subr_witness.c revision 127323
165557Sjasone/*- 265557Sjasone * Copyright (c) 1998 Berkeley Software Design, Inc. All rights reserved. 365557Sjasone * 465557Sjasone * Redistribution and use in source and binary forms, with or without 565557Sjasone * modification, are permitted provided that the following conditions 665557Sjasone * are met: 765557Sjasone * 1. Redistributions of source code must retain the above copyright 865557Sjasone * notice, this list of conditions and the following disclaimer. 965557Sjasone * 2. Redistributions in binary form must reproduce the above copyright 1065557Sjasone * notice, this list of conditions and the following disclaimer in the 1165557Sjasone * documentation and/or other materials provided with the distribution. 1265557Sjasone * 3. Berkeley Software Design Inc's name may not be used to endorse or 1365557Sjasone * promote products derived from this software without specific prior 1465557Sjasone * written permission. 1565557Sjasone * 1665557Sjasone * THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``AS IS'' AND 1765557Sjasone * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1865557Sjasone * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1965557Sjasone * ARE DISCLAIMED. IN NO EVENT SHALL BERKELEY SOFTWARE DESIGN INC BE LIABLE 2065557Sjasone * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2165557Sjasone * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2265557Sjasone * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2365557Sjasone * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2465557Sjasone * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2565557Sjasone * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2665557Sjasone * SUCH DAMAGE. 2765557Sjasone * 2865557Sjasone * from BSDI $Id: mutex_witness.c,v 1.1.2.20 2000/04/27 03:10:27 cp Exp $ 2967352Sjhb * and BSDI $Id: synch_machdep.c,v 2.3.2.39 2000/04/27 03:10:25 cp Exp $ 3065557Sjasone */ 3165557Sjasone 3265557Sjasone/* 3374912Sjhb * Implementation of the `witness' lock verifier. Originally implemented for 3474912Sjhb * mutexes in BSD/OS. Extended to handle generic lock objects and lock 3574912Sjhb * classes in FreeBSD. 3672200Sbmilekic */ 3772200Sbmilekic 3872200Sbmilekic/* 3965557Sjasone * Main Entry: witness 4065557Sjasone * Pronunciation: 'wit-n&s 4165557Sjasone * Function: noun 4265557Sjasone * Etymology: Middle English witnesse, from Old English witnes knowledge, 4365557Sjasone * testimony, witness, from 2wit 4465557Sjasone * Date: before 12th century 4565557Sjasone * 1 : attestation of a fact or event : TESTIMONY 4665557Sjasone * 2 : one that gives evidence; specifically : one who testifies in 4765557Sjasone * a cause or before a judicial tribunal 4865557Sjasone * 3 : one asked to be present at a transaction so as to be able to 4965557Sjasone * testify to its having taken place 5065557Sjasone * 4 : one who has personal knowledge of something 5165557Sjasone * 5 a : something serving as evidence or proof : SIGN 5265557Sjasone * b : public affirmation by word or example of usually 5365557Sjasone * religious faith or conviction <the heroic witness to divine 5465557Sjasone * life -- Pilot> 5565557Sjasone * 6 capitalized : a member of the Jehovah's Witnesses 5665557Sjasone */ 5765557Sjasone 58111881Sjhb/* 59111881Sjhb * Special rules concerning Giant and lock orders: 60111881Sjhb * 61111881Sjhb * 1) Giant must be acquired before any other mutexes. Stated another way, 62111881Sjhb * no other mutex may be held when Giant is acquired. 63111881Sjhb * 64111881Sjhb * 2) Giant must be released when blocking on a sleepable lock. 65111881Sjhb * 66111881Sjhb * This rule is less obvious, but is a result of Giant providing the same 67111881Sjhb * semantics as spl(). Basically, when a thread sleeps, it must release 68111881Sjhb * Giant. When a thread blocks on a sleepable lock, it sleeps. Hence rule 69111881Sjhb * 2). 70111881Sjhb * 71111881Sjhb * 3) Giant may be acquired before or after sleepable locks. 72111881Sjhb * 73111881Sjhb * This rule is also not quite as obvious. Giant may be acquired after 74111881Sjhb * a sleepable lock because it is a non-sleepable lock and non-sleepable 75111881Sjhb * locks may always be acquired while holding a sleepable lock. The second 76111881Sjhb * case, Giant before a sleepable lock, follows from rule 2) above. Suppose 77111881Sjhb * you have two threads T1 and T2 and a sleepable lock X. Suppose that T1 78111881Sjhb * acquires X and blocks on Giant. Then suppose that T2 acquires Giant and 79111881Sjhb * blocks on X. When T2 blocks on X, T2 will release Giant allowing T1 to 80111881Sjhb * execute. Thus, acquiring Giant both before and after a sleepable lock 81111881Sjhb * will not result in a lock order reversal. 82111881Sjhb */ 83111881Sjhb 84116182Sobrien#include <sys/cdefs.h> 85116182Sobrien__FBSDID("$FreeBSD: head/sys/kern/subr_witness.c 127323 2004-03-23 00:32:27Z alfred $"); 86116182Sobrien 8768790Sjhb#include "opt_ddb.h" 8867676Sjhb#include "opt_witness.h" 8967676Sjhb 9065557Sjasone#include <sys/param.h> 9167352Sjhb#include <sys/bus.h> 9267352Sjhb#include <sys/kernel.h> 9374912Sjhb#include <sys/ktr.h> 9474912Sjhb#include <sys/lock.h> 9567352Sjhb#include <sys/malloc.h> 9674912Sjhb#include <sys/mutex.h> 9765557Sjasone#include <sys/proc.h> 9867676Sjhb#include <sys/sysctl.h> 9965557Sjasone#include <sys/systm.h> 10065557Sjasone 10168790Sjhb#include <ddb/ddb.h> 10268790Sjhb 103111881Sjhb#include <machine/stdarg.h> 104111881Sjhb 105105508Sphk/* Define this to check for blessed mutexes */ 106105508Sphk#undef BLESSING 107105508Sphk 10874912Sjhb#define WITNESS_COUNT 200 10974912Sjhb#define WITNESS_CHILDCOUNT (WITNESS_COUNT * 4) 11065557Sjasone/* 11183798Sjhb * XXX: This is somewhat bogus, as we assume here that at most 1024 threads 11274912Sjhb * will hold LOCK_NCHILDREN * 2 locks. We handle failure ok, and we should 11374912Sjhb * probably be safe for the most part, but it's still a SWAG. 11467352Sjhb */ 11574912Sjhb#define LOCK_CHILDCOUNT (MAXCPU + 1024) * 2 11671352Sjasone 11774912Sjhb#define WITNESS_NCHILDREN 6 11871352Sjasone 11974912Sjhbstruct witness_child_list_entry; 12071352Sjasone 12174912Sjhbstruct witness { 12274912Sjhb const char *w_name; 12374912Sjhb struct lock_class *w_class; 12474912Sjhb STAILQ_ENTRY(witness) w_list; /* List of all witnesses. */ 12574912Sjhb STAILQ_ENTRY(witness) w_typelist; /* Witnesses of a type. */ 12674912Sjhb struct witness_child_list_entry *w_children; /* Great evilness... */ 12774912Sjhb const char *w_file; 12874912Sjhb int w_line; 12974912Sjhb u_int w_level; 13074912Sjhb u_int w_refcount; 13174912Sjhb u_char w_Giant_squawked:1; 13274912Sjhb u_char w_other_squawked:1; 13374912Sjhb u_char w_same_squawked:1; 134112118Sjhb u_char w_displayed:1; 13574912Sjhb}; 13671352Sjasone 13774912Sjhbstruct witness_child_list_entry { 13874912Sjhb struct witness_child_list_entry *wcl_next; 13974912Sjhb struct witness *wcl_children[WITNESS_NCHILDREN]; 14074912Sjhb u_int wcl_count; 14174912Sjhb}; 14271352Sjasone 14374912SjhbSTAILQ_HEAD(witness_list, witness); 14471352Sjasone 145105508Sphk#ifdef BLESSING 14674912Sjhbstruct witness_blessed { 14774912Sjhb const char *b_lock1; 14874912Sjhb const char *b_lock2; 14974912Sjhb}; 150105508Sphk#endif 15171352Sjasone 15274912Sjhbstruct witness_order_list_entry { 15374912Sjhb const char *w_name; 15474912Sjhb struct lock_class *w_class; 15574912Sjhb}; 15671352Sjasone 157112117Sjhb#ifdef BLESSING 158112117Sjhbstatic int blessed(struct witness *, struct witness *); 159112117Sjhb#endif 160112117Sjhbstatic int depart(struct witness *w); 16174912Sjhbstatic struct witness *enroll(const char *description, 16274912Sjhb struct lock_class *lock_class); 163112117Sjhbstatic int insertchild(struct witness *parent, struct witness *child); 164112117Sjhbstatic int isitmychild(struct witness *parent, struct witness *child); 165112117Sjhbstatic int isitmydescendant(struct witness *parent, struct witness *child); 16674912Sjhbstatic int itismychild(struct witness *parent, struct witness *child); 167112117Sjhbstatic int rebalancetree(struct witness_list *list); 16874912Sjhbstatic void removechild(struct witness *parent, struct witness *child); 169112117Sjhbstatic int reparentchildren(struct witness *newparent, 170112117Sjhb struct witness *oldparent); 171112562Sjhbstatic int sysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS); 17274912Sjhbstatic void witness_displaydescendants(void(*)(const char *fmt, ...), 173112118Sjhb struct witness *, int indent); 174112116Sjhbstatic const char *fixup_filename(const char *file); 17574912Sjhbstatic void witness_leveldescendents(struct witness *parent, int level); 17674912Sjhbstatic void witness_levelall(void); 17774912Sjhbstatic struct witness *witness_get(void); 17874912Sjhbstatic void witness_free(struct witness *m); 17974912Sjhbstatic struct witness_child_list_entry *witness_child_get(void); 18074912Sjhbstatic void witness_child_free(struct witness_child_list_entry *wcl); 18174912Sjhbstatic struct lock_list_entry *witness_lock_list_get(void); 18274912Sjhbstatic void witness_lock_list_free(struct lock_list_entry *lle); 18376272Sjhbstatic struct lock_instance *find_instance(struct lock_list_entry *lock_list, 18476272Sjhb struct lock_object *lock); 185111881Sjhbstatic void witness_list_lock(struct lock_instance *instance); 186112115Sjhb#ifdef DDB 187112061Sjhbstatic void witness_list(struct thread *td); 188100011Smpstatic void witness_display_list(void(*prnt)(const char *fmt, ...), 189100011Smp struct witness_list *list); 190100011Smpstatic void witness_display(void(*)(const char *fmt, ...)); 191100011Smp#endif 19272200Sbmilekic 19374912SjhbMALLOC_DEFINE(M_WITNESS, "witness", "witness structure"); 19472200Sbmilekic 195112562Sjhb/* 196112562Sjhb * If set to 0, witness is disabled. If set to 1, witness performs full lock 197112562Sjhb * order checking for all locks. If set to 2 or higher, then witness skips 198112562Sjhb * the full lock order check if the lock being acquired is at a higher level 199112562Sjhb * (i.e. farther down in the tree) than the current lock. This last mode is 200112562Sjhb * somewhat experimental and not considered fully safe. At runtime, this 201112562Sjhb * value may be set to 0 to turn off witness. witness is not allowed be 202112562Sjhb * turned on once it is turned off, however. 203112562Sjhb */ 20477843Speterstatic int witness_watch = 1; 20577900SpeterTUNABLE_INT("debug.witness_watch", &witness_watch); 206112562SjhbSYSCTL_PROC(_debug, OID_AUTO, witness_watch, CTLFLAG_RW | CTLTYPE_INT, NULL, 0, 207112562Sjhb sysctl_debug_witness_watch, "I", "witness is watching lock operations"); 20871352Sjasone 20967352Sjhb#ifdef DDB 21072200Sbmilekic/* 21167676Sjhb * When DDB is enabled and witness_ddb is set to 1, it will cause the system to 21265557Sjasone * drop into kdebug() when: 21365557Sjasone * - a lock heirarchy violation occurs 21465557Sjasone * - locks are held when going to sleep. 21565557Sjasone */ 21667676Sjhb#ifdef WITNESS_DDB 21777843Speterint witness_ddb = 1; 21867676Sjhb#else 21977843Speterint witness_ddb = 0; 22065557Sjasone#endif 22177900SpeterTUNABLE_INT("debug.witness_ddb", &witness_ddb); 22267676SjhbSYSCTL_INT(_debug, OID_AUTO, witness_ddb, CTLFLAG_RW, &witness_ddb, 0, ""); 223110779Speter 224110779Speter/* 225110779Speter * When DDB is enabled and witness_trace is set to 1, it will cause the system 226110779Speter * to print a stack trace: 227110779Speter * - a lock heirarchy violation occurs 228110779Speter * - locks are held when going to sleep. 229110779Speter */ 230110779Speterint witness_trace = 1; 231110779SpeterTUNABLE_INT("debug.witness_trace", &witness_trace); 232110779SpeterSYSCTL_INT(_debug, OID_AUTO, witness_trace, CTLFLAG_RW, &witness_trace, 0, ""); 23367676Sjhb#endif /* DDB */ 23465557Sjasone 23567676Sjhb#ifdef WITNESS_SKIPSPIN 23677843Speterint witness_skipspin = 1; 23767676Sjhb#else 23877843Speterint witness_skipspin = 0; 23965557Sjasone#endif 24077900SpeterTUNABLE_INT("debug.witness_skipspin", &witness_skipspin); 241121307SsilbySYSCTL_INT(_debug, OID_AUTO, witness_skipspin, CTLFLAG_RDTUN, &witness_skipspin, 0, 24267676Sjhb ""); 24365557Sjasone 24474912Sjhbstatic struct mtx w_mtx; 24574912Sjhbstatic struct witness_list w_free = STAILQ_HEAD_INITIALIZER(w_free); 24674912Sjhbstatic struct witness_list w_all = STAILQ_HEAD_INITIALIZER(w_all); 24774912Sjhbstatic struct witness_list w_spin = STAILQ_HEAD_INITIALIZER(w_spin); 24874912Sjhbstatic struct witness_list w_sleep = STAILQ_HEAD_INITIALIZER(w_sleep); 24974912Sjhbstatic struct witness_child_list_entry *w_child_free = NULL; 25074912Sjhbstatic struct lock_list_entry *w_lock_list_free = NULL; 25165557Sjasone 25274912Sjhbstatic struct witness w_data[WITNESS_COUNT]; 25374912Sjhbstatic struct witness_child_list_entry w_childdata[WITNESS_CHILDCOUNT]; 25474912Sjhbstatic struct lock_list_entry w_locklistdata[LOCK_CHILDCOUNT]; 25565557Sjasone 25674912Sjhbstatic struct witness_order_list_entry order_lists[] = { 25774912Sjhb { "proctree", &lock_class_sx }, 25874912Sjhb { "allproc", &lock_class_sx }, 259111951Sjhb { "Giant", &lock_class_mtx_sleep }, 260108184Skris { "filedesc structure", &lock_class_mtx_sleep }, 261108184Skris { "pipe mutex", &lock_class_mtx_sleep }, 26296122Salfred { "sigio lock", &lock_class_mtx_sleep }, 26391140Stanimura { "process group", &lock_class_mtx_sleep }, 26474912Sjhb { "process lock", &lock_class_mtx_sleep }, 26591140Stanimura { "session", &lock_class_mtx_sleep }, 26674912Sjhb { "uidinfo hash", &lock_class_mtx_sleep }, 26774912Sjhb { "uidinfo struct", &lock_class_mtx_sleep }, 268113275Smike { "allprison", &lock_class_mtx_sleep }, 26974912Sjhb { NULL, NULL }, 27075464Sjhb /* 27175464Sjhb * spin locks 27275464Sjhb */ 27384331Sjhb#ifdef SMP 27484331Sjhb { "ap boot", &lock_class_mtx_spin }, 27572224Sjhb#endif 27674912Sjhb { "sio", &lock_class_mtx_spin }, 27772224Sjhb#ifdef __i386__ 27874912Sjhb { "cy", &lock_class_mtx_spin }, 27972224Sjhb#endif 280124972Sru { "uart_hwmtx", &lock_class_mtx_spin }, 281103091Sjake { "sabtty", &lock_class_mtx_spin }, 282109015Sjake { "zstty", &lock_class_mtx_spin }, 28374912Sjhb { "ng_node", &lock_class_mtx_spin }, 28474912Sjhb { "ng_worklist", &lock_class_mtx_spin }, 285119813Ssam { "taskqueue_fast", &lock_class_mtx_spin }, 286122001Sjhb { "intr table", &lock_class_mtx_spin }, 28774912Sjhb { "ithread table lock", &lock_class_mtx_spin }, 288126324Sjhb { "sleepq chain", &lock_class_mtx_spin }, 28974912Sjhb { "sched lock", &lock_class_mtx_spin }, 290122514Sjhb { "turnstile chain", &lock_class_mtx_spin }, 291122514Sjhb { "td_contested", &lock_class_mtx_spin }, 29274912Sjhb { "callout", &lock_class_mtx_spin }, 293122917Smarkm { "entropy harvest", &lock_class_mtx_spin }, 294122917Smarkm { "entropy harvest buffers", &lock_class_mtx_spin }, 29565557Sjasone /* 29665557Sjasone * leaf locks 29765557Sjasone */ 29890278Sjhb { "allpmaps", &lock_class_mtx_spin }, 29999416Salc { "vm page queue free mutex", &lock_class_mtx_spin }, 30088322Sjhb { "icu", &lock_class_mtx_spin }, 30172224Sjhb#ifdef SMP 30274912Sjhb { "smp rendezvous", &lock_class_mtx_spin }, 303122849Speter#if defined(__i386__) || defined(__amd64__) 30499862Speter { "tlb", &lock_class_mtx_spin }, 305112993Speter { "lazypmap", &lock_class_mtx_spin }, 306112993Speter#endif 307108187Sjake#ifdef __sparc64__ 308108187Sjake { "ipi", &lock_class_mtx_spin }, 30999862Speter#endif 310108187Sjake#endif 31178785Sjhb { "clk", &lock_class_mtx_spin }, 31295473Sdes { "mutex profiling lock", &lock_class_mtx_spin }, 313111028Sjeff { "kse zombie lock", &lock_class_mtx_spin }, 314103786Sjeff { "ALD Queue", &lock_class_mtx_spin }, 315104951Speter#ifdef __ia64__ 316104951Speter { "MCA spin lock", &lock_class_mtx_spin }, 317104951Speter#endif 318115425Speter#if defined(__i386__) || defined(__amd64__) 319111068Speter { "pcicfg", &lock_class_mtx_spin }, 320111068Speter#endif 32174912Sjhb { NULL, NULL }, 32274912Sjhb { NULL, NULL } 32365557Sjasone}; 32465557Sjasone 325105508Sphk#ifdef BLESSING 32665557Sjasone/* 32765557Sjasone * Pairs of locks which have been blessed 32865557Sjasone * Don't complain about order problems with blessed locks 32965557Sjasone */ 33065856Sjhbstatic struct witness_blessed blessed_list[] = { 33165557Sjasone}; 33272200Sbmilekicstatic int blessed_count = 33372200Sbmilekic sizeof(blessed_list) / sizeof(struct witness_blessed); 334105508Sphk#endif 33565557Sjasone 33674912Sjhb/* 33774912Sjhb * List of all locks in the system. 33874912Sjhb */ 33997963SjhbTAILQ_HEAD(, lock_object) all_locks = TAILQ_HEAD_INITIALIZER(all_locks); 34074912Sjhb 34174912Sjhbstatic struct mtx all_mtx = { 34274912Sjhb { &lock_class_mtx_sleep, /* mtx_object.lo_class */ 34374912Sjhb "All locks list", /* mtx_object.lo_name */ 34493811Sjhb "All locks list", /* mtx_object.lo_type */ 34574912Sjhb LO_INITIALIZED, /* mtx_object.lo_flags */ 34697963Sjhb { NULL, NULL }, /* mtx_object.lo_list */ 34774912Sjhb NULL }, /* mtx_object.lo_witness */ 348122514Sjhb MTX_UNOWNED, 0 /* mtx_lock, mtx_recurse */ 34974912Sjhb}; 35074912Sjhb 35174912Sjhb/* 35274912Sjhb * This global is set to 0 once it becomes safe to use the witness code. 35374912Sjhb */ 35474912Sjhbstatic int witness_cold = 1; 35574912Sjhb 35674912Sjhb/* 35774912Sjhb * Global variables for book keeping. 35874912Sjhb */ 35974912Sjhbstatic int lock_cur_cnt; 36074912Sjhbstatic int lock_max_cnt; 36174912Sjhb 36274912Sjhb/* 36374912Sjhb * The WITNESS-enabled diagnostic code. 36474912Sjhb */ 36571352Sjasonestatic void 36674912Sjhbwitness_initialize(void *dummy __unused) 36765557Sjasone{ 36874912Sjhb struct lock_object *lock; 36974912Sjhb struct witness_order_list_entry *order; 37074912Sjhb struct witness *w, *w1; 37174912Sjhb int i; 37265557Sjasone 37374912Sjhb /* 37474912Sjhb * We have to release Giant before initializing its witness 37574912Sjhb * structure so that WITNESS doesn't get confused. 37674912Sjhb */ 37774912Sjhb mtx_unlock(&Giant); 37874912Sjhb mtx_assert(&Giant, MA_NOTOWNED); 37974912Sjhb 38087593Sobrien CTR1(KTR_WITNESS, "%s: initializing witness", __func__); 38197963Sjhb TAILQ_INSERT_HEAD(&all_locks, &all_mtx.mtx_object, lo_list); 38293811Sjhb mtx_init(&w_mtx, "witness lock", NULL, MTX_SPIN | MTX_QUIET | 38393811Sjhb MTX_NOWITNESS); 38474912Sjhb for (i = 0; i < WITNESS_COUNT; i++) 38574912Sjhb witness_free(&w_data[i]); 38674912Sjhb for (i = 0; i < WITNESS_CHILDCOUNT; i++) 38774912Sjhb witness_child_free(&w_childdata[i]); 38874912Sjhb for (i = 0; i < LOCK_CHILDCOUNT; i++) 38974912Sjhb witness_lock_list_free(&w_locklistdata[i]); 39074912Sjhb 39174912Sjhb /* First add in all the specified order lists. */ 39274912Sjhb for (order = order_lists; order->w_name != NULL; order++) { 39374912Sjhb w = enroll(order->w_name, order->w_class); 39475569Sjhb if (w == NULL) 39575569Sjhb continue; 39674912Sjhb w->w_file = "order list"; 39774912Sjhb for (order++; order->w_name != NULL; order++) { 39874912Sjhb w1 = enroll(order->w_name, order->w_class); 39975569Sjhb if (w1 == NULL) 40075569Sjhb continue; 40174912Sjhb w1->w_file = "order list"; 402112117Sjhb if (!itismychild(w, w1)) 403112117Sjhb panic("Not enough memory for static orders!"); 40474912Sjhb w = w1; 40565557Sjasone } 40665557Sjasone } 40765557Sjasone 40874912Sjhb /* Iterate through all locks and add them to witness. */ 40974912Sjhb mtx_lock(&all_mtx); 41097963Sjhb TAILQ_FOREACH(lock, &all_locks, lo_list) { 41174912Sjhb if (lock->lo_flags & LO_WITNESS) 41293811Sjhb lock->lo_witness = enroll(lock->lo_type, 41374912Sjhb lock->lo_class); 41474912Sjhb else 41574912Sjhb lock->lo_witness = NULL; 41674912Sjhb } 41774912Sjhb mtx_unlock(&all_mtx); 41874912Sjhb 41974912Sjhb /* Mark the witness code as being ready for use. */ 42074912Sjhb atomic_store_rel_int(&witness_cold, 0); 42174912Sjhb 42274912Sjhb mtx_lock(&Giant); 42365557Sjasone} 42474912SjhbSYSINIT(witness_init, SI_SUB_WITNESS, SI_ORDER_FIRST, witness_initialize, NULL) 42565557Sjasone 426112562Sjhbstatic int 427112562Sjhbsysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS) 428112562Sjhb{ 429112562Sjhb int error, value; 430112562Sjhb 431112562Sjhb value = witness_watch; 432112562Sjhb error = sysctl_handle_int(oidp, &value, 0, req); 433112562Sjhb if (error != 0 || req->newptr == NULL) 434112562Sjhb return (error); 435112562Sjhb error = suser(req->td); 436112562Sjhb if (error != 0) 437112562Sjhb return (error); 438112562Sjhb if (value == witness_watch) 439112562Sjhb return (0); 440112562Sjhb if (value != 0) 441112562Sjhb return (EINVAL); 442112562Sjhb witness_watch = 0; 443112562Sjhb return (0); 444112562Sjhb} 445112562Sjhb 44674912Sjhbvoid 44774912Sjhbwitness_init(struct lock_object *lock) 44874912Sjhb{ 44974912Sjhb struct lock_class *class; 45074912Sjhb 45174912Sjhb class = lock->lo_class; 45274912Sjhb if (lock->lo_flags & LO_INITIALIZED) 45382284Sjhb panic("%s: lock (%s) %s is already initialized", __func__, 45474912Sjhb class->lc_name, lock->lo_name); 45574912Sjhb if ((lock->lo_flags & LO_RECURSABLE) != 0 && 45674912Sjhb (class->lc_flags & LC_RECURSABLE) == 0) 45782284Sjhb panic("%s: lock (%s) %s can not be recursable", __func__, 45874912Sjhb class->lc_name, lock->lo_name); 45974912Sjhb if ((lock->lo_flags & LO_SLEEPABLE) != 0 && 46074912Sjhb (class->lc_flags & LC_SLEEPABLE) == 0) 46182284Sjhb panic("%s: lock (%s) %s can not be sleepable", __func__, 46274912Sjhb class->lc_name, lock->lo_name); 46382244Sjhb if ((lock->lo_flags & LO_UPGRADABLE) != 0 && 46482244Sjhb (class->lc_flags & LC_UPGRADABLE) == 0) 46582284Sjhb panic("%s: lock (%s) %s can not be upgradable", __func__, 46682244Sjhb class->lc_name, lock->lo_name); 46782244Sjhb 46874912Sjhb mtx_lock(&all_mtx); 46997963Sjhb TAILQ_INSERT_TAIL(&all_locks, lock, lo_list); 47074912Sjhb lock->lo_flags |= LO_INITIALIZED; 47174912Sjhb lock_cur_cnt++; 47274912Sjhb if (lock_cur_cnt > lock_max_cnt) 47374912Sjhb lock_max_cnt = lock_cur_cnt; 47474912Sjhb mtx_unlock(&all_mtx); 475112562Sjhb if (!witness_cold && witness_watch != 0 && panicstr == NULL && 47674912Sjhb (lock->lo_flags & LO_WITNESS) != 0) 47793811Sjhb lock->lo_witness = enroll(lock->lo_type, class); 47874912Sjhb else 47974912Sjhb lock->lo_witness = NULL; 48074912Sjhb} 48174912Sjhb 48274912Sjhbvoid 48374912Sjhbwitness_destroy(struct lock_object *lock) 48474912Sjhb{ 48575362Sjhb struct witness *w; 48674912Sjhb 48774912Sjhb if (witness_cold) 48874912Sjhb panic("lock (%s) %s destroyed while witness_cold", 48974912Sjhb lock->lo_class->lc_name, lock->lo_name); 49074912Sjhb if ((lock->lo_flags & LO_INITIALIZED) == 0) 49182284Sjhb panic("%s: lock (%s) %s is not initialized", __func__, 49274912Sjhb lock->lo_class->lc_name, lock->lo_name); 49374912Sjhb 49476272Sjhb /* XXX: need to verify that no one holds the lock */ 49575362Sjhb w = lock->lo_witness; 49675362Sjhb if (w != NULL) { 49775362Sjhb mtx_lock_spin(&w_mtx); 49897948Sjhb MPASS(w->w_refcount > 0); 49975362Sjhb w->w_refcount--; 500112117Sjhb 501112117Sjhb /* 502112117Sjhb * Lock is already released if we have an allocation failure 503112117Sjhb * and depart() fails. 504112117Sjhb */ 505112117Sjhb if (w->w_refcount != 0 || depart(w)) 506112117Sjhb mtx_unlock_spin(&w_mtx); 50775362Sjhb } 50875362Sjhb 50974912Sjhb mtx_lock(&all_mtx); 51074912Sjhb lock_cur_cnt--; 51197963Sjhb TAILQ_REMOVE(&all_locks, lock, lo_list); 51280055Sjhb lock->lo_flags &= ~LO_INITIALIZED; 51374912Sjhb mtx_unlock(&all_mtx); 51474912Sjhb} 51574912Sjhb 516112115Sjhb#ifdef DDB 51771352Sjasonestatic void 51874912Sjhbwitness_display_list(void(*prnt)(const char *fmt, ...), 51974912Sjhb struct witness_list *list) 52071352Sjasone{ 521112118Sjhb struct witness *w; 52271352Sjasone 52374912Sjhb STAILQ_FOREACH(w, list, w_typelist) { 524112118Sjhb if (w->w_file == NULL || w->w_level > 0) 52571352Sjasone continue; 52671352Sjasone /* 52771352Sjasone * This lock has no anscestors, display its descendants. 52871352Sjasone */ 529112118Sjhb witness_displaydescendants(prnt, w, 0); 53071352Sjasone } 53174912Sjhb} 53272224Sjhb 53374912Sjhbstatic void 53474912Sjhbwitness_display(void(*prnt)(const char *fmt, ...)) 53574912Sjhb{ 53674912Sjhb struct witness *w; 53774912Sjhb 53882284Sjhb KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 53974912Sjhb witness_levelall(); 54074912Sjhb 541112118Sjhb /* Clear all the displayed flags. */ 542112118Sjhb STAILQ_FOREACH(w, &w_all, w_list) { 543112118Sjhb w->w_displayed = 0; 544112118Sjhb } 545112118Sjhb 54672224Sjhb /* 54774930Sjhb * First, handle sleep locks which have been acquired at least 54874912Sjhb * once. 54974912Sjhb */ 55074912Sjhb prnt("Sleep locks:\n"); 55174912Sjhb witness_display_list(prnt, &w_sleep); 55274912Sjhb 55374912Sjhb /* 55474930Sjhb * Now do spin locks which have been acquired at least once. 55572224Sjhb */ 55674912Sjhb prnt("\nSpin locks:\n"); 55774912Sjhb witness_display_list(prnt, &w_spin); 55872224Sjhb 55972224Sjhb /* 56074930Sjhb * Finally, any locks which have not been acquired yet. 56172224Sjhb */ 56274912Sjhb prnt("\nLocks which were never acquired:\n"); 56374912Sjhb STAILQ_FOREACH(w, &w_all, w_list) { 56497948Sjhb if (w->w_file != NULL || w->w_refcount == 0) 56571352Sjasone continue; 56674912Sjhb prnt("%s\n", w->w_name); 56771352Sjasone } 56871352Sjasone} 569112115Sjhb#endif /* DDB */ 57071352Sjasone 571112116Sjhb/* Trim useless garbage from filenames. */ 572112116Sjhbstatic const char * 573112116Sjhbfixup_filename(const char *file) 574112116Sjhb{ 575112116Sjhb 576112116Sjhb if (file == NULL) 577112116Sjhb return (NULL); 578112116Sjhb while (strncmp(file, "../", 3) == 0) 579112116Sjhb file += 3; 580112116Sjhb return (file); 581112116Sjhb} 582112116Sjhb 583125160Sjhbint 584125160Sjhbwitness_defineorder(struct lock_object *lock1, struct lock_object *lock2) 585125160Sjhb{ 586125160Sjhb 587125160Sjhb if (witness_watch == 0 || panicstr != NULL) 588125160Sjhb return (0); 589125160Sjhb 590125160Sjhb /* Require locks that witness knows about. */ 591125160Sjhb if (lock1 == NULL || lock1->lo_witness == NULL || lock2 == NULL || 592125160Sjhb lock2->lo_witness == NULL) 593125160Sjhb return (EINVAL); 594125160Sjhb 595125160Sjhb MPASS(!mtx_owned(&w_mtx)); 596125160Sjhb mtx_lock_spin(&w_mtx); 597125160Sjhb 598125160Sjhb /* 599125160Sjhb * If we already have either an explicit or implied lock order that 600125160Sjhb * is the other way around, then return an error. 601125160Sjhb */ 602125160Sjhb if (isitmydescendant(lock2->lo_witness, lock1->lo_witness)) { 603125160Sjhb mtx_unlock_spin(&w_mtx); 604125160Sjhb return (EDOOFUS); 605125160Sjhb } 606125160Sjhb 607125160Sjhb /* Try to add the new order. */ 608125160Sjhb CTR3(KTR_WITNESS, "%s: adding %s as a child of %s", __func__, 609125160Sjhb lock2->lo_type, lock1->lo_type); 610125160Sjhb if (!itismychild(lock1->lo_witness, lock2->lo_witness)) 611125160Sjhb return (ENOMEM); 612125160Sjhb mtx_unlock_spin(&w_mtx); 613125160Sjhb return (0); 614125160Sjhb} 615125160Sjhb 61665557Sjasonevoid 617125160Sjhbwitness_checkorder(struct lock_object *lock, int flags, const char *file, 618125160Sjhb int line) 61965557Sjasone{ 62074912Sjhb struct lock_list_entry **lock_list, *lle; 62176272Sjhb struct lock_instance *lock1, *lock2; 62274912Sjhb struct lock_class *class; 62365856Sjhb struct witness *w, *w1; 62483366Sjulian struct thread *td; 62574912Sjhb int i, j; 62665557Sjasone 627112562Sjhb if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL || 62880747Sjhb panicstr != NULL) 62971320Sjasone return; 630125160Sjhb 631125160Sjhb /* 632125160Sjhb * Try locks do not block if they fail to acquire the lock, thus 633125160Sjhb * there is no danger of deadlocks or of switching while holding a 634125160Sjhb * spin lock if we acquire a lock via a try operation. This 635125160Sjhb * function shouldn't even be called for try locks, so panic if 636125160Sjhb * that happens. 637125160Sjhb */ 638125160Sjhb if (flags & LOP_TRYLOCK) 639125160Sjhb panic("%s should not be called for try lock operations", 640125160Sjhb __func__); 641125160Sjhb 64274912Sjhb w = lock->lo_witness; 64374912Sjhb class = lock->lo_class; 64483366Sjulian td = curthread; 645112116Sjhb file = fixup_filename(file); 64665557Sjasone 64774912Sjhb if (class->lc_flags & LC_SLEEPLOCK) { 64893676Sjhb /* 64993676Sjhb * Since spin locks include a critical section, this check 65093676Sjhb * impliclty enforces a lock order of all sleep locks before 65193676Sjhb * all spin locks. 65293676Sjhb */ 653125160Sjhb if (td->td_critnest != 0) 65474912Sjhb panic("blockable sleep lock (%s) %s @ %s:%d", 65574912Sjhb class->lc_name, lock->lo_name, file, line); 65683366Sjulian lock_list = &td->td_sleeplocks; 65788899Sjhb } else 65888899Sjhb lock_list = PCPU_PTR(spinlocks); 65965557Sjasone 66076772Sjhb /* 66174912Sjhb * Is this the first lock acquired? If so, then no order checking 66274912Sjhb * is needed. 66365557Sjasone */ 66474912Sjhb if (*lock_list == NULL) 665125160Sjhb return; 66665557Sjasone 66774912Sjhb /* 668125160Sjhb * Check to see if we are recursing on a lock we already own. If 669125160Sjhb * so, make sure that we don't mismatch exclusive and shared lock 670125160Sjhb * acquires. 67176272Sjhb */ 67276272Sjhb lock1 = find_instance(*lock_list, lock); 67376272Sjhb if (lock1 != NULL) { 67476272Sjhb if ((lock1->li_flags & LI_EXCLUSIVE) != 0 && 67576272Sjhb (flags & LOP_EXCLUSIVE) == 0) { 67676272Sjhb printf("shared lock of (%s) %s @ %s:%d\n", 67776272Sjhb class->lc_name, lock->lo_name, file, line); 67876272Sjhb printf("while exclusively locked from %s:%d\n", 67976272Sjhb lock1->li_file, lock1->li_line); 68076272Sjhb panic("share->excl"); 68176272Sjhb } 68276272Sjhb if ((lock1->li_flags & LI_EXCLUSIVE) == 0 && 68376272Sjhb (flags & LOP_EXCLUSIVE) != 0) { 68476272Sjhb printf("exclusive lock of (%s) %s @ %s:%d\n", 68576272Sjhb class->lc_name, lock->lo_name, file, line); 68676272Sjhb printf("while share locked from %s:%d\n", 68776272Sjhb lock1->li_file, lock1->li_line); 68876272Sjhb panic("excl->share"); 68976272Sjhb } 69076272Sjhb return; 69176272Sjhb } 69276272Sjhb 69376272Sjhb /* 694112112Sjhb * Try locks do not block if they fail to acquire the lock, thus 695112112Sjhb * there is no danger of deadlocks or of switching while holding a 696112112Sjhb * spin lock if we acquire a lock via a try operation. 697112112Sjhb */ 698112112Sjhb if (flags & LOP_TRYLOCK) 699125160Sjhb return; 700112112Sjhb 701112112Sjhb /* 70274912Sjhb * Check for duplicate locks of the same type. Note that we only 70374912Sjhb * have to check for this on the last lock we just acquired. Any 70474912Sjhb * other cases will be caught as lock order violations. 70574912Sjhb */ 70676272Sjhb lock1 = &(*lock_list)->ll_children[(*lock_list)->ll_count - 1]; 70776272Sjhb w1 = lock1->li_lock->lo_witness; 70874912Sjhb if (w1 == w) { 70993273Sjeff if (w->w_same_squawked || (lock->lo_flags & LO_DUPOK)) 710125160Sjhb return; 71165557Sjasone w->w_same_squawked = 1; 71275755Sjhb printf("acquiring duplicate lock of same type: \"%s\"\n", 71393811Sjhb lock->lo_type); 71493811Sjhb printf(" 1st %s @ %s:%d\n", lock1->li_lock->lo_name, 71593811Sjhb lock1->li_file, lock1->li_line); 71693811Sjhb printf(" 2nd %s @ %s:%d\n", lock->lo_name, file, line); 71767676Sjhb#ifdef DDB 718125160Sjhb goto debugger; 719125160Sjhb#else 720125160Sjhb return; 721112115Sjhb#endif 72265557Sjasone } 72365557Sjasone MPASS(!mtx_owned(&w_mtx)); 72474912Sjhb mtx_lock_spin(&w_mtx); 72565557Sjasone /* 72665557Sjasone * If we have a known higher number just say ok 72765557Sjasone */ 72865557Sjasone if (witness_watch > 1 && w->w_level > w1->w_level) { 72974912Sjhb mtx_unlock_spin(&w_mtx); 730125160Sjhb return; 73165557Sjasone } 732111881Sjhb /* 733111881Sjhb * If we know that the the lock we are acquiring comes after 734111881Sjhb * the lock we most recently acquired in the lock order tree, 735111881Sjhb * then there is no need for any further checks. 736111881Sjhb */ 73774912Sjhb if (isitmydescendant(w1, w)) { 73874912Sjhb mtx_unlock_spin(&w_mtx); 739125160Sjhb return; 74065557Sjasone } 74174912Sjhb for (j = 0, lle = *lock_list; lle != NULL; lle = lle->ll_next) { 74274912Sjhb for (i = lle->ll_count - 1; i >= 0; i--, j++) { 74365557Sjasone 74474912Sjhb MPASS(j < WITNESS_COUNT); 74576272Sjhb lock1 = &lle->ll_children[i]; 74676272Sjhb w1 = lock1->li_lock->lo_witness; 74774912Sjhb 74874912Sjhb /* 74974912Sjhb * If this lock doesn't undergo witness checking, 75074912Sjhb * then skip it. 75174912Sjhb */ 75274912Sjhb if (w1 == NULL) { 75376272Sjhb KASSERT((lock1->li_lock->lo_flags & LO_WITNESS) == 0, 75474912Sjhb ("lock missing witness structure")); 75574912Sjhb continue; 75674912Sjhb } 75776272Sjhb /* 758111881Sjhb * If we are locking Giant and this is a sleepable 75976272Sjhb * lock, then skip it. 76076272Sjhb */ 761111881Sjhb if ((lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0 && 76276272Sjhb lock == &Giant.mtx_object) 76376272Sjhb continue; 76493690Sjhb /* 76593690Sjhb * If we are locking a sleepable lock and this lock 766111881Sjhb * is Giant, then skip it. 76793690Sjhb */ 768111881Sjhb if ((lock->lo_flags & LO_SLEEPABLE) != 0 && 769111887Sjhb lock1->li_lock == &Giant.mtx_object) 770111881Sjhb continue; 771111881Sjhb /* 772111881Sjhb * If we are locking a sleepable lock and this lock 773111881Sjhb * isn't sleepable, we want to treat it as a lock 774111881Sjhb * order violation to enfore a general lock order of 775111881Sjhb * sleepable locks before non-sleepable locks. 776111881Sjhb */ 77793690Sjhb if (!((lock->lo_flags & LO_SLEEPABLE) != 0 && 778111881Sjhb (lock1->li_lock->lo_flags & LO_SLEEPABLE) == 0)) 779111881Sjhb /* 780111881Sjhb * Check the lock order hierarchy for a reveresal. 781111881Sjhb */ 782111881Sjhb if (!isitmydescendant(w, w1)) 78374912Sjhb continue; 78474912Sjhb /* 78574912Sjhb * We have a lock order violation, check to see if it 78674912Sjhb * is allowed or has already been yelled about. 78774912Sjhb */ 78874912Sjhb mtx_unlock_spin(&w_mtx); 789105508Sphk#ifdef BLESSING 790125160Sjhb /* 791125160Sjhb * If the lock order is blessed, just bail. We don't 792125160Sjhb * look for other lock order violations though, which 793125160Sjhb * may be a bug. 794125160Sjhb */ 79565557Sjasone if (blessed(w, w1)) 796125160Sjhb return; 797105508Sphk#endif 79876272Sjhb if (lock1->li_lock == &Giant.mtx_object) { 79965557Sjasone if (w1->w_Giant_squawked) 800125160Sjhb return; 80165557Sjasone else 80265557Sjasone w1->w_Giant_squawked = 1; 80365557Sjasone } else { 80465557Sjasone if (w1->w_other_squawked) 805125160Sjhb return; 80665557Sjasone else 80765557Sjasone w1->w_other_squawked = 1; 80865557Sjasone } 80974912Sjhb /* 81074912Sjhb * Ok, yell about it. 81174912Sjhb */ 81265557Sjasone printf("lock order reversal\n"); 81374912Sjhb /* 81474912Sjhb * Try to locate an earlier lock with 81574912Sjhb * witness w in our list. 81674912Sjhb */ 81774912Sjhb do { 81876272Sjhb lock2 = &lle->ll_children[i]; 81976272Sjhb MPASS(lock2->li_lock != NULL); 82076272Sjhb if (lock2->li_lock->lo_witness == w) 82174912Sjhb break; 82274912Sjhb if (i == 0 && lle->ll_next != NULL) { 82374912Sjhb lle = lle->ll_next; 82474912Sjhb i = lle->ll_count - 1; 825106781Sjhb MPASS(i >= 0 && i < LOCK_NCHILDREN); 826125160Sjhb } else 827125160Sjhb i--; 82874912Sjhb } while (i >= 0); 82976272Sjhb if (i < 0) { 83093811Sjhb printf(" 1st %p %s (%s) @ %s:%d\n", 83193811Sjhb lock1->li_lock, lock1->li_lock->lo_name, 83293811Sjhb lock1->li_lock->lo_type, lock1->li_file, 83376272Sjhb lock1->li_line); 83493811Sjhb printf(" 2nd %p %s (%s) @ %s:%d\n", lock, 83593811Sjhb lock->lo_name, lock->lo_type, file, line); 83676272Sjhb } else { 83793811Sjhb printf(" 1st %p %s (%s) @ %s:%d\n", 83893811Sjhb lock2->li_lock, lock2->li_lock->lo_name, 83993811Sjhb lock2->li_lock->lo_type, lock2->li_file, 84076272Sjhb lock2->li_line); 84193811Sjhb printf(" 2nd %p %s (%s) @ %s:%d\n", 84293811Sjhb lock1->li_lock, lock1->li_lock->lo_name, 84393811Sjhb lock1->li_lock->lo_type, lock1->li_file, 84476272Sjhb lock1->li_line); 84593811Sjhb printf(" 3rd %p %s (%s) @ %s:%d\n", lock, 84693811Sjhb lock->lo_name, lock->lo_type, file, line); 84776272Sjhb } 84867676Sjhb#ifdef DDB 849125160Sjhb goto debugger; 850125160Sjhb#else 851125160Sjhb return; 852112115Sjhb#endif 85365557Sjasone } 85465557Sjasone } 85576272Sjhb lock1 = &(*lock_list)->ll_children[(*lock_list)->ll_count - 1]; 85678871Sjhb /* 857125160Sjhb * If requested, build a new lock order. However, don't build a new 858125160Sjhb * relationship between a sleepable lock and Giant if it is in the 859125160Sjhb * wrong direction. The correct lock order is that sleepable locks 860125160Sjhb * always come before Giant. 86178871Sjhb */ 862125160Sjhb if (flags & LOP_NEWORDER && 863125160Sjhb !(lock1->li_lock == &Giant.mtx_object && 864112117Sjhb (lock->lo_flags & LO_SLEEPABLE) != 0)) { 86587593Sobrien CTR3(KTR_WITNESS, "%s: adding %s as a child of %s", __func__, 86693811Sjhb lock->lo_type, lock1->li_lock->lo_type); 86778871Sjhb if (!itismychild(lock1->li_lock->lo_witness, w)) 868112117Sjhb /* Witness is dead. */ 869112117Sjhb return; 87078871Sjhb } 871112117Sjhb mtx_unlock_spin(&w_mtx); 872125160Sjhb return; 87365557Sjasone 87467676Sjhb#ifdef DDB 875125160Sjhbdebugger: 876125160Sjhb if (witness_trace) 877125160Sjhb backtrace(); 878125160Sjhb if (witness_ddb) 879125160Sjhb Debugger(__func__); 880125160Sjhb#endif 881125160Sjhb} 882125160Sjhb 883125160Sjhbvoid 884125160Sjhbwitness_lock(struct lock_object *lock, int flags, const char *file, int line) 885125160Sjhb{ 886125160Sjhb struct lock_list_entry **lock_list, *lle; 887125160Sjhb struct lock_instance *instance; 888125160Sjhb struct witness *w; 889125160Sjhb struct thread *td; 890125160Sjhb 891125160Sjhb if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL || 892125160Sjhb panicstr != NULL) 893125160Sjhb return; 894125160Sjhb w = lock->lo_witness; 895125160Sjhb td = curthread; 896125160Sjhb file = fixup_filename(file); 897125160Sjhb 898125160Sjhb /* Determine lock list for this lock. */ 899125160Sjhb if (lock->lo_class->lc_flags & LC_SLEEPLOCK) 900125160Sjhb lock_list = &td->td_sleeplocks; 901125160Sjhb else 902125160Sjhb lock_list = PCPU_PTR(spinlocks); 903125160Sjhb 904125160Sjhb /* Check to see if we are recursing on a lock we already own. */ 905125160Sjhb instance = find_instance(*lock_list, lock); 906125160Sjhb if (instance != NULL) { 907125160Sjhb instance->li_flags++; 908125160Sjhb CTR4(KTR_WITNESS, "%s: pid %d recursed on %s r=%d", __func__, 909125160Sjhb td->td_proc->p_pid, lock->lo_name, 910125160Sjhb instance->li_flags & LI_RECURSEMASK); 911125160Sjhb instance->li_file = file; 912125160Sjhb instance->li_line = line; 913125160Sjhb return; 914110779Speter } 915125160Sjhb 916125160Sjhb /* Update per-witness last file and line acquire. */ 91765557Sjasone w->w_file = file; 91865557Sjasone w->w_line = line; 919125160Sjhb 920125160Sjhb /* Find the next open lock instance in the list and fill it. */ 92174912Sjhb lle = *lock_list; 92276272Sjhb if (lle == NULL || lle->ll_count == LOCK_NCHILDREN) { 92378785Sjhb lle = witness_lock_list_get(); 92478785Sjhb if (lle == NULL) 92565557Sjasone return; 92678785Sjhb lle->ll_next = *lock_list; 92787593Sobrien CTR3(KTR_WITNESS, "%s: pid %d added lle %p", __func__, 92884680Sjhb td->td_proc->p_pid, lle); 92978785Sjhb *lock_list = lle; 93065557Sjasone } 931125160Sjhb instance = &lle->ll_children[lle->ll_count++]; 932125160Sjhb instance->li_lock = lock; 933125160Sjhb instance->li_line = line; 934125160Sjhb instance->li_file = file; 93576272Sjhb if ((flags & LOP_EXCLUSIVE) != 0) 936125160Sjhb instance->li_flags = LI_EXCLUSIVE; 93776272Sjhb else 938125160Sjhb instance->li_flags = 0; 93987593Sobrien CTR4(KTR_WITNESS, "%s: pid %d added %s as lle[%d]", __func__, 94084680Sjhb td->td_proc->p_pid, lock->lo_name, lle->ll_count - 1); 94165557Sjasone} 94265557Sjasone 94365557Sjasonevoid 94482244Sjhbwitness_upgrade(struct lock_object *lock, int flags, const char *file, int line) 94582244Sjhb{ 94682244Sjhb struct lock_instance *instance; 94782244Sjhb struct lock_class *class; 94882244Sjhb 94982284Sjhb KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 950112562Sjhb if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 95182244Sjhb return; 95282244Sjhb class = lock->lo_class; 953112116Sjhb file = fixup_filename(file); 95482244Sjhb if ((lock->lo_flags & LO_UPGRADABLE) == 0) 95582244Sjhb panic("upgrade of non-upgradable lock (%s) %s @ %s:%d", 95682244Sjhb class->lc_name, lock->lo_name, file, line); 95782244Sjhb if ((flags & LOP_TRYLOCK) == 0) 95882244Sjhb panic("non-try upgrade of lock (%s) %s @ %s:%d", class->lc_name, 95982244Sjhb lock->lo_name, file, line); 96082244Sjhb if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0) 96182244Sjhb panic("upgrade of non-sleep lock (%s) %s @ %s:%d", 96282244Sjhb class->lc_name, lock->lo_name, file, line); 96383366Sjulian instance = find_instance(curthread->td_sleeplocks, lock); 96482244Sjhb if (instance == NULL) 96582244Sjhb panic("upgrade of unlocked lock (%s) %s @ %s:%d", 96682244Sjhb class->lc_name, lock->lo_name, file, line); 96782244Sjhb if ((instance->li_flags & LI_EXCLUSIVE) != 0) 96882244Sjhb panic("upgrade of exclusive lock (%s) %s @ %s:%d", 96982244Sjhb class->lc_name, lock->lo_name, file, line); 97082244Sjhb if ((instance->li_flags & LI_RECURSEMASK) != 0) 97182244Sjhb panic("upgrade of recursed lock (%s) %s r=%d @ %s:%d", 97282244Sjhb class->lc_name, lock->lo_name, 97382244Sjhb instance->li_flags & LI_RECURSEMASK, file, line); 97482244Sjhb instance->li_flags |= LI_EXCLUSIVE; 97582244Sjhb} 97682244Sjhb 97782244Sjhbvoid 97882244Sjhbwitness_downgrade(struct lock_object *lock, int flags, const char *file, 97982244Sjhb int line) 98082244Sjhb{ 98182244Sjhb struct lock_instance *instance; 98282244Sjhb struct lock_class *class; 98382244Sjhb 98482284Sjhb KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 985112562Sjhb if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 98682244Sjhb return; 98782244Sjhb class = lock->lo_class; 988112116Sjhb file = fixup_filename(file); 98982244Sjhb if ((lock->lo_flags & LO_UPGRADABLE) == 0) 99082244Sjhb panic("downgrade of non-upgradable lock (%s) %s @ %s:%d", 99182244Sjhb class->lc_name, lock->lo_name, file, line); 99282244Sjhb if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0) 99382244Sjhb panic("downgrade of non-sleep lock (%s) %s @ %s:%d", 99482244Sjhb class->lc_name, lock->lo_name, file, line); 99583366Sjulian instance = find_instance(curthread->td_sleeplocks, lock); 99682244Sjhb if (instance == NULL) 99782244Sjhb panic("downgrade of unlocked lock (%s) %s @ %s:%d", 99882244Sjhb class->lc_name, lock->lo_name, file, line); 99982244Sjhb if ((instance->li_flags & LI_EXCLUSIVE) == 0) 100082244Sjhb panic("downgrade of shared lock (%s) %s @ %s:%d", 100182244Sjhb class->lc_name, lock->lo_name, file, line); 100282244Sjhb if ((instance->li_flags & LI_RECURSEMASK) != 0) 100382244Sjhb panic("downgrade of recursed lock (%s) %s r=%d @ %s:%d", 100482244Sjhb class->lc_name, lock->lo_name, 100582244Sjhb instance->li_flags & LI_RECURSEMASK, file, line); 100682244Sjhb instance->li_flags &= ~LI_EXCLUSIVE; 100782244Sjhb} 100882244Sjhb 100982244Sjhbvoid 101074912Sjhbwitness_unlock(struct lock_object *lock, int flags, const char *file, int line) 101165557Sjasone{ 101274912Sjhb struct lock_list_entry **lock_list, *lle; 101376272Sjhb struct lock_instance *instance; 101474912Sjhb struct lock_class *class; 101583366Sjulian struct thread *td; 101692858Simp register_t s; 101774912Sjhb int i, j; 101865557Sjasone 1019112562Sjhb if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL || 102080747Sjhb panicstr != NULL) 102171352Sjasone return; 102283366Sjulian td = curthread; 102374912Sjhb class = lock->lo_class; 1024112116Sjhb file = fixup_filename(file); 1025125160Sjhb 1026125160Sjhb /* Find lock instance associated with this lock. */ 102776272Sjhb if (class->lc_flags & LC_SLEEPLOCK) 102883366Sjulian lock_list = &td->td_sleeplocks; 102976272Sjhb else 103074912Sjhb lock_list = PCPU_PTR(spinlocks); 103174912Sjhb for (; *lock_list != NULL; lock_list = &(*lock_list)->ll_next) 103276272Sjhb for (i = 0; i < (*lock_list)->ll_count; i++) { 103376272Sjhb instance = &(*lock_list)->ll_children[i]; 1034125160Sjhb if (instance->li_lock == lock) 1035125160Sjhb goto found; 103676272Sjhb } 103776272Sjhb panic("lock (%s) %s not locked @ %s:%d", class->lc_name, lock->lo_name, 103876272Sjhb file, line); 1039125160Sjhbfound: 1040125160Sjhb 1041125160Sjhb /* First, check for shared/exclusive mismatches. */ 1042125160Sjhb if ((instance->li_flags & LI_EXCLUSIVE) != 0 && 1043125160Sjhb (flags & LOP_EXCLUSIVE) == 0) { 1044125160Sjhb printf("shared unlock of (%s) %s @ %s:%d\n", class->lc_name, 1045125160Sjhb lock->lo_name, file, line); 1046125160Sjhb printf("while exclusively locked from %s:%d\n", 1047125160Sjhb instance->li_file, instance->li_line); 1048125160Sjhb panic("excl->ushare"); 1049125160Sjhb } 1050125160Sjhb if ((instance->li_flags & LI_EXCLUSIVE) == 0 && 1051125160Sjhb (flags & LOP_EXCLUSIVE) != 0) { 1052125160Sjhb printf("exclusive unlock of (%s) %s @ %s:%d\n", class->lc_name, 1053125160Sjhb lock->lo_name, file, line); 1054125160Sjhb printf("while share locked from %s:%d\n", instance->li_file, 1055125160Sjhb instance->li_line); 1056125160Sjhb panic("share->uexcl"); 1057125160Sjhb } 1058125160Sjhb 1059125160Sjhb /* If we are recursed, unrecurse. */ 1060125160Sjhb if ((instance->li_flags & LI_RECURSEMASK) > 0) { 1061125160Sjhb CTR4(KTR_WITNESS, "%s: pid %d unrecursed on %s r=%d", __func__, 1062125160Sjhb td->td_proc->p_pid, instance->li_lock->lo_name, 1063125160Sjhb instance->li_flags); 1064125160Sjhb instance->li_flags--; 1065125160Sjhb return; 1066125160Sjhb } 1067125160Sjhb 1068125160Sjhb /* Otherwise, remove this item from the list. */ 1069125160Sjhb s = intr_disable(); 1070125160Sjhb CTR4(KTR_WITNESS, "%s: pid %d removed %s from lle[%d]", __func__, 1071125160Sjhb td->td_proc->p_pid, instance->li_lock->lo_name, 1072125160Sjhb (*lock_list)->ll_count - 1); 1073125160Sjhb for (j = i; j < (*lock_list)->ll_count - 1; j++) 1074125160Sjhb (*lock_list)->ll_children[j] = 1075125160Sjhb (*lock_list)->ll_children[j + 1]; 1076125160Sjhb (*lock_list)->ll_count--; 1077125160Sjhb intr_restore(s); 1078125160Sjhb 1079125160Sjhb /* If this lock list entry is now empty, free it. */ 1080125160Sjhb if ((*lock_list)->ll_count == 0) { 1081125160Sjhb lle = *lock_list; 1082125160Sjhb *lock_list = lle->ll_next; 1083125160Sjhb CTR3(KTR_WITNESS, "%s: pid %d removed lle %p", __func__, 1084125160Sjhb td->td_proc->p_pid, lle); 1085125160Sjhb witness_lock_list_free(lle); 1086125160Sjhb } 108765557Sjasone} 108865557Sjasone 108974912Sjhb/* 1090111881Sjhb * Warn if any locks other than 'lock' are held. Flags can be passed in to 1091111881Sjhb * exempt Giant and sleepable locks from the checks as well. If any 1092111881Sjhb * non-exempt locks are held, then a supplied message is printed to the 1093111881Sjhb * console along with a list of the offending locks. If indicated in the 1094111881Sjhb * flags then a failure results in a panic as well. 109574912Sjhb */ 109665557Sjasoneint 1097111881Sjhbwitness_warn(int flags, struct lock_object *lock, const char *fmt, ...) 109865557Sjasone{ 1099111881Sjhb struct lock_list_entry *lle; 110076272Sjhb struct lock_instance *lock1; 110183366Sjulian struct thread *td; 1102111881Sjhb va_list ap; 110374912Sjhb int i, n; 110465557Sjasone 1105112562Sjhb if (witness_cold || witness_watch == 0 || panicstr != NULL) 110674912Sjhb return (0); 110774912Sjhb n = 0; 110883366Sjulian td = curthread; 1109111881Sjhb for (lle = td->td_sleeplocks; lle != NULL; lle = lle->ll_next) 111074912Sjhb for (i = lle->ll_count - 1; i >= 0; i--) { 111176272Sjhb lock1 = &lle->ll_children[i]; 1112111881Sjhb if (lock1->li_lock == lock) 1113111881Sjhb continue; 1114111881Sjhb if (flags & WARN_GIANTOK && 111576272Sjhb lock1->li_lock == &Giant.mtx_object) 111674912Sjhb continue; 1117111881Sjhb if (flags & WARN_SLEEPOK && 1118111881Sjhb (lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0) 111976272Sjhb continue; 1120111881Sjhb if (n == 0) { 1121111881Sjhb va_start(ap, fmt); 1122111881Sjhb vprintf(fmt, ap); 1123111881Sjhb va_end(ap); 1124111881Sjhb printf(" with the following"); 1125111881Sjhb if (flags & WARN_SLEEPOK) 1126111881Sjhb printf(" non-sleepable"); 1127118441Sjhb printf(" locks held:\n"); 112876272Sjhb } 112974912Sjhb n++; 1130111881Sjhb witness_list_lock(lock1); 113174912Sjhb } 1132111881Sjhb if (PCPU_GET(spinlocks) != NULL) { 113397006Sjhb /* 113497006Sjhb * Since we already hold a spinlock preemption is 113597006Sjhb * already blocked. 113697006Sjhb */ 1137111881Sjhb if (n == 0) { 1138111881Sjhb va_start(ap, fmt); 1139111881Sjhb vprintf(fmt, ap); 1140111881Sjhb va_end(ap); 1141111881Sjhb printf(" with the following"); 1142111881Sjhb if (flags & WARN_SLEEPOK) 1143111881Sjhb printf(" non-sleepable"); 1144118441Sjhb printf(" locks held:\n"); 1145111881Sjhb } 1146111881Sjhb n += witness_list_locks(PCPU_PTR(spinlocks)); 114765557Sjasone } 1148111881Sjhb if (flags & WARN_PANIC && n) 1149111881Sjhb panic("witness_warn"); 115067676Sjhb#ifdef DDB 1151111881Sjhb else if (witness_ddb && n) 115275711Sjhb Debugger(__func__); 1153127323Salfred else if (witness_trace && n) 1154127323Salfred backtrace(); 1155111881Sjhb#endif 115665557Sjasone return (n); 115765557Sjasone} 115865557Sjasone 1159102448Siedowseconst char * 1160102448Siedowsewitness_file(struct lock_object *lock) 1161102448Siedowse{ 1162102448Siedowse struct witness *w; 1163102448Siedowse 1164112562Sjhb if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL) 1165102448Siedowse return ("?"); 1166102448Siedowse w = lock->lo_witness; 1167102448Siedowse return (w->w_file); 1168102448Siedowse} 1169102448Siedowse 1170102448Siedowseint 1171102448Siedowsewitness_line(struct lock_object *lock) 1172102448Siedowse{ 1173102448Siedowse struct witness *w; 1174102448Siedowse 1175112562Sjhb if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL) 1176102448Siedowse return (0); 1177102448Siedowse w = lock->lo_witness; 1178102448Siedowse return (w->w_line); 1179102448Siedowse} 1180102448Siedowse 118165856Sjhbstatic struct witness * 118274912Sjhbenroll(const char *description, struct lock_class *lock_class) 118365557Sjasone{ 118474912Sjhb struct witness *w; 118565557Sjasone 1186125348Sjhb if (witness_watch == 0 || panicstr != NULL) 118765557Sjasone return (NULL); 118874912Sjhb if ((lock_class->lc_flags & LC_SPINLOCK) && witness_skipspin) 118965557Sjasone return (NULL); 119074912Sjhb mtx_lock_spin(&w_mtx); 119174912Sjhb STAILQ_FOREACH(w, &w_all, w_list) { 119297948Sjhb if (w->w_name == description || (w->w_refcount > 0 && 119397948Sjhb strcmp(description, w->w_name) == 0)) { 119475362Sjhb w->w_refcount++; 119574912Sjhb mtx_unlock_spin(&w_mtx); 119674912Sjhb if (lock_class != w->w_class) 119774912Sjhb panic( 119874912Sjhb "lock (%s) %s does not match earlier (%s) lock", 119974912Sjhb description, lock_class->lc_name, 120074912Sjhb w->w_class->lc_name); 120165557Sjasone return (w); 120265557Sjasone } 120365557Sjasone } 120474912Sjhb /* 120574912Sjhb * This isn't quite right, as witness_cold is still 0 while we 120674912Sjhb * enroll all the locks initialized before witness_initialize(). 120774912Sjhb */ 120875364Sbp if ((lock_class->lc_flags & LC_SPINLOCK) && !witness_cold) { 120975364Sbp mtx_unlock_spin(&w_mtx); 121074912Sjhb panic("spin lock %s not in order list", description); 121175364Sbp } 121265557Sjasone if ((w = witness_get()) == NULL) 121365557Sjasone return (NULL); 121474912Sjhb w->w_name = description; 121574912Sjhb w->w_class = lock_class; 121675362Sjhb w->w_refcount = 1; 121774912Sjhb STAILQ_INSERT_HEAD(&w_all, w, w_list); 121874912Sjhb if (lock_class->lc_flags & LC_SPINLOCK) 121974912Sjhb STAILQ_INSERT_HEAD(&w_spin, w, w_typelist); 122074912Sjhb else if (lock_class->lc_flags & LC_SLEEPLOCK) 122174912Sjhb STAILQ_INSERT_HEAD(&w_sleep, w, w_typelist); 122275364Sbp else { 122375364Sbp mtx_unlock_spin(&w_mtx); 122474912Sjhb panic("lock class %s is not sleep or spin", 122574912Sjhb lock_class->lc_name); 122675364Sbp } 122774912Sjhb mtx_unlock_spin(&w_mtx); 122865557Sjasone return (w); 122965557Sjasone} 123065557Sjasone 1231112117Sjhb/* Don't let the door bang you on the way out... */ 123265557Sjasonestatic int 1233112117Sjhbdepart(struct witness *w) 123465557Sjasone{ 1235112117Sjhb struct witness_child_list_entry *wcl, *nwcl; 123674912Sjhb struct witness_list *list; 1237112117Sjhb struct witness *parent; 123865557Sjasone 1239112117Sjhb MPASS(w->w_refcount == 0); 1240112117Sjhb if (w->w_class->lc_flags & LC_SLEEPLOCK) 1241112117Sjhb list = &w_sleep; 1242112117Sjhb else 1243112117Sjhb list = &w_spin; 1244112117Sjhb /* 1245112117Sjhb * First, we run through the entire tree looking for any 1246112117Sjhb * witnesses that the outgoing witness is a child of. For 1247112117Sjhb * each parent that we find, we reparent all the direct 1248112117Sjhb * children of the outgoing witness to its parent. 1249112117Sjhb */ 1250112117Sjhb STAILQ_FOREACH(parent, list, w_typelist) { 1251112117Sjhb if (!isitmychild(parent, w)) 1252112117Sjhb continue; 1253112117Sjhb removechild(parent, w); 1254112117Sjhb if (!reparentchildren(parent, w)) 1255112117Sjhb return (0); 1256112117Sjhb } 1257112117Sjhb 1258112117Sjhb /* 1259112117Sjhb * Now we go through and free up the child list of the 1260112117Sjhb * outgoing witness. 1261112117Sjhb */ 1262112117Sjhb for (wcl = w->w_children; wcl != NULL; wcl = nwcl) { 1263112117Sjhb nwcl = wcl->wcl_next; 1264112117Sjhb witness_child_free(wcl); 1265112117Sjhb } 1266112117Sjhb 1267112117Sjhb /* 1268112117Sjhb * Detach from various lists and free. 1269112117Sjhb */ 1270112117Sjhb STAILQ_REMOVE(list, w, witness, w_typelist); 1271112117Sjhb STAILQ_REMOVE(&w_all, w, witness, w_list); 1272112117Sjhb witness_free(w); 1273112117Sjhb 1274112117Sjhb /* Finally, fixup the tree. */ 1275112117Sjhb return (rebalancetree(list)); 1276112117Sjhb} 1277112117Sjhb 1278112117Sjhb/* 1279112117Sjhb * Prune an entire lock order tree. We look for cases where a lock 1280112117Sjhb * is now both a descendant and a direct child of a given lock. In 1281112117Sjhb * that case, we want to remove the direct child link from the tree. 1282112117Sjhb * 1283112117Sjhb * Returns false if insertchild() fails. 1284112117Sjhb */ 1285112117Sjhbstatic int 1286112117Sjhbrebalancetree(struct witness_list *list) 1287112117Sjhb{ 1288112117Sjhb struct witness *child, *parent; 1289112117Sjhb 1290112117Sjhb STAILQ_FOREACH(child, list, w_typelist) { 1291112117Sjhb STAILQ_FOREACH(parent, list, w_typelist) { 1292112117Sjhb if (!isitmychild(parent, child)) 1293112117Sjhb continue; 1294112117Sjhb removechild(parent, child); 1295112117Sjhb if (isitmydescendant(parent, child)) 1296112117Sjhb continue; 1297112117Sjhb if (!insertchild(parent, child)) 1298112117Sjhb return (0); 1299112117Sjhb } 1300112117Sjhb } 1301112117Sjhb witness_levelall(); 1302112117Sjhb return (1); 1303112117Sjhb} 1304112117Sjhb 1305112117Sjhb/* 1306112117Sjhb * Add "child" as a direct child of "parent". Returns false if 1307112117Sjhb * we fail due to out of memory. 1308112117Sjhb */ 1309112117Sjhbstatic int 1310112117Sjhbinsertchild(struct witness *parent, struct witness *child) 1311112117Sjhb{ 1312112117Sjhb struct witness_child_list_entry **wcl; 1313112117Sjhb 131474912Sjhb MPASS(child != NULL && parent != NULL); 131574912Sjhb 131665557Sjasone /* 131765557Sjasone * Insert "child" after "parent" 131865557Sjasone */ 131974912Sjhb wcl = &parent->w_children; 132074912Sjhb while (*wcl != NULL && (*wcl)->wcl_count == WITNESS_NCHILDREN) 132174912Sjhb wcl = &(*wcl)->wcl_next; 132274912Sjhb if (*wcl == NULL) { 132374912Sjhb *wcl = witness_child_get(); 132474912Sjhb if (*wcl == NULL) 1325112117Sjhb return (0); 132665557Sjasone } 132774912Sjhb (*wcl)->wcl_children[(*wcl)->wcl_count++] = child; 132874912Sjhb 1329112117Sjhb return (1); 1330112117Sjhb} 1331112117Sjhb 1332112117Sjhb/* 1333112117Sjhb * Make all the direct descendants of oldparent be direct descendants 1334112117Sjhb * of newparent. 1335112117Sjhb */ 1336112117Sjhbstatic int 1337112117Sjhbreparentchildren(struct witness *newparent, struct witness *oldparent) 1338112117Sjhb{ 1339112117Sjhb struct witness_child_list_entry *wcl; 1340112117Sjhb int i; 1341112117Sjhb 1342112117Sjhb /* Avoid making a witness a child of itself. */ 1343112117Sjhb MPASS(!isitmychild(oldparent, newparent)); 1344112117Sjhb 1345112117Sjhb for (wcl = oldparent->w_children; wcl != NULL; wcl = wcl->wcl_next) 1346112117Sjhb for (i = 0; i < wcl->wcl_count; i++) 1347112117Sjhb if (!insertchild(newparent, wcl->wcl_children[i])) 1348112117Sjhb return (0); 1349112117Sjhb return (1); 1350112117Sjhb} 1351112117Sjhb 1352112117Sjhbstatic int 1353112117Sjhbitismychild(struct witness *parent, struct witness *child) 1354112117Sjhb{ 1355112117Sjhb struct witness_list *list; 1356112117Sjhb 1357112117Sjhb MPASS(child != NULL && parent != NULL); 1358112117Sjhb if ((parent->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK)) != 1359112117Sjhb (child->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK))) 1360112117Sjhb panic( 1361112117Sjhb "%s: parent (%s) and child (%s) are not the same lock type", 1362112117Sjhb __func__, parent->w_class->lc_name, 1363112117Sjhb child->w_class->lc_name); 1364112117Sjhb 1365112117Sjhb if (!insertchild(parent, child)) 136665557Sjasone return (0); 1367112117Sjhb 136874912Sjhb if (parent->w_class->lc_flags & LC_SLEEPLOCK) 136974912Sjhb list = &w_sleep; 137074912Sjhb else 137174912Sjhb list = &w_spin; 1372112117Sjhb return (rebalancetree(list)); 137365557Sjasone} 137465557Sjasone 137565557Sjasonestatic void 137665856Sjhbremovechild(struct witness *parent, struct witness *child) 137765557Sjasone{ 137874912Sjhb struct witness_child_list_entry **wcl, *wcl1; 137965557Sjasone int i; 138065557Sjasone 138174912Sjhb for (wcl = &parent->w_children; *wcl != NULL; wcl = &(*wcl)->wcl_next) 138274912Sjhb for (i = 0; i < (*wcl)->wcl_count; i++) 138374912Sjhb if ((*wcl)->wcl_children[i] == child) 138465557Sjasone goto found; 138565557Sjasone return; 138665557Sjasonefound: 138774912Sjhb (*wcl)->wcl_count--; 138874912Sjhb if ((*wcl)->wcl_count > i) 138974912Sjhb (*wcl)->wcl_children[i] = 139074912Sjhb (*wcl)->wcl_children[(*wcl)->wcl_count]; 139174912Sjhb MPASS((*wcl)->wcl_children[i] != NULL); 139274912Sjhb if ((*wcl)->wcl_count != 0) 139365557Sjasone return; 139474912Sjhb wcl1 = *wcl; 139574912Sjhb *wcl = wcl1->wcl_next; 139674912Sjhb witness_child_free(wcl1); 139765557Sjasone} 139865557Sjasone 139965557Sjasonestatic int 140065856Sjhbisitmychild(struct witness *parent, struct witness *child) 140165557Sjasone{ 140274912Sjhb struct witness_child_list_entry *wcl; 140365557Sjasone int i; 140465557Sjasone 140574912Sjhb for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) { 140674912Sjhb for (i = 0; i < wcl->wcl_count; i++) { 140774912Sjhb if (wcl->wcl_children[i] == child) 140865557Sjasone return (1); 140965557Sjasone } 141065557Sjasone } 141165557Sjasone return (0); 141265557Sjasone} 141365557Sjasone 141465557Sjasonestatic int 141565856Sjhbisitmydescendant(struct witness *parent, struct witness *child) 141665557Sjasone{ 141774912Sjhb struct witness_child_list_entry *wcl; 141874912Sjhb int i, j; 141965557Sjasone 142074912Sjhb if (isitmychild(parent, child)) 142174912Sjhb return (1); 142274912Sjhb j = 0; 142374912Sjhb for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) { 142467352Sjhb MPASS(j < 1000); 142574912Sjhb for (i = 0; i < wcl->wcl_count; i++) { 142674912Sjhb if (isitmydescendant(wcl->wcl_children[i], child)) 142765557Sjasone return (1); 142865557Sjasone } 142974912Sjhb j++; 143065557Sjasone } 143165557Sjasone return (0); 143265557Sjasone} 143365557Sjasone 1434104094Sphkstatic void 143565557Sjasonewitness_levelall (void) 143665557Sjasone{ 143774912Sjhb struct witness_list *list; 143865856Sjhb struct witness *w, *w1; 143965557Sjasone 144074912Sjhb /* 144174912Sjhb * First clear all levels. 144274912Sjhb */ 144374912Sjhb STAILQ_FOREACH(w, &w_all, w_list) { 144474912Sjhb w->w_level = 0; 144574912Sjhb } 144674912Sjhb 144774912Sjhb /* 144874912Sjhb * Look for locks with no parent and level all their descendants. 144974912Sjhb */ 145074912Sjhb STAILQ_FOREACH(w, &w_all, w_list) { 145174912Sjhb /* 145274912Sjhb * This is just an optimization, technically we could get 145374912Sjhb * away just walking the all list each time. 145474912Sjhb */ 145574912Sjhb if (w->w_class->lc_flags & LC_SLEEPLOCK) 145674912Sjhb list = &w_sleep; 145774912Sjhb else 145874912Sjhb list = &w_spin; 145974912Sjhb STAILQ_FOREACH(w1, list, w_typelist) { 146065557Sjasone if (isitmychild(w1, w)) 146174912Sjhb goto skip; 146265557Sjasone } 146365557Sjasone witness_leveldescendents(w, 0); 146474912Sjhb skip: 146595541Smarcel ; /* silence GCC 3.x */ 146665557Sjasone } 146765557Sjasone} 146865557Sjasone 146965557Sjasonestatic void 147065856Sjhbwitness_leveldescendents(struct witness *parent, int level) 147165557Sjasone{ 147274912Sjhb struct witness_child_list_entry *wcl; 147365557Sjasone int i; 147465557Sjasone 147565557Sjasone if (parent->w_level < level) 147665557Sjasone parent->w_level = level; 147765557Sjasone level++; 147874912Sjhb for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) 147974912Sjhb for (i = 0; i < wcl->wcl_count; i++) 148074912Sjhb witness_leveldescendents(wcl->wcl_children[i], level); 148165557Sjasone} 148265557Sjasone 148365557Sjasonestatic void 148465856Sjhbwitness_displaydescendants(void(*prnt)(const char *fmt, ...), 1485112118Sjhb struct witness *parent, int indent) 148665557Sjasone{ 148774912Sjhb struct witness_child_list_entry *wcl; 148874912Sjhb int i, level; 148965557Sjasone 149095543Sjhb level = parent->w_level; 149174912Sjhb prnt("%-2d", level); 1492112118Sjhb for (i = 0; i < indent; i++) 149365557Sjasone prnt(" "); 1494112118Sjhb if (parent->w_refcount > 0) 1495112118Sjhb prnt("%s", parent->w_name); 1496112118Sjhb else 1497112118Sjhb prnt("(dead)"); 1498112118Sjhb if (parent->w_displayed) { 1499112118Sjhb prnt(" -- (already displayed)\n"); 1500112118Sjhb return; 1501112118Sjhb } 1502112118Sjhb parent->w_displayed = 1; 150397948Sjhb if (parent->w_refcount > 0) { 150497948Sjhb if (parent->w_file != NULL) 1505112118Sjhb prnt(" -- last acquired @ %s:%d", parent->w_file, 150697948Sjhb parent->w_line); 1507112118Sjhb } 1508112118Sjhb prnt("\n"); 150974912Sjhb for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) 151074912Sjhb for (i = 0; i < wcl->wcl_count; i++) 151174912Sjhb witness_displaydescendants(prnt, 1512112118Sjhb wcl->wcl_children[i], indent + 1); 151374912Sjhb} 151465557Sjasone 1515105508Sphk#ifdef BLESSING 151665557Sjasonestatic int 151765856Sjhbblessed(struct witness *w1, struct witness *w2) 151865557Sjasone{ 151965557Sjasone int i; 152065856Sjhb struct witness_blessed *b; 152165557Sjasone 152265557Sjasone for (i = 0; i < blessed_count; i++) { 152365557Sjasone b = &blessed_list[i]; 152474912Sjhb if (strcmp(w1->w_name, b->b_lock1) == 0) { 152574912Sjhb if (strcmp(w2->w_name, b->b_lock2) == 0) 152665557Sjasone return (1); 152765557Sjasone continue; 152865557Sjasone } 152974912Sjhb if (strcmp(w1->w_name, b->b_lock2) == 0) 153074912Sjhb if (strcmp(w2->w_name, b->b_lock1) == 0) 153165557Sjasone return (1); 153265557Sjasone } 153365557Sjasone return (0); 153465557Sjasone} 1535105508Sphk#endif 153665557Sjasone 153765856Sjhbstatic struct witness * 153874912Sjhbwitness_get(void) 153965557Sjasone{ 154065856Sjhb struct witness *w; 154165557Sjasone 1542112562Sjhb if (witness_watch == 0) { 154376481Sjhb mtx_unlock_spin(&w_mtx); 154476481Sjhb return (NULL); 154576481Sjhb } 154674912Sjhb if (STAILQ_EMPTY(&w_free)) { 1547112562Sjhb witness_watch = 0; 154874912Sjhb mtx_unlock_spin(&w_mtx); 154974912Sjhb printf("%s: witness exhausted\n", __func__); 155065557Sjasone return (NULL); 155165557Sjasone } 155274912Sjhb w = STAILQ_FIRST(&w_free); 155374912Sjhb STAILQ_REMOVE_HEAD(&w_free, w_list); 155465856Sjhb bzero(w, sizeof(*w)); 155565557Sjasone return (w); 155665557Sjasone} 155765557Sjasone 155865557Sjasonestatic void 155965856Sjhbwitness_free(struct witness *w) 156065557Sjasone{ 156174912Sjhb 156274912Sjhb STAILQ_INSERT_HEAD(&w_free, w, w_list); 156365557Sjasone} 156465557Sjasone 156574912Sjhbstatic struct witness_child_list_entry * 156674912Sjhbwitness_child_get(void) 156765557Sjasone{ 156874912Sjhb struct witness_child_list_entry *wcl; 156965557Sjasone 1570112562Sjhb if (witness_watch == 0) { 157176481Sjhb mtx_unlock_spin(&w_mtx); 157276481Sjhb return (NULL); 157376481Sjhb } 157474912Sjhb wcl = w_child_free; 157574912Sjhb if (wcl == NULL) { 1576112562Sjhb witness_watch = 0; 157774912Sjhb mtx_unlock_spin(&w_mtx); 157874912Sjhb printf("%s: witness exhausted\n", __func__); 157974912Sjhb return (NULL); 158065557Sjasone } 158174912Sjhb w_child_free = wcl->wcl_next; 158274912Sjhb bzero(wcl, sizeof(*wcl)); 158374912Sjhb return (wcl); 158474912Sjhb} 158569881Sjake 158674912Sjhbstatic void 158774912Sjhbwitness_child_free(struct witness_child_list_entry *wcl) 158874912Sjhb{ 158974912Sjhb 159074912Sjhb wcl->wcl_next = w_child_free; 159174912Sjhb w_child_free = wcl; 159265557Sjasone} 159365557Sjasone 159474912Sjhbstatic struct lock_list_entry * 159574912Sjhbwitness_lock_list_get(void) 159674912Sjhb{ 159774912Sjhb struct lock_list_entry *lle; 159871709Sjhb 1599112562Sjhb if (witness_watch == 0) 160076481Sjhb return (NULL); 160174912Sjhb mtx_lock_spin(&w_mtx); 160274912Sjhb lle = w_lock_list_free; 160374912Sjhb if (lle == NULL) { 1604112562Sjhb witness_watch = 0; 160574912Sjhb mtx_unlock_spin(&w_mtx); 160674912Sjhb printf("%s: witness exhausted\n", __func__); 160774912Sjhb return (NULL); 160874912Sjhb } 160974912Sjhb w_lock_list_free = lle->ll_next; 161074912Sjhb mtx_unlock_spin(&w_mtx); 161174912Sjhb bzero(lle, sizeof(*lle)); 161274912Sjhb return (lle); 161374912Sjhb} 161474912Sjhb 161574912Sjhbstatic void 161674912Sjhbwitness_lock_list_free(struct lock_list_entry *lle) 161771709Sjhb{ 161871709Sjhb 161974912Sjhb mtx_lock_spin(&w_mtx); 162074912Sjhb lle->ll_next = w_lock_list_free; 162174912Sjhb w_lock_list_free = lle; 162274912Sjhb mtx_unlock_spin(&w_mtx); 162371709Sjhb} 162471709Sjhb 162576272Sjhbstatic struct lock_instance * 162676272Sjhbfind_instance(struct lock_list_entry *lock_list, struct lock_object *lock) 162776272Sjhb{ 162876272Sjhb struct lock_list_entry *lle; 162976272Sjhb struct lock_instance *instance; 163076272Sjhb int i; 163176272Sjhb 163276272Sjhb for (lle = lock_list; lle != NULL; lle = lle->ll_next) 163376272Sjhb for (i = lle->ll_count - 1; i >= 0; i--) { 163476272Sjhb instance = &lle->ll_children[i]; 163576272Sjhb if (instance->li_lock == lock) 163676272Sjhb return (instance); 163776272Sjhb } 163876272Sjhb return (NULL); 163976272Sjhb} 164076272Sjhb 1641111881Sjhbstatic void 1642111881Sjhbwitness_list_lock(struct lock_instance *instance) 1643111881Sjhb{ 1644111881Sjhb struct lock_object *lock; 1645111881Sjhb 1646111881Sjhb lock = instance->li_lock; 1647111881Sjhb printf("%s %s %s", (instance->li_flags & LI_EXCLUSIVE) != 0 ? 1648111881Sjhb "exclusive" : "shared", lock->lo_class->lc_name, lock->lo_name); 1649111881Sjhb if (lock->lo_type != lock->lo_name) 1650111881Sjhb printf(" (%s)", lock->lo_type); 1651111881Sjhb printf(" r = %d (%p) locked @ %s:%d\n", 1652111881Sjhb instance->li_flags & LI_RECURSEMASK, lock, instance->li_file, 1653111881Sjhb instance->li_line); 1654111881Sjhb} 1655111881Sjhb 165674912Sjhbint 165775273Sjhbwitness_list_locks(struct lock_list_entry **lock_list) 165872224Sjhb{ 165975273Sjhb struct lock_list_entry *lle; 166074912Sjhb int i, nheld; 166172224Sjhb 166274912Sjhb nheld = 0; 166374912Sjhb for (lle = *lock_list; lle != NULL; lle = lle->ll_next) 166474912Sjhb for (i = lle->ll_count - 1; i >= 0; i--) { 1665111881Sjhb witness_list_lock(&lle->ll_children[i]); 166674912Sjhb nheld++; 166774912Sjhb } 166875273Sjhb return (nheld); 166975273Sjhb} 167075273Sjhb 1671118271Sjhb/* 1672118271Sjhb * This is a bit risky at best. We call this function when we have timed 1673118271Sjhb * out acquiring a spin lock, and we assume that the other CPU is stuck 1674118271Sjhb * with this lock held. So, we go groveling around in the other CPU's 1675118271Sjhb * per-cpu data to try to find the lock instance for this spin lock to 1676118271Sjhb * see when it was last acquired. 1677118271Sjhb */ 167865557Sjasonevoid 1679118271Sjhbwitness_display_spinlock(struct lock_object *lock, struct thread *owner) 1680118271Sjhb{ 1681118271Sjhb struct lock_instance *instance; 1682118271Sjhb struct pcpu *pc; 1683118271Sjhb 1684118271Sjhb if (owner->td_critnest == 0 || owner->td_oncpu == NOCPU) 1685118271Sjhb return; 1686118271Sjhb pc = pcpu_find(owner->td_oncpu); 1687118271Sjhb instance = find_instance(pc->pc_spinlocks, lock); 1688118271Sjhb if (instance != NULL) 1689118271Sjhb witness_list_lock(instance); 1690118271Sjhb} 1691118271Sjhb 1692118271Sjhbvoid 169374912Sjhbwitness_save(struct lock_object *lock, const char **filep, int *linep) 169465557Sjasone{ 169576272Sjhb struct lock_instance *instance; 169671320Sjasone 169782284Sjhb KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 1698112562Sjhb if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 169971352Sjasone return; 170082243Sjhb if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0) 170182243Sjhb panic("%s: lock (%s) %s is not a sleep lock", __func__, 170282243Sjhb lock->lo_class->lc_name, lock->lo_name); 170383366Sjulian instance = find_instance(curthread->td_sleeplocks, lock); 170482243Sjhb if (instance == NULL) 170582243Sjhb panic("%s: lock (%s) %s not locked", __func__, 170682243Sjhb lock->lo_class->lc_name, lock->lo_name); 170776272Sjhb *filep = instance->li_file; 170876272Sjhb *linep = instance->li_line; 170965557Sjasone} 171065557Sjasone 171165557Sjasonevoid 171274912Sjhbwitness_restore(struct lock_object *lock, const char *file, int line) 171365557Sjasone{ 171476272Sjhb struct lock_instance *instance; 171571320Sjasone 171682284Sjhb KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 1717112562Sjhb if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 171871352Sjasone return; 171982243Sjhb if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0) 172082243Sjhb panic("%s: lock (%s) %s is not a sleep lock", __func__, 172182243Sjhb lock->lo_class->lc_name, lock->lo_name); 172283366Sjulian instance = find_instance(curthread->td_sleeplocks, lock); 172382243Sjhb if (instance == NULL) 172482243Sjhb panic("%s: lock (%s) %s not locked", __func__, 172582243Sjhb lock->lo_class->lc_name, lock->lo_name); 172674912Sjhb lock->lo_witness->w_file = file; 172774912Sjhb lock->lo_witness->w_line = line; 172876272Sjhb instance->li_file = file; 172976272Sjhb instance->li_line = line; 173065557Sjasone} 173165557Sjasone 173278871Sjhbvoid 173378871Sjhbwitness_assert(struct lock_object *lock, int flags, const char *file, int line) 173478871Sjhb{ 173578871Sjhb#ifdef INVARIANT_SUPPORT 173678871Sjhb struct lock_instance *instance; 173778871Sjhb 1738112562Sjhb if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 173978941Sjhb return; 174078871Sjhb if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) != 0) 174183366Sjulian instance = find_instance(curthread->td_sleeplocks, lock); 174278871Sjhb else if ((lock->lo_class->lc_flags & LC_SPINLOCK) != 0) 174378871Sjhb instance = find_instance(PCPU_GET(spinlocks), lock); 174486422Sjhb else { 174578871Sjhb panic("Lock (%s) %s is not sleep or spin!", 174678871Sjhb lock->lo_class->lc_name, lock->lo_name); 174786422Sjhb } 1748112116Sjhb file = fixup_filename(file); 174978871Sjhb switch (flags) { 175078871Sjhb case LA_UNLOCKED: 175178871Sjhb if (instance != NULL) 175278871Sjhb panic("Lock (%s) %s locked @ %s:%d.", 175378871Sjhb lock->lo_class->lc_name, lock->lo_name, file, line); 175478871Sjhb break; 175578871Sjhb case LA_LOCKED: 175678871Sjhb case LA_LOCKED | LA_RECURSED: 175778871Sjhb case LA_LOCKED | LA_NOTRECURSED: 175878871Sjhb case LA_SLOCKED: 175978871Sjhb case LA_SLOCKED | LA_RECURSED: 176078871Sjhb case LA_SLOCKED | LA_NOTRECURSED: 176178871Sjhb case LA_XLOCKED: 176278871Sjhb case LA_XLOCKED | LA_RECURSED: 176378871Sjhb case LA_XLOCKED | LA_NOTRECURSED: 176486422Sjhb if (instance == NULL) { 176578871Sjhb panic("Lock (%s) %s not locked @ %s:%d.", 176678871Sjhb lock->lo_class->lc_name, lock->lo_name, file, line); 176786422Sjhb break; 176886422Sjhb } 176978871Sjhb if ((flags & LA_XLOCKED) != 0 && 177078871Sjhb (instance->li_flags & LI_EXCLUSIVE) == 0) 177178871Sjhb panic("Lock (%s) %s not exclusively locked @ %s:%d.", 177278871Sjhb lock->lo_class->lc_name, lock->lo_name, file, line); 177378871Sjhb if ((flags & LA_SLOCKED) != 0 && 177478871Sjhb (instance->li_flags & LI_EXCLUSIVE) != 0) 177578871Sjhb panic("Lock (%s) %s exclusively locked @ %s:%d.", 177678871Sjhb lock->lo_class->lc_name, lock->lo_name, file, line); 177778871Sjhb if ((flags & LA_RECURSED) != 0 && 177878871Sjhb (instance->li_flags & LI_RECURSEMASK) == 0) 177978871Sjhb panic("Lock (%s) %s not recursed @ %s:%d.", 178078871Sjhb lock->lo_class->lc_name, lock->lo_name, file, line); 178178871Sjhb if ((flags & LA_NOTRECURSED) != 0 && 178278871Sjhb (instance->li_flags & LI_RECURSEMASK) != 0) 178378871Sjhb panic("Lock (%s) %s recursed @ %s:%d.", 178478871Sjhb lock->lo_class->lc_name, lock->lo_name, file, line); 178578871Sjhb break; 178678871Sjhb default: 178778871Sjhb panic("Invalid lock assertion at %s:%d.", file, line); 178878871Sjhb 178978871Sjhb } 179078871Sjhb#endif /* INVARIANT_SUPPORT */ 179178871Sjhb} 179278871Sjhb 179374912Sjhb#ifdef DDB 1794112061Sjhbstatic void 1795112061Sjhbwitness_list(struct thread *td) 1796112061Sjhb{ 179774912Sjhb 1798112061Sjhb KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 1799112061Sjhb KASSERT(db_active, ("%s: not in the debugger", __func__)); 1800112061Sjhb 1801112562Sjhb if (witness_watch == 0) 1802112061Sjhb return; 1803112061Sjhb 1804112061Sjhb witness_list_locks(&td->td_sleeplocks); 1805112061Sjhb 1806112061Sjhb /* 1807112061Sjhb * We only handle spinlocks if td == curthread. This is somewhat broken 1808112061Sjhb * if td is currently executing on some other CPU and holds spin locks 1809112061Sjhb * as we won't display those locks. If we had a MI way of getting 1810112061Sjhb * the per-cpu data for a given cpu then we could use 1811113339Sjulian * td->td_oncpu to get the list of spinlocks for this thread 1812112061Sjhb * and "fix" this. 1813112061Sjhb * 1814112061Sjhb * That still wouldn't really fix this unless we locked sched_lock 1815112061Sjhb * or stopped the other CPU to make sure it wasn't changing the list 1816112061Sjhb * out from under us. It is probably best to just not try to handle 1817112061Sjhb * threads on other CPU's for now. 1818112061Sjhb */ 1819112061Sjhb if (td == curthread && PCPU_GET(spinlocks) != NULL) 1820112061Sjhb witness_list_locks(PCPU_PTR(spinlocks)); 1821112061Sjhb} 1822112061Sjhb 182374930SjhbDB_SHOW_COMMAND(locks, db_witness_list) 182474912Sjhb{ 182583366Sjulian struct thread *td; 182683366Sjulian pid_t pid; 182775273Sjhb struct proc *p; 182874912Sjhb 182975273Sjhb if (have_addr) { 183075273Sjhb pid = (addr % 16) + ((addr >> 4) % 16) * 10 + 183175273Sjhb ((addr >> 8) % 16) * 100 + ((addr >> 12) % 16) * 1000 + 183275273Sjhb ((addr >> 16) % 16) * 10000; 183375273Sjhb /* sx_slock(&allproc_lock); */ 183483366Sjulian FOREACH_PROC_IN_SYSTEM(p) { 183575273Sjhb if (p->p_pid == pid) 183675273Sjhb break; 183775273Sjhb } 183875273Sjhb /* sx_sunlock(&allproc_lock); */ 183975273Sjhb if (p == NULL) { 184075273Sjhb db_printf("pid %d not found\n", pid); 184175273Sjhb return; 184275273Sjhb } 184390361Sjulian FOREACH_THREAD_IN_PROC(p, td) { 184490361Sjulian witness_list(td); 184590361Sjulian } 184683366Sjulian } else { 184783366Sjulian td = curthread; 184890361Sjulian witness_list(td); 184983366Sjulian } 185074912Sjhb} 185174912Sjhb 185274912SjhbDB_SHOW_COMMAND(witness, db_witness_display) 185374912Sjhb{ 185474912Sjhb 185574912Sjhb witness_display(db_printf); 185674912Sjhb} 185774912Sjhb#endif 1858