src/runtime/thread.c

   1 #include <stdlib.h>
   2 #include <stdio.h>
   3 #include <string.h>
   4 #include <sched.h>
   5 #include <signal.h>
   6 #include <stddef.h>
   7 #include <errno.h>
   8 #include <sys/types.h>
   9 #include <sys/wait.h>
  10
  11 #include "sbcl.h"
  12 #include "runtime.h"
  13 #include "validate.h"           /* for CONTROL_STACK_SIZE etc */
  14 #include "alloc.h"
  15 #include "thread.h"
  16 #include "arch.h"
  17 #include "target-arch-os.h"
  18 #include "os.h"
  19 #include "globals.h"
  20 #include "dynbind.h"
  21 #include "genesis/cons.h"
  22 #include "genesis/fdefn.h"
  23 #include "interr.h"             /* for lose() */
  24 #include "gc-internal.h"
  25
  26 #define ALIEN_STACK_SIZE (1*1024*1024) /* 1Mb size chosen at random */
  27
  28 struct freeable_stack {
  29     os_thread_t os_thread;
  30     os_vm_address_t stack;
  31 };
  32
  33 static struct freeable_stack * volatile freeable_stack = 0;
  34
  35 int dynamic_values_bytes=4096*sizeof(lispobj);  /* same for all threads */
  36 struct thread * volatile all_threads;
  37 extern struct interrupt_data * global_interrupt_data;
  38 extern int linux_no_threads_p;
  39
  40 #ifdef LISP_FEATURE_SB_THREAD
  41
  42 pthread_mutex_t all_threads_lock = PTHREAD_MUTEX_INITIALIZER;
  43
  44 /* When trying to get all_threads_lock one should make sure that
  45  * SIG_STOP_FOR_GC is not blocked. Else there would be a possible
  46  * deadlock: gc locks it, other thread blocks signals, gc sends stop
  47  * request to other thread and waits, other thread blocks on lock. */
  48 void check_sig_stop_for_gc_can_arrive_or_lose()
  49 {
  50     /* Get the current sigmask, by blocking the empty set. */
  51     sigset_t empty,current;
  52     sigemptyset(&empty);
  53     thread_sigmask(SIG_BLOCK, &empty, &current);
  54     if (sigismember(&current,SIG_STOP_FOR_GC))
  55         lose("SIG_STOP_FOR_GC cannot arrive: it is blocked\n");
  56     if (SymbolValue(GC_INHIBIT,arch_os_get_current_thread()) != NIL)
  57         lose("SIG_STOP_FOR_GC cannot arrive: gc is inhibited\n");
  58     if (arch_pseudo_atomic_atomic(NULL))
  59         lose("SIG_STOP_FOR_GC cannot arrive: in pseudo atomic\n");
  60 }
  61
  62 #define GET_ALL_THREADS_LOCK(name) \
  63     { \
  64         sigset_t _newset,_oldset; \
  65         sigemptyset(&_newset); \
  66         sigaddset_deferrable(&_newset); \
  67         thread_sigmask(SIG_BLOCK, &_newset, &_oldset); \
  68         check_sig_stop_for_gc_can_arrive_or_lose(); \
  69         FSHOW_SIGNAL((stderr,"/%s:waiting on lock=%ld, thread=%lu\n",name, \
  70                all_threads_lock,arch_os_get_current_thread()->os_thread)); \
  71         pthread_mutex_lock(&all_threads_lock); \
  72         FSHOW_SIGNAL((stderr,"/%s:got lock, thread=%lu\n", \
  73                name,arch_os_get_current_thread()->os_thread));
  74
  75 #define RELEASE_ALL_THREADS_LOCK(name) \
  76         FSHOW_SIGNAL((stderr,"/%s:released lock\n",name)); \
  77         pthread_mutex_unlock(&all_threads_lock); \
  78         thread_sigmask(SIG_SETMASK,&_oldset,0); \
  79     }
  80
  81 #if defined(LISP_FEATURE_X86) || defined(LISP_FEATURE_X86_64)
  82 extern lispobj call_into_lisp_first_time(lispobj fun, lispobj *args, int nargs);
  83 #endif
  84
  85 #endif
  86
  87 static void
  88 link_thread(struct thread *th)
  89 {
  90     if (all_threads) all_threads->prev=th;
  91     th->next=all_threads;
  92     th->prev=0;
  93     all_threads=th;
  94 }
  95
  96 #ifdef LISP_FEATURE_SB_THREAD
  97 static void
  98 unlink_thread(struct thread *th)
  99 {
 100     if (th->prev)
 101         th->prev->next = th->next;
 102     else
 103         all_threads = th->next;
 104     if (th->next)
 105         th->next->prev = th->prev;
 106 }
 107 #endif
 108
 109 static int
 110 initial_thread_trampoline(struct thread *th)
 111 {
 112     lispobj function;
 113 #if defined(LISP_FEATURE_X86) || defined(LISP_FEATURE_X86_64)
 114     lispobj *args = NULL;
 115 #endif
 116     function = th->no_tls_value_marker;
 117     th->no_tls_value_marker = NO_TLS_VALUE_MARKER_WIDETAG;
 118     if(arch_os_thread_init(th)==0) return 1;
 119     link_thread(th);
 120     th->os_thread=thread_self();
 121     protect_control_stack_guard_page(1);
 122     th->state = STATE_RUNNING;
 123
 124 #if defined(LISP_FEATURE_X86) || defined(LISP_FEATURE_X86_64)
 125     return call_into_lisp_first_time(function,args,0);
 126 #else
 127     return funcall0(function);
 128 #endif
 129 }
 130
 131 #define THREAD_STRUCT_SIZE (THREAD_CONTROL_STACK_SIZE + BINDING_STACK_SIZE + \
 132                             ALIEN_STACK_SIZE + dynamic_values_bytes + \
 133                             32 * SIGSTKSZ)
 134
 135 #ifdef LISP_FEATURE_SB_THREAD
 136
 137 static void
 138 free_thread_stack_later(struct thread *thread_to_be_cleaned_up)
 139 {
 140     struct freeable_stack *new_freeable_stack = 0;
 141     if (thread_to_be_cleaned_up) {
 142         new_freeable_stack = (struct freeable_stack *)
 143             os_validate(0, sizeof(struct freeable_stack));
 144         new_freeable_stack->os_thread = thread_to_be_cleaned_up->os_thread;
 145         new_freeable_stack->stack = (os_vm_address_t)
 146             thread_to_be_cleaned_up->control_stack_start;
 147     }
 148     new_freeable_stack = (struct freeable_stack *)
 149         swap_lispobjs((lispobj *)(void *)&freeable_stack,
 150                       (lispobj)new_freeable_stack);
 151     if (new_freeable_stack) {
 152         FSHOW((stderr,"/reaping %lu\n", new_freeable_stack->os_thread));
 153         /* Under NPTL pthread_join really waits until the thread
 154          * exists and the stack can be safely freed. This is sadly not
 155          * mandated by the pthread spec. */
 156         gc_assert(pthread_join(new_freeable_stack->os_thread, NULL) == 0);
 157         os_invalidate(new_freeable_stack->stack, THREAD_STRUCT_SIZE);
 158         os_invalidate((os_vm_address_t) new_freeable_stack,
 159                       sizeof(struct freeable_stack));
 160     }
 161 }
 162
 163 /* this is the first thing that runs in the child (which is why the
 164  * silly calling convention).  Basically it calls the user's requested
 165  * lisp function after doing arch_os_thread_init and whatever other
 166  * bookkeeping needs to be done
 167  */
 168 int
 169 new_thread_trampoline(struct thread *th)
 170 {
 171     lispobj function;
 172     int result;
 173     FSHOW((stderr,"/creating thread %lu\n", thread_self()));
 174     function = th->no_tls_value_marker;
 175     th->no_tls_value_marker = NO_TLS_VALUE_MARKER_WIDETAG;
 176     if(arch_os_thread_init(th)==0) {
 177         /* FIXME: handle error */
 178         lose("arch_os_thread_init failed\n");
 179     }
 180
 181     th->os_thread=thread_self();
 182     protect_control_stack_guard_page(1);
 183     /* This thread is in STATE_STARTING so the GC is not sending it
 184      * SIG_STOP_FOR_GC => no danger of deadlocking even with
 185      * SIG_STOP_FOR_GC blocked. The lock is acquired in order not to
 186      * enter running state with the gc running. */
 187     pthread_mutex_lock(&all_threads_lock);
 188     th->state = STATE_RUNNING;
 189     pthread_mutex_unlock(&all_threads_lock);
 190     /* Now that we entered STATE_RUNNING let the gc suspend this
 191      * thread. */
 192     {
 193         sigset_t sigset;
 194         sigemptyset(&sigset);
 195         sigaddset(&sigset, SIG_STOP_FOR_GC);
 196         thread_sigmask(SIG_UNBLOCK, &sigset, 0);
 197     }
 198
 199     result = funcall0(function);
 200     th->state=STATE_DEAD;
 201
 202     /* SIG_STOP_FOR_GC is blocked and GC might be waiting for this
 203      * thread, but since we are already dead it won't wait long. */
 204     pthread_mutex_lock(&all_threads_lock);
 205     unlink_thread(th);
 206     pthread_mutex_unlock(&all_threads_lock);
 207
 208     gc_alloc_update_page_tables(0, &th->alloc_region);
 209     if(th->tls_cookie>=0) arch_os_thread_cleanup(th);
 210     os_invalidate((os_vm_address_t)th->interrupt_data,
 211                   (sizeof (struct interrupt_data)));
 212     free_thread_stack_later(th);
 213     FSHOW((stderr,"/exiting thread %lu\n", thread_self()));
 214     return result;
 215 }
 216
 217 #endif /* LISP_FEATURE_SB_THREAD */
 218
 219 static void
 220 free_thread_struct(struct thread *th)
 221 {
 222     if (th->interrupt_data)
 223         os_invalidate((os_vm_address_t) th->interrupt_data,
 224                       (sizeof (struct interrupt_data)));
 225     os_invalidate((os_vm_address_t) th->control_stack_start,
 226                   THREAD_STRUCT_SIZE);
 227 }
 228
 229 /* this is called from any other thread to create the new one, and
 230  * initialize all parts of it that can be initialized from another
 231  * thread
 232  */
 233
 234 static struct thread *
 235 create_thread_struct(lispobj initial_function) {
 236     union per_thread_data *per_thread;
 237     struct thread *th=0;        /*  subdue gcc */
 238     void *spaces=0;
 239
 240     /* may as well allocate all the spaces at once: it saves us from
 241      * having to decide what to do if only some of the allocations
 242      * succeed */
 243     spaces=os_validate(0, THREAD_STRUCT_SIZE);
 244     if(!spaces)
 245          return NULL;
 246     per_thread=(union per_thread_data *)
 247         (spaces+
 248          THREAD_CONTROL_STACK_SIZE+
 249          BINDING_STACK_SIZE+
 250          ALIEN_STACK_SIZE);
 251
 252     if(all_threads) {
 253         memcpy(per_thread,arch_os_get_current_thread(),
 254                dynamic_values_bytes);
 255     } else {
 256 #ifdef LISP_FEATURE_SB_THREAD
 257         int i;
 258         for(i=0;i<(dynamic_values_bytes/sizeof(lispobj));i++)
 259             per_thread->dynamic_values[i]=NO_TLS_VALUE_MARKER_WIDETAG;
 260         if(SymbolValue(FREE_TLS_INDEX,0)==UNBOUND_MARKER_WIDETAG) {
 261             SetSymbolValue
 262                 (FREE_TLS_INDEX,
 263                  make_fixnum(MAX_INTERRUPTS+
 264                              sizeof(struct thread)/sizeof(lispobj)),
 265                  0);
 266             SetSymbolValue(TLS_INDEX_LOCK,make_fixnum(0),0);
 267         }
 268 #define STATIC_TLS_INIT(sym,field) \
 269   ((struct symbol *)(sym-OTHER_POINTER_LOWTAG))->tls_index= \
 270   make_fixnum(THREAD_SLOT_OFFSET_WORDS(field))
 271
 272         STATIC_TLS_INIT(BINDING_STACK_START,binding_stack_start);
 273         STATIC_TLS_INIT(BINDING_STACK_POINTER,binding_stack_pointer);
 274         STATIC_TLS_INIT(CONTROL_STACK_START,control_stack_start);
 275         STATIC_TLS_INIT(CONTROL_STACK_END,control_stack_end);
 276         STATIC_TLS_INIT(ALIEN_STACK,alien_stack_pointer);
 277 #if defined(LISP_FEATURE_X86) || defined (LISP_FEATURE_X86_64)
 278         STATIC_TLS_INIT(PSEUDO_ATOMIC_ATOMIC,pseudo_atomic_atomic);
 279         STATIC_TLS_INIT(PSEUDO_ATOMIC_INTERRUPTED,pseudo_atomic_interrupted);
 280 #endif
 281 #undef STATIC_TLS_INIT
 282 #endif
 283     }
 284
 285     th=&per_thread->thread;
 286     th->control_stack_start = spaces;
 287     th->binding_stack_start=
 288         (lispobj*)((void*)th->control_stack_start+THREAD_CONTROL_STACK_SIZE);
 289     th->control_stack_end = th->binding_stack_start;
 290     th->alien_stack_start=
 291         (lispobj*)((void*)th->binding_stack_start+BINDING_STACK_SIZE);
 292     th->binding_stack_pointer=th->binding_stack_start;
 293     th->this=th;
 294     th->os_thread=0;
 295     th->state=STATE_STARTING;
 296 #ifdef LISP_FEATURE_STACK_GROWS_DOWNWARD_NOT_UPWARD
 297     th->alien_stack_pointer=((void *)th->alien_stack_start
 298                              + ALIEN_STACK_SIZE-N_WORD_BYTES);
 299 #else
 300     th->alien_stack_pointer=((void *)th->alien_stack_start);
 301 #endif
 302 #if defined(LISP_FEATURE_X86) || defined (LISP_FEATURE_X86_64)
 303     th->pseudo_atomic_interrupted=0;
 304     th->pseudo_atomic_atomic=0;
 305 #endif
 306 #ifdef LISP_FEATURE_GENCGC
 307     gc_set_region_empty(&th->alloc_region);
 308 #endif
 309
 310 #ifndef LISP_FEATURE_SB_THREAD
 311     /* the tls-points-into-struct-thread trick is only good for threaded
 312      * sbcl, because unithread sbcl doesn't have tls.  So, we copy the
 313      * appropriate values from struct thread here, and make sure that
 314      * we use the appropriate SymbolValue macros to access any of the
 315      * variable quantities from the C runtime.  It's not quite OAOOM,
 316      * it just feels like it */
 317     SetSymbolValue(BINDING_STACK_START,(lispobj)th->binding_stack_start,th);
 318     SetSymbolValue(CONTROL_STACK_START,(lispobj)th->control_stack_start,th);
 319     SetSymbolValue(CONTROL_STACK_END,(lispobj)th->control_stack_end,th);
 320 #if defined(LISP_FEATURE_X86) || defined (LISP_FEATURE_X86_64)
 321     SetSymbolValue(BINDING_STACK_POINTER,(lispobj)th->binding_stack_pointer,th);
 322     SetSymbolValue(ALIEN_STACK,(lispobj)th->alien_stack_pointer,th);
 323     SetSymbolValue(PSEUDO_ATOMIC_ATOMIC,(lispobj)th->pseudo_atomic_atomic,th);
 324     SetSymbolValue(PSEUDO_ATOMIC_INTERRUPTED,th->pseudo_atomic_interrupted,th);
 325 #else
 326     current_binding_stack_pointer=th->binding_stack_pointer;
 327     current_control_stack_pointer=th->control_stack_start;
 328 #endif
 329 #endif
 330     bind_variable(CURRENT_CATCH_BLOCK,make_fixnum(0),th);
 331     bind_variable(CURRENT_UNWIND_PROTECT_BLOCK,make_fixnum(0),th);
 332     bind_variable(FREE_INTERRUPT_CONTEXT_INDEX,make_fixnum(0),th);
 333     bind_variable(INTERRUPT_PENDING, NIL,th);
 334     bind_variable(INTERRUPTS_ENABLED,T,th);
 335     bind_variable(GC_PENDING,NIL,th);
 336 #ifdef LISP_FEATURE_SB_THREAD
 337     bind_variable(STOP_FOR_GC_PENDING,NIL,th);
 338 #endif
 339
 340     th->interrupt_data = (struct interrupt_data *)
 341         os_validate(0,(sizeof (struct interrupt_data)));
 342     if (!th->interrupt_data) {
 343         free_thread_struct(th);
 344         return 0;
 345     }
 346     th->interrupt_data->pending_handler = 0;
 347     th->no_tls_value_marker=initial_function;
 348     return th;
 349 }
 350
 351 void create_initial_thread(lispobj initial_function) {
 352     struct thread *th=create_thread_struct(initial_function);
 353     if(th) {
 354         initial_thread_trampoline(th); /* no return */
 355     } else lose("can't create initial thread");
 356 }
 357
 358 #ifdef LISP_FEATURE_SB_THREAD
 359
 360 #ifndef __USE_XOPEN2K
 361 extern int pthread_attr_setstack (pthread_attr_t *__attr, void *__stackaddr,
 362                                   size_t __stacksize);
 363 #endif
 364
 365 boolean create_os_thread(struct thread *th,os_thread_t *kid_tid)
 366 {
 367     /* The new thread inherits the restrictive signal mask set here,
 368      * and enables signals again when it is set up properly. */
 369     pthread_attr_t attr;
 370     sigset_t newset,oldset;
 371     boolean r=1;
 372     sigemptyset(&newset);
 373     sigaddset_blockable(&newset);
 374     thread_sigmask(SIG_BLOCK, &newset, &oldset);
 375
 376     if((pthread_attr_init(&attr)) ||
 377        (pthread_attr_setstack(&attr,th->control_stack_start,
 378                               THREAD_CONTROL_STACK_SIZE-16)) ||
 379        (pthread_create
 380         (kid_tid,&attr,(void *(*)(void *))new_thread_trampoline,th)))
 381         r=0;
 382     thread_sigmask(SIG_SETMASK,&oldset,0);
 383     return r;
 384 }
 385
 386 os_thread_t create_thread(lispobj initial_function) {
 387     struct thread *th;
 388     os_thread_t kid_tid;
 389
 390     if(linux_no_threads_p) return 0;
 391
 392     /* The new thread must be linked immediately onto all_threads for
 393      * gc. */
 394     GET_ALL_THREADS_LOCK("create_thread")
 395     /* If it is too slow most of the allocation/initialization can
 396      * be done without the lock. */
 397     th = create_thread_struct(initial_function);
 398     if (th)
 399         link_thread(th);
 400     RELEASE_ALL_THREADS_LOCK("create_thread")
 401     if(th==0) return 0;
 402
 403     if (create_os_thread(th,&kid_tid)) {
 404         return kid_tid;
 405     } else {
 406         GET_ALL_THREADS_LOCK("create_thread")
 407         unlink_thread(th);
 408         RELEASE_ALL_THREADS_LOCK("create_thread")
 409         free_thread_struct(th);
 410         return 0;
 411     }
 412 }
 413
 414 /* Send the signo to os_thread, retry if the rt signal queue is
 415  * full. */
 416 static int kill_thread_safely(os_thread_t os_thread, int signo)
 417 {
 418     int r;
 419     /* The man page does not mention EAGAIN as a valid return value
 420      * for either pthread_kill or kill. But that's theory, this is
 421      * practice. By waiting here we assume that the delivery of this
 422      * signal is not necessary for the delivery of the signals in the
 423      * queue. In other words, we _assume_ there are no deadlocks. */
 424     while ((r=pthread_kill(os_thread,signo))==EAGAIN) {
 425         /* wait a bit then try again in the hope of the rt signal
 426          * queue not being full */
 427         FSHOW_SIGNAL((stderr,"/rt signal queue full\n"));
 428         /* FIXME: some kind of backoff (random, exponential) would be
 429          * nice. */
 430         sleep(1);
 431     }
 432     return r;
 433 }
 434
 435 int signal_interrupt_thread(os_thread_t os_thread)
 436 {
 437     int status = kill_thread_safely(os_thread, SIG_INTERRUPT_THREAD);
 438     if (status == 0) {
 439         return 0;
 440     } else if (status == ESRCH) {
 441         return -1;
 442     } else {
 443         lose("cannot send SIG_INTERRUPT_THREAD to thread=%lu: %d, %s",
 444              os_thread, status, strerror(status));
 445     }
 446 }
 447
 448 /* stopping the world is a two-stage process.  From this thread we signal
 449  * all the others with SIG_STOP_FOR_GC.  The handler for this signal does
 450  * the usual pseudo-atomic checks (we don't want to stop a thread while
 451  * it's in the middle of allocation) then waits for another SIG_STOP_FOR_GC.
 452  */
 453
 454 /* To avoid deadlocks when gc stops the world all clients of each
 455  * mutex must enable or disable SIG_STOP_FOR_GC for the duration of
 456  * holding the lock, but they must agree on which. */
 457 void gc_stop_the_world()
 458 {
 459     struct thread *p,*th=arch_os_get_current_thread();
 460     int status;
 461     FSHOW_SIGNAL((stderr,"/gc_stop_the_world:waiting on lock, thread=%lu\n",
 462                   th->os_thread));
 463     /* keep threads from starting while the world is stopped. */
 464     pthread_mutex_lock(&all_threads_lock); \
 465     FSHOW_SIGNAL((stderr,"/gc_stop_the_world:got lock, thread=%lu\n",
 466                   th->os_thread));
 467     /* stop all other threads by sending them SIG_STOP_FOR_GC */
 468     for(p=all_threads; p; p=p->next) {
 469         if((p!=th) && ((p->state==STATE_RUNNING))) {
 470             FSHOW_SIGNAL((stderr,"/gc_stop_the_world: suspending %lu\n",
 471                           p->os_thread));
 472             status=kill_thread_safely(p->os_thread,SIG_STOP_FOR_GC);
 473             if (status==ESRCH) {
 474                 /* This thread has exited. */
 475                 gc_assert(p->state==STATE_DEAD);
 476             } else if (status) {
 477                 lose("cannot send suspend thread=%lu: %d, %s",
 478                      p->os_thread,status,strerror(status));
 479             }
 480         }
 481     }
 482     FSHOW_SIGNAL((stderr,"/gc_stop_the_world:signals sent\n"));
 483     /* wait for the running threads to stop or finish */
 484     for(p=all_threads;p;) {
 485         if((p!=th) && (p->state==STATE_RUNNING)) {
 486             gc_assert(p->os_thread!=0);
 487             sched_yield();
 488         } else {
 489             p=p->next;
 490         }
 491     }
 492     FSHOW_SIGNAL((stderr,"/gc_stop_the_world:end\n"));
 493 }
 494
 495 void gc_start_the_world()
 496 {
 497     struct thread *p,*th=arch_os_get_current_thread();
 498     int status;
 499     /* if a resumed thread creates a new thread before we're done with
 500      * this loop, the new thread will get consed on the front of
 501      * all_threads, but it won't have been stopped so won't need
 502      * restarting */
 503     FSHOW_SIGNAL((stderr,"/gc_start_the_world:begin\n"));
 504     for(p=all_threads;p;p=p->next) {
 505         if((p!=th) && (p->state!=STATE_STARTING) && (p->state!=STATE_DEAD)) {
 506             gc_assert(p->os_thread!=0);
 507             if(p->state!=STATE_SUSPENDED) {
 508                 lose("gc_start_the_world: wrong thread state is %d\n",
 509                      fixnum_value(p->state));
 510             }
 511             FSHOW_SIGNAL((stderr, "/gc_start_the_world: resuming %lu\n",
 512                           p->os_thread));
 513             p->state=STATE_RUNNING;
 514             status=kill_thread_safely(p->os_thread,SIG_STOP_FOR_GC);
 515             if (status) {
 516                 lose("cannot resume thread=%lu: %d, %s",
 517                      p->os_thread,status,strerror(status));
 518             }
 519         }
 520     }
 521     /* If we waited here until all threads leave STATE_SUSPENDED, then
 522      * SIG_STOP_FOR_GC wouldn't need to be a rt signal. That has some
 523      * performance implications, but does away with the 'rt signal
 524      * queue full' problem. */
 525     pthread_mutex_unlock(&all_threads_lock); \
 526     FSHOW_SIGNAL((stderr,"/gc_start_the_world:end\n"));
 527 }
 528 #endif