     stevel 	p->p_lwprcnt--;
   1409      0     stevel 
   1410      0     stevel 	/*
   1411      0     stevel 	 * Check for bail-out conditions as outlined above.
   1412      0     stevel 	 */
   1413      0     stevel 	if (holdcheck(0) != 0) {
   1414      0     stevel 		mutex_exit(&p->p_lock);
   1415      0     stevel 		return (-1);
   1416      0     stevel 	}
   1417      0     stevel 
   1418      0     stevel 	if (!(p->p_flag & SHOLDWATCH)) {
   1419      0     stevel 		/*
   1420      0     stevel 		 * We are the master watchpoint thread.  Set SHOLDWATCH and poke
   1421      0     stevel 		 * the other threads.
   1422      0     stevel 		 */
   1423      0     stevel 		p->p_flag |= SHOLDWATCH;
   1424      0     stevel 		pokelwps(p);
   1425      0     stevel 
   1426      0     stevel 		/*
   1427      0     stevel 		 * Wait for all threads to be stopped or have TP_WATCHSTOP set.
   1428      0     stevel 		 */
   1429      0     stevel 		while (pr_allstopped(p, 1) > 0) {
   1430      0     stevel 			if (holdcheck(SHOLDWATCH) != 0) {
   1431      0     stevel 				p->p_flag &= ~SHOLDWATCH;
   1432      0     stevel 				mutex_exit(&p->p_lock);
   1433      0     stevel 				return (-1);
   1434      0     stevel 			}
   1435      0     stevel 
   1436      0     stevel 			cv_wait(&p->p_holdlwps, &p->p_lock);
   1437      0     stevel 		}
   1438      0     stevel 
   1439      0     stevel 		/*
   1440      0     stevel 		 * All threads are now stopped or in the process of stopping.
   1441      0     stevel 		 * Set SWATCHOK and let them stop completely.
   1442      0     stevel 		 */
   1443      0     stevel 		p->p_flag |= SWATCHOK;
   1444      0     stevel 		t->t_proc_flag &= ~TP_WATCHSTOP;
   1445      0     stevel 		cv_broadcast(&p->p_holdlwps);
   1446      0     stevel 
   1447      0     stevel 		while (pr_allstopped(p, 0) > 0) {
   1448      0     stevel 			/*
   1449      0     stevel 			 * At first glance, it may appear that we don't need a
   1450      0     stevel 			 * call to holdcheck() here.  But if the process gets a
   1451      0     stevel 			 * SIGKILL signal, one of our stopped threads may have
   1452      0     stevel 			 * been awakened and is waiting in exitlwps(), which
   1453      0     stevel 			 * takes precedence over watchpoints.
   1454      0     stevel 			 */
   1455      0     stevel 			if (holdcheck(SHOLDWATCH | SWATCHOK) != 0) {
   1456      0     stevel 				p->p_flag &= ~(SHOLDWATCH | SWATCHOK);
   1457      0     stevel 				mutex_exit(&p->p_lock);
   1458      0     stevel 				return (-1);
   1459      0     stevel 			}
   1460      0     stevel 
   1461      0     stevel 			cv_wait(&p->p_holdlwps, &p->p_lock);
   1462      0     stevel 		}
   1463      0     stevel 
   1464      0     stevel 		/*
   1465      0     stevel 		 * All threads are now completely stopped.
   1466      0     stevel 		 */
   1467      0     stevel 		p->p_flag &= ~SWATCHOK;
   1468      0     stevel 		p->p_flag &= ~SHOLDWATCH;
   1469      0     stevel 		p->p_lwprcnt++;
   1470      0     stevel 
   1471      0     stevel 	} else if (!(p->p_flag & SWATCHOK)) {
   1472      0     stevel 
   1473      0     stevel 		/*
   1474      0     stevel 		 * SHOLDWATCH is set, so another thread is trying to do
   1475      0     stevel 		 * watchpoint activity.  Indicate this thread is stopping, and
   1476      0     stevel 		 * wait for the OK from the master thread.
   1477      0     stevel 		 */
   1478      0     stevel 		t->t_proc_flag |= TP_WATCHSTOP;
   1479      0     stevel 		cv_broadcast(&p->p_holdlwps);
   1480      0     stevel 
   1481      0     stevel 		while (!(p->p_flag & SWATCHOK)) {
   1482      0     stevel 			if (holdcheck(0) != 0) {
   1483      0     stevel 				t->t_proc_flag &= ~TP_WATCHSTOP;
   1484      0     stevel 				mutex_exit(&p->p_lock);
   1485      0     stevel 				return (-1);
   1486      0     stevel 			}
   1487      0     stevel 
   1488      0     stevel 			cv_wait(&p->p_holdlwps, &p->p_lock);
   1489      0     stevel 		}
   1490      0     stevel 
   1491      0     stevel 		/*
   1492      0     stevel 		 * Once the master thread has given the OK, this thread can
   1493      0     stevel 		 * actually call stop().
   1494      0     stevel 		 */
   1495      0     stevel 		t->t_proc_flag &= ~TP_WATCHSTOP;
   1496      0     stevel 		p->p_lwprcnt++;
   1497      0     stevel 
   1498      0     stevel 		stop(PR_SUSPENDED, SUSPEND_NORMAL);
   1499      0     stevel 
   1500      0     stevel 		/*
   1501      0     stevel 		 * It's not OK to do watchpoint activity, notify caller to
   1502      0     stevel 		 * retry.
   1503      0     stevel 		 */
   1504      0     stevel 		ret = -1;
   1505      0     stevel 
   1506      0     stevel 	} else {
   1507      0     stevel 
   1508      0     stevel 		/*
   1509      0     stevel 		 * The only way we can hit the case where SHOLDWATCH is set and
   1510      0     stevel 		 * SWATCHOK is set is if we are triggering this from within a
   1511      0     stevel 		 * stop() call.  Assert that this is the case.
   1512      0     stevel 		 */
   1513      0     stevel 
   1514      0     stevel 		ASSERT(t->t_proc_flag & TP_STOPPING);
   1515      0     stevel 		p->p_lwprcnt++;
   1516      0     stevel 	}
   1517      0     stevel 
   1518      0     stevel 	mutex_exit(&p->p_lock);
   1519      0     stevel 
   1520      0     stevel 	return (ret);
   1521      0     stevel }
   1522      0     stevel 
   1523      0     stevel /*
   1524      0     stevel  * force all interruptible lwps to trap into the kernel.
   1525      0     stevel  */
   1526      0     stevel void
   1527      0     stevel pokelwps(proc_t *p)
   1528      0     stevel {
   1529      0     stevel 	kthread_t *t;
   1530      0     stevel 
   1531      0     stevel 	ASSERT(MUTEX_HELD(&p->p_lock));
   1532      0     stevel 
   1533      0     stevel 	t = p->p_tlist;
   1534      0     stevel 	do {
   1535      0     stevel 		if (t == curthread)
   1536      0     stevel 			continue;
   1537      0     stevel 		thread_lock(t);
   1538      0     stevel 		aston(t);	/* make thread trap or do post_syscall */
   1539   3792      akolb 		if (ISWAKEABLE(t) || ISWAITING(t)) {
   1540   3792      akolb 			setrun_locked(t);
   1541      0     stevel 		} else if (t->t_state == TS_STOPPED) {
   1542      0     stevel 			/*
   1543      0     stevel 			 * Ensure that proc_exit() is not blocked by lwps
   1544      0     stevel 			 * that were stopped via jobcontrol or /proc.
   1545      0     stevel 			 */
   1546      0     stevel 			if (p->p_flag & SEXITLWPS) {
   1547      0     stevel 				p->p_stopsig = 0;
   1548      0     stevel 				t->t_schedflag |= (TS_XSTART | TS_PSTART);
   1549      0     stevel 				setrun_locked(t);
   1550      0     stevel 			}
   1551      0     stevel 			/*
   1552      0     stevel 			 * If we are holding lwps for a forkall(),
   1553      0     stevel 			 * force lwps that have been suspended via
   1554      0     stevel 			 * lwp_suspend() and are suspended inside
   1555      0     stevel 			 * of a system call to proceed to their
   1556      0     stevel 			 * holdlwp() points where they are clonable.
   1557      0     stevel 			 */
   1558      0     stevel 			if ((p->p_flag & SHOLDFORK) && SUSPENDED(t)) {
   1559      0     stevel 				if ((t->t_schedflag & TS_CSTART) == 0) {
   1560      0     stevel 					p->p_lwprcnt++;
   1561      0     stevel 					t->t_schedflag |= TS_CSTART;
   1562      0     stevel 					setrun_locked(t);
   1563      0     stevel 				}
   1564      0     stevel 			}
   1565      0     stevel 		} else if (t->t_state == TS_ONPROC) {
   1566      0     stevel 			if (t->t_cpu != CPU)
   1567      0     stevel 				poke_cpu(t->t_cpu->cpu_id);
   1568      0     stevel 		}
   1569      0     stevel 		thread_unlock(t);
   1570      0     stevel 	} while ((t = t->t_forw) != p->p_tlist);
   1571      0     stevel }
   1572      0     stevel 
   1573      0     stevel /*
   1574      0     stevel  * undo the effects of holdlwps() or holdwatch().
   1575      0     stevel  */
   1576      0     stevel void
   1577      0     stevel continuelwps(proc_t *p)
   1578      0     stevel {
   1579      0     stevel 	kthread_t *t;
   1580      0     stevel 
   1581      0     stevel 	/*
   1582      0     stevel 	 * If this flag is set, then the original holdwatch() didn't actually
   1583      0     stevel 	 * stop the process.  See comments for holdwatch().
   1584      0     stevel 	 */
   1585      0     stevel 	if (p->p_flag & SWATCHOK) {
   1586      0     stevel 		ASSERT(curthread->t_proc_flag & TP_STOPPING);
   1587      0     stevel 		return;
   1588      0     stevel 	}
   1589      0     stevel 
   1590      0     stevel 	ASSERT(MUTEX_HELD(&p->p_lock));
   1591      0     stevel 	ASSERT((p->p_flag & (SHOLDFORK | SHOLDFORK1 | SHOLDWATCH)) == 0);
   1592      0     stevel 
   1593      0     stevel 	t = p->p_tlist;
   1594      0     stevel 	do {
   1595      0     stevel 		thread_lock(t);		/* SUSPENDED looks at t_schedflag */
   1596      0     stevel 		if (SUSPENDED(t) && !(t->t_proc_flag & TP_HOLDLWP)) {
   1597      0     stevel 			p->p_lwprcnt++;
   1598      0     stevel 			t->t_schedflag |= TS_CSTART;
   1599      0     stevel 			setrun_locked(t);
   1600      0     stevel 		}
   1601      0     stevel 		thread_unlock(t);
   1602      0     stevel 	} while ((t = t->t_forw) != p->p_tlist);
   1603      0     stevel }
   1604      0     stevel 
   1605      0     stevel /*
   1606      0     stevel  * Force all other LWPs in the current process other than the caller to exit,
   1607      0     stevel  * and then cv_wait() on p_holdlwps for them to exit.  The exitlwps() function
   1608      0     stevel  * is typically used in these situations:
   1609      0     stevel  *
   1610      0     stevel  *   (a) prior to an exec() system call
   1611      0     stevel  *   (b) prior to dumping a core file
   1612      0     stevel  *   (c) prior to a uadmin() shutdown
   1613      0     stevel  *
   1614      0     stevel  * If the 'coredump' flag is set, other LWPs are quiesced but not destroyed.
   1615      0     stevel  * Multiple threads in the process can call this function at one time by
   1616      0     stevel  * triggering execs or core dumps simultaneously, so the SEXITLWPS bit is used
   1617      0     stevel  * to declare one particular thread the winner who gets to kill the others.
   1618      0     stevel  * If a thread wins the exitlwps() dance, zero is returned; otherwise an
   1619      0     stevel  * appropriate errno value is returned to caller for its system call to return.
   1620      0     stevel  */
   1621      0     stevel int
   1622      0     stevel exitlwps(int coredump)
   1623      0     stevel {
   1624      0     stevel 	proc_t *p = curproc;
   1625      0     stevel 	int heldcnt;
   1626      0     stevel 
   1627      0     stevel 	if (curthread->t_door)
   1628      0     stevel 		door_slam();
   1629      0     stevel 	if (p->p_door_list)
   1630      0     stevel 		door_revoke_all();
   1631      0     stevel 	if (curthread->t_schedctl != NULL)
   1632      0     stevel 		schedctl_lwp_cleanup(curthread);
   1633      0     stevel 
   1634      0     stevel 	/*
   1635      0     stevel 	 * Ensure that before starting to wait for other lwps to exit,
   1636      0     stevel 	 * cleanup all upimutexes held by curthread. Otherwise, some other
   1637      0     stevel 	 * lwp could be waiting (uninterruptibly) for a upimutex held by
   1638      0     stevel 	 * curthread, and the call to pokelwps() below would deadlock.
   1639      0     stevel 	 * Even if a blocked upimutex_lock is made interruptible,
   1640      0     stevel 	 * curthread's upimutexes need to be unlocked: do it here.
   1641      0     stevel 	 */
   1642      0     stevel 	if (curthread->t_upimutex != NULL)
   1643      0     stevel 		upimutex_cleanup();
   1644      0     stevel 
   1645      0     stevel 	/*
   1646      0     stevel 	 * Grab p_lock in order to check and set SEXITLWPS to declare a winner.
   1647      0     stevel 	 * We must also block any further /proc access from this point forward.
   1648      0     stevel 	 */
   1649      0     stevel 	mutex_enter(&p->p_lock);
   1650      0     stevel 	prbarrier(p);
   1651      0     stevel 
   1652      0     stevel 	if (p->p_flag & SEXITLWPS) {
   1653      0     stevel 		mutex_exit(&p->p_lock);
   1654      0     stevel 		aston(curthread);	/* force a trip through post_syscall */
   1655      0     stevel 		return (set_errno(EINTR));
   1656      0     stevel 	}
   1657      0     stevel 
   1658      0     stevel 	p->p_flag |= SEXITLWPS;
   1659      0     stevel 	if (coredump)		/* tell other lwps to stop, not exit */
   1660      0     stevel 		p->p_flag |= SCOREDUMP;
   1661      0     stevel 
   1662      0     stevel 	/*
   1663      0     stevel 	 * Give precedence to exitlwps() if a holdlwps() is
   1664      0     stevel 	 * in progress. The lwp doing the holdlwps() operation
   1665      0     stevel 	 * is aborted when it is awakened.
   1666      0     stevel 	 */
   1667      0     stevel 	while (p->p_flag & (SHOLDFORK | SHOLDFORK1 | SHOLDWATCH)) {
   1668      0     stevel 		cv_broadcast(&p->p_holdlwps);
   1669      0     stevel 		cv_wait(&p->p_holdlwps, &p->p_lock);
   1670    390        raf 		prbarrier(p);
   1671      0     stevel 	}
   1672      0     stevel 	p->p_flag |= SHOLDFORK;
   1673      0     stevel 	pokelwps(p);
   1674      0     stevel 
   1675      0     stevel 	/*
   1676      0     stevel 	 * Wait for process to become quiescent.
   1677      0     stevel 	 */
   1678      0     stevel 	--p->p_lwprcnt;
   1679    390        raf 	while (p->p_lwprcnt > 0) {
   1680      0     stevel 		cv_wait(&p->p_holdlwps, &p->p_lock);
   1681    390        raf 		prbarrier(p);
   1682    390        raf 	}
   1683      0     stevel 	p->p_lwprcnt++;
   1684      0     stevel 	ASSERT(p->p_lwprcnt == 1);
   1685      0     stevel 
   1686      0     stevel 	/*
   1687      0     stevel 	 * The SCOREDUMP flag puts the process into a quiescent
   1688      0     stevel 	 * state.  The process's lwps remain attached to this
   1689      0     stevel 	 * process until exitlwps() is called again without the
   1690      0     stevel 	 * 'coredump' flag set, then the lwps are terminated
   1691      0     stevel 	 * and the process can exit.
   1692      0     stevel 	 */
   1693      0     stevel 	if (coredump) {
   1694      0     stevel 		p->p_flag &= ~(SCOREDUMP | SHOLDFORK | SEXITLWPS);
   1695      0     stevel 		goto out;
   1696      0     stevel 	}
   1697      0     stevel 
   1698      0     stevel 	/*
   1699      0     stevel 	 * Determine if there are any lwps left dangling in
   1700      0     stevel 	 * the stopped state.  This happens when exitlwps()
   1701      0     stevel 	 * aborts a holdlwps() operation.
   1702      0     stevel 	 */
   1703      0     stevel 	p->p_flag &= ~SHOLDFORK;
   1704      0     stevel 	if ((heldcnt = p->p_lwpcnt) > 1) {
   1705      0     stevel 		kthread_t *t;
   1706      0     stevel 		for (t = curthread->t_forw; --heldcnt > 0; t = t->t_forw) {
   1707      0     stevel 			t->t_proc_flag &= ~TP_TWAIT;
   1708      0     stevel 			lwp_continue(t);
   1709      0     stevel 		}
   1710      0     stevel 	}
   1711      0     stevel 
   1712      0     stevel 	/*
   1713      0     stevel 	 * Wait for all other lwps to exit.
   1714      0     stevel 	 */
   1715      0     stevel 	--p->p_lwprcnt;
   1716    390        raf 	while (p->p_lwpcnt > 1) {
   1717      0     stevel 		cv_wait(&p->p_holdlwps, &p->p_lock);
   1718    390        raf 		prbarrier(p);
   1719    390        raf 	}
   1720      0     stevel 	++p->p_lwprcnt;
   1721      0     stevel 	ASSERT(p->p_lwpcnt == 1 && p->p_lwprcnt == 1);
   1722      0     stevel 
   1723      0     stevel 	p->p_flag &= ~SEXITLWPS;
   1724      0     stevel 	curthread->t_proc_flag &= ~TP_TWAIT;
   1725      0     stevel 
   1726      0     stevel out:
   1727      0     stevel 	if (!coredump && p->p_zombcnt) {	/* cleanup the zombie lwps */
   1728      0     stevel 		lwpdir_t *ldp;
   1729      0     stevel 		lwpent_t *lep;
   1730      0     stevel 		int i;
   1731      0     stevel 
   1732      0     stevel 		for (ldp = p->p_lwpdir, i = 0; i < p->p_lwpdir_sz; i++, ldp++) {
   1733      0     stevel 			lep = ldp->ld_entry;
   1734      0     stevel 			if (lep != NULL && lep->le_thread != curthread) {
   1735      0     stevel 				ASSERT(lep->le_thread == NULL);
   1736      0     stevel 				p->p_zombcnt--;
   1737      0     stevel 				lwp_hash_out(p, lep->le_lwpid);
   1738      0     stevel 			}
   1739      0     stevel 		}
   1740      0     stevel 		ASSERT(p->p_zombcnt == 0);
   1741      0     stevel 	}
   1742      0     stevel 
   1743      0     stevel 	/*
   1744      0     stevel 	 * If some other LWP in the process wanted us to suspend ourself,
   1745      0     stevel 	 * then we will not do it.  The other LWP is now terminated and
   1746      0     stevel 	 * no one will ever continue us again if we suspend ourself.
   1747      0     stevel 	 */
   1748      0     stevel 	curthread->t_proc_flag &= ~TP_HOLDLWP;
   1749      0     stevel 	p->p_flag &= ~(SHOLDFORK | SHOLDFORK1 | SHOLDWATCH | SLWPWRAP);
   1750      0     stevel 	mutex_exit(&p->p_lock);
   1751      0     stevel 	return (0);
   1752      0     stevel }
   1753      0     stevel 
   1754      0     stevel /*
   1755      0     stevel  * duplicate a lwp.
   1756      0     stevel  */
   1757      0     stevel klwp_t *
   1758      0     stevel forklwp(klwp_t *lwp, proc_t *cp, id_t lwpid)
   1759      0     stevel {
   1760      0     stevel 	klwp_t *clwp;
   1761      0     stevel 	void *tregs, *tfpu;
   1762      0     stevel 	kthread_t *t = lwptot(lwp);
   1763      0     stevel 	kthread_t *ct;
   1764      0     stevel 	proc_t *p = lwptoproc(lwp);
   1765      0     stevel 	int cid;
   1766      0     stevel 	void *bufp;
   1767   2712    nn35248 	void *brand_data;
   1768      0     stevel 	int val;
   1769      0     stevel 
   1770      0     stevel 	ASSERT(p == curproc);
   1771      0     stevel 	ASSERT(t == curthread || (SUSPENDED(t) && lwp->lwp_asleep == 0));
   1772      0     stevel 
   1773      0     stevel #if defined(__sparc)
   1774      0     stevel 	if (t == curthread)
   1775      0     stevel 		(void) flush_user_windows_to_stack(NULL);
   1776      0     stevel #endif
   1777      0     stevel 
   1778      0     stevel 	if (t == curthread)
   1779      0     stevel 		/* copy args out of registers first */
   1780      0     stevel 		(void) save_syscall_args();
   1781   2712    nn35248 
   1782      0     stevel 	clwp = lwp_create(cp->p_lwpcnt == 0 ? lwp_rtt_initial : lwp_rtt,
   1783      0     stevel 	    NULL, 0, cp, TS_STOPPED, t->t_pri, &t->t_hold, NOCLASS, lwpid);
   1784      0     stevel 	if (clwp == NULL)
   1785      0     stevel 		return (NULL);
   1786      0     stevel 
   1787      0     stevel 	/*
   1788      0     stevel 	 * most of the parent's lwp can be copied to its duplicate,
   1789      0     stevel 	 * except for the fields that are unique to each lwp, like
   1790      0     stevel 	 * lwp_thread, lwp_procp, lwp_regs, and lwp_ap.
   1791      0     stevel 	 */
   1792      0     stevel 	ct = clwp->lwp_thread;
   1793      0     stevel 	tregs = clwp->lwp_regs;
   1794      0     stevel 	tfpu = clwp->lwp_fpu;
   1795   2712    nn35248 	brand_data = clwp->lwp_brand;
   1796      0     stevel 
   1797   3551   johansen 	/*
   1798   3551   johansen 	 * Copy parent lwp to child lwp.  Hold child's p_lock to prevent
   1799   3551   johansen 	 * mstate_aggr_state() from reading stale mstate entries copied
   1800   3551   johansen 	 * from lwp to clwp.
   1801   3551   johansen 	 */
   1802   3551   johansen 	mutex_enter(&cp->p_lock);
   1803      0     stevel 	*clwp = *lwp;
   1804   3551   johansen 
   1805   3551   johansen 	/* clear microstate and resource usage data in new lwp */
   1806   3551   johansen 	init_mstate(ct, LMS_STOPPED);
   1807   3551   johansen 	bzero(&clwp->lwp_ru, sizeof (clwp->lwp_ru));
   1808   3551   johansen 	mutex_exit(&cp->p_lock);
   1809      0     stevel 
   1810      0     stevel 	/* fix up child's lwp */
   1811      0     stevel 
   1812   4503    sudheer 	clwp->lwp_pcb.pcb_flags = 0;
   1813   4503    sudheer #if defined(__sparc)
   1814      0     stevel 	clwp->lwp_pcb.pcb_step = STEP_NONE;
   1815      0     stevel #endif
   1816      0     stevel 	clwp->lwp_cursig = 0;
   1817      0     stevel 	clwp->lwp_extsig = 0;
   1818      0     stevel 	clwp->lwp_curinfo = (struct sigqueue *)0;
   1819      0     stevel 	clwp->lwp_thread = ct;
   1820      0     stevel 	ct->t_sysnum = t->t_sysnum;
   1821      0     stevel 	clwp->lwp_regs = tregs;
   1822      0     stevel 	clwp->lwp_fpu = tfpu;
   1823   2712    nn35248 	clwp->lwp_brand = brand_data;
   1824      0     stevel 	clwp->lwp_ap = clwp->lwp_arg;
   1825      0     stevel 	clwp->lwp_procp = cp;
   1826      0     stevel 	bzero(clwp->lwp_timer, sizeof (clwp->lwp_timer));
   1827      0     stevel 	clwp->lwp_lastfault = 0;
   1828      0     stevel 	clwp->lwp_lastfaddr = 0;
   1829      0     stevel 
   1830      0     stevel 	/* copy parent's struct regs to child. */
   1831      0     stevel 	lwp_forkregs(lwp, clwp);
   1832      0     stevel 
   1833      0     stevel 	/*
   1834   1217        rab 	 * Fork thread context ops, if any.
   1835      0     stevel 	 */
   1836      0     stevel 	if (t->t_ctx)
   1837      0     stevel 		forkctx(t, ct);
   1838      0     stevel 
   1839      0     stevel 	/* fix door state in the child */
   1840      0     stevel 	if (t->t_door)
   1841      0     stevel 		door_fork(t, ct);
   1842      0     stevel 
   1843      0     stevel 	/* copy current contract templates, clear latest contracts */
   1844      0     stevel 	lwp_ctmpl_copy(clwp, lwp);
   1845      0     stevel 
   1846      0     stevel 	mutex_enter(&cp->p_lock);
   1847      0     stevel 	/* lwp_create() set the TP_HOLDLWP flag */
   1848      0     stevel 	if (!(t->t_proc_flag & TP_HOLDLWP))
   1849      0     stevel 		ct->t_proc_flag &= ~TP_HOLDLWP;
   1850      0     stevel 	if (cp->p_flag & SMSACCT)
   1851      0     stevel 		ct->t_proc_flag |= TP_MSACCT;
   1852      0     stevel 	mutex_exit(&cp->p_lock);
   1853   2712    nn35248 
   1854   2712    nn35248 	/* Allow brand to propagate brand-specific state */
   1855   2712    nn35248 	if (PROC_IS_BRANDED(p))
   1856   2712    nn35248 		BROP(p)->b_forklwp(lwp, clwp);
   1857      0     stevel 
   1858      0     stevel retry:
   1859      0     stevel 	cid = t->t_cid;
   1860      0     stevel 
   1861      0     stevel 	val = CL_ALLOC(&bufp, cid, KM_SLEEP);
   1862      0     stevel 	ASSERT(val == 0);
   1863      0     stevel 
   1864      0     stevel 	mutex_enter(&p->p_lock);
   1865      0     stevel 	if (cid != t->t_cid) {
   1866      0     stevel 		/*
   1867      0     stevel 		 * Someone just changed this thread's scheduling class,
   1868      0     stevel 		 * so try pre-allocating the buffer again.  Hopefully we
   1869      0     stevel 		 * don't hit this often.
   1870      0     stevel 		 */
   1871      0     stevel 		mutex_exit(&p->p_lock);
   1872      0     stevel 		CL_FREE(cid, bufp);
   1873      0     stevel 		goto retry;
   1874      0     stevel 	}
   1875      0     stevel 
   1876      0     stevel 	ct->t_unpark = t->t_unpark;
   1877      0     stevel 	ct->t_clfuncs = t->t_clfuncs;
   1878      0     stevel 	CL_FORK(t, ct, bufp);
   1879      0     stevel 	ct->t_cid = t->t_cid;	/* after data allocated so prgetpsinfo works */
   1880      0     stevel 	mutex_exit(&p->p_lock);
   1881      0     stevel 
   1882      0     stevel 	return (clwp);
   1883      0     stevel }
   1884      0     stevel 
   1885      0     stevel /*
   1886      0     stevel  * Add a new lwp entry to the lwp directory and to the lwpid hash table.
   1887      0     stevel  */
   1888      0     stevel void
   1889   9393      Roger lwp_hash_in(proc_t *p, lwpent_t *lep, tidhash_t *tidhash, uint_t tidhash_sz,
   1890   9393      Roger     int do_lock)
   1891      0     stevel {
   1892   9393      Roger 	tidhash_t *thp = &tidhash[TIDHASH(lep->le_lwpid, tidhash_sz)];
   1893      0     stevel 	lwpdir_t **ldpp;
   1894      0     stevel 	lwpdir_t *ldp;
   1895      0     stevel 	kthread_t *t;
   1896      0     stevel 
   1897      0     stevel 	/*
   1898      0     stevel 	 * Allocate a directory element from the free list.
   1899      0     stevel 	 * Code elsewhere guarantees a free slot.
   1900      0     stevel 	 */
   1901      0     stevel 	ldp = p->p_lwpfree;
   1902      0     stevel 	p->p_lwpfree = ldp->ld_next;
   1903      0     stevel 	ASSERT(ldp->ld_entry == NULL);
   1904      0     stevel 	ldp->ld_entry = lep;
   1905      0     stevel 
   1906   9393      Roger 	if (do_lock)
   1907   9393      Roger 		mutex_enter(&thp->th_lock);
   1908   9393      Roger 
   1909      0     stevel 	/*
   1910      0     stevel 	 * Insert it into the lwpid hash table.
   1911      0     stevel 	 */
   1912   9393      Roger 	ldpp = &thp->th_list;
   1913      0     stevel 	ldp->ld_next = *ldpp;
   1914      0     stevel 	*ldpp = ldp;
   1915      0     stevel 
   1916      0     stevel 	/*
   1917      0     stevel 	 * Set the active thread's directory slot entry.
   1918      0     stevel 	 */
   1919      0     stevel 	if ((t = lep->le_thread) != NULL) {
   1920      0     stevel 		ASSERT(lep->le_lwpid == t->t_tid);
   1921      0     stevel 		t->t_dslot = (int)(ldp - p->p_lwpdir);
   1922      0     stevel 	}
   1923   9393      Roger 
   1924   9393      Roger 	if (do_lock)
   1925   9393      Roger 		mutex_exit(&thp->th_lock);
   1926      0     stevel }
   1927      0     stevel 
   1928      0     stevel /*
   1929      0     stevel  * Remove an lwp from the lwpid hash table and free its directory entry.
   1930      0     stevel  * This is done when a detached lwp exits in lwp_exit() or
   1931      0     stevel  * when a non-detached lwp is waited for in lwp_wait() or
   1932      0     stevel  * when a zombie lwp is detached in lwp_detach().
   1933      0     stevel  */
   1934      0     stevel void
   1935      0     stevel lwp_hash_out(proc_t *p, id_t lwpid)
   1936      0     stevel {
   1937   9393      Roger 	tidhash_t *thp = &p->p_tidhash[TIDHASH(lwpid, p->p_tidhash_sz)];
   1938      0     stevel 	lwpdir_t **ldpp;
   1939      0     stevel 	lwpdir_t *ldp;
   1940      0     stevel 	lwpent_t *lep;
   1941      0     stevel 
   1942   9393      Roger 	mutex_enter(&thp->th_lock);
   1943   9393      Roger 	for (ldpp = &thp->th_list;
   1944      0     stevel 	    (ldp = *ldpp) != NULL; ldpp = &ldp->ld_next) {
   1945      0     stevel 		lep = ldp->ld_entry;
   1946      0     stevel 		if (lep->le_lwpid == lwpid) {
   1947      0     stevel 			prlwpfree(p, lep);	/* /proc deals with le_trace */
   1948      0     stevel 			*ldpp = ldp->ld_next;
   1949      0     stevel 			ldp->ld_entry = NULL;
   1950      0     stevel 			ldp->ld_next = p->p_lwpfree;
   1951      0     stevel 			p->p_lwpfree = ldp;
   1952      0     stevel 			kmem_free(lep, sizeof (*lep));
   1953      0     stevel 			break;
   1954      0     stevel 		}
   1955      0     stevel 	}
   1956   9393      Roger 	mutex_exit(&thp->th_lock);
   1957      0     stevel }
   1958      0     stevel 
   1959      0     stevel /*
   1960      0     stevel  * Lookup an lwp in the lwpid hash table by lwpid.
   1961      0     stevel  */
   1962      0     stevel lwpdir_t *
   1963      0     stevel lwp_hash_lookup(proc_t *p, id_t lwpid)
   1964      0     stevel {
   1965   9393      Roger 	tidhash_t *thp;
   1966      0     stevel 	lwpdir_t *ldp;
   1967      0     stevel 
   1968      0     stevel 	/*
   1969      0     stevel 	 * The process may be exiting, after p_tidhash has been set to NULL in
   1970      0     stevel 	 * proc_exit() but before prfee() has been called.  Return failure in
   1971      0     stevel 	 * this case.
   1972      0     stevel 	 */
   1973      0     stevel 	if (p->p_tidhash == NULL)
   1974      0     stevel 		return (NULL);
   1975      0     stevel 
   1976   9393      Roger 	thp = &p->p_tidhash[TIDHASH(lwpid, p->p_tidhash_sz)];
   1977   9393      Roger 	for (ldp = thp->th_list; ldp != NULL; ldp = ldp->ld_next) {
   1978      0     stevel 		if (ldp->ld_entry->le_lwpid == lwpid)
   1979      0     stevel 			return (ldp);
   1980      0     stevel 	}
   1981      0     stevel 
   1982   9393      Roger 	return (NULL);
   1983   9393      Roger }
   1984   9393      Roger 
   1985   9393      Roger /*
   1986   9393      Roger  * Same as lwp_hash_lookup(), but acquire and return
   1987   9393      Roger  * the tid hash table entry lock on success.
   1988   9393      Roger  */
   1989   9393      Roger lwpdir_t *
   1990   9393      Roger lwp_hash_lookup_and_lock(proc_t *p, id_t lwpid, kmutex_t **mpp)
   1991   9393      Roger {
   1992   9393      Roger 	tidhash_t *tidhash;
   1993   9393      Roger 	uint_t tidhash_sz;
   1994   9393      Roger 	tidhash_t *thp;
   1995   9393      Roger 	lwpdir_t *ldp;
   1996   9393      Roger 
   1997   9393      Roger top:
   1998   9393      Roger 	tidhash_sz = p->p_tidhash_sz;
   1999   9393      Roger 	membar_consumer();
   2000   9393      Roger 	if ((tidhash = p->p_tidhash) == NULL)
   2001   9393      Roger 		return (NULL);
   2002   9393      Roger 
   2003   9393      Roger 	thp = &tidhash[TIDHASH(lwpid, tidhash_sz)];
   2004   9393      Roger 	mutex_enter(&thp->th_lock);
   2005   9393      Roger 
   2006   9393      Roger 	/*
   2007   9393      Roger 	 * Since we are not holding p->p_lock, the tid hash table
   2008   9393      Roger 	 * may have changed.  If so, start over.  If not, then
   2009   9393      Roger 	 * it cannot change until after we drop &thp->th_lock;
   2010   9393      Roger 	 */
   2011   9393      Roger 	if (tidhash != p->p_tidhash || tidhash_sz != p->p_tidhash_sz) {
   2012   9393      Roger 		mutex_exit(&thp->th_lock);
   2013   9393      Roger 		goto top;
   2014   9393      Roger 	}
   2015   9393      Roger 
   2016   9393      Roger 	for (ldp = thp->th_list; ldp != NULL; ldp = ldp->ld_next) {
   2017   9393      Roger 		if (ldp->ld_entry->le_lwpid == lwpid) {
   2018   9393      Roger 			*mpp = &thp->th_lock;
   2019   9393      Roger 			return (ldp);
   2020   9393      Roger 		}
   2021   9393      Roger 	}
   2022   9393      Roger 
   2023   9393      Roger 	mutex_exit(&thp->th_lock);
   2024      0     stevel 	return (NULL);
   2025      0     stevel }
   2026      0     stevel 
   2027      0     stevel /*
   2028      0     stevel  * Update the indicated LWP usage statistic for the current LWP.
   2029      0     stevel  */
   2030      0     stevel void
   2031      0     stevel lwp_stat_update(lwp_stat_id_t lwp_stat_id, long inc)
   2032      0     stevel {
   2033      0     stevel 	klwp_t *lwp = ttolwp(curthread);
   2034      0     stevel 
   2035      0     stevel 	if (lwp == NULL)
   2036      0     stevel 		return;
   2037      0     stevel 
   2038      0     stevel 	switch (lwp_stat_id) {
   2039      0     stevel 	case LWP_STAT_INBLK:
   2040      0     stevel 		lwp->lwp_ru.inblock += inc;
   2041      0     stevel 		break;
   2042      0     stevel 	case LWP_STAT_OUBLK:
   2043      0     stevel 		lwp->lwp_ru.oublock += inc;
   2044      0     stevel 		break;
   2045      0     stevel 	case LWP_STAT_MSGRCV:
   2046      0     stevel 		lwp->lwp_ru.msgrcv += inc;
   2047      0     stevel 		break;
   2048      0     stevel 	case LWP_STAT_MSGSND:
   2049      0     stevel 		lwp->lwp_ru.msgsnd += inc;
   2050      0     stevel 		break;
   2051      0     stevel 	default:
   2052      0     stevel 		panic("lwp_stat_update: invalid lwp_stat_id 0x%x", lwp_stat_id);
   2053      0     stevel 	}
   2054      0     stevel }
   2055