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 "thread.h"
  15 #include "arch.h"
  16 #include "target-arch-os.h"
  17 #include "os.h"
  18 #include "globals.h"
  19 #include "dynbind.h"
  20 #include "genesis/cons.h"
  21 #include "genesis/fdefn.h"
  22 #include "interr.h"             /* for lose() */
  23 #include "gc-internal.h"
  24
  25 #define ALIEN_STACK_SIZE (1*1024*1024) /* 1Mb size chosen at random */
  26
  27 int dynamic_values_bytes=4096*sizeof(lispobj);  /* same for all threads */
  28 struct thread *all_threads;
  29 volatile lispobj all_threads_lock;
  30 extern struct interrupt_data * global_interrupt_data;
  31
  32 int
  33 initial_thread_trampoline(struct thread *th)
  34 {
  35     lispobj function;
  36     lispobj *args = NULL;
  37     function = th->unbound_marker;
  38     th->unbound_marker = UNBOUND_MARKER_WIDETAG;
  39     if(arch_os_thread_init(th)==0) return 1;
  40
  41     if(th->pid < 1) lose("th->pid not set up right");
  42     th->state=STATE_RUNNING;
  43 #if defined(LISP_FEATURE_X86)
  44     return call_into_lisp_first_time(function,args,0);
  45 #else
  46     return funcall0(function);
  47 #endif
  48 }
  49
  50 /* this is the first thing that clone() runs in the child (which is
  51  * why the silly calling convention).  Basically it calls the user's
  52  * requested lisp function after doing arch_os_thread_init and
  53  * whatever other bookkeeping needs to be done
  54  */
  55
  56 #ifdef LISP_FEATURE_SB_THREAD
  57 int
  58 new_thread_trampoline(struct thread *th)
  59 {
  60     lispobj function;
  61     function = th->unbound_marker;
  62     th->unbound_marker = UNBOUND_MARKER_WIDETAG;
  63     if(arch_os_thread_init(th)==0) return 1;
  64
  65     /* wait here until our thread is linked into all_threads: see below */
  66     while(th->pid<1) sched_yield();
  67
  68     th->state=STATE_RUNNING;
  69     return funcall0(function);
  70 }
  71 #endif /* LISP_FEATURE_SB_THREAD */
  72
  73 /* this is called from any other thread to create the new one, and
  74  * initialize all parts of it that can be initialized from another
  75  * thread
  76  */
  77
  78 struct thread * create_thread_struct(lispobj initial_function) {
  79     union per_thread_data *per_thread;
  80     struct thread *th=0;        /*  subdue gcc */
  81     void *spaces=0;
  82
  83     /* may as well allocate all the spaces at once: it saves us from
  84      * having to decide what to do if only some of the allocations
  85      * succeed */
  86     spaces=os_validate(0,
  87                        THREAD_CONTROL_STACK_SIZE+
  88                        BINDING_STACK_SIZE+
  89                        ALIEN_STACK_SIZE+
  90                        dynamic_values_bytes+
  91                        32*SIGSTKSZ
  92                        );
  93     if(!spaces) goto cleanup;
  94     per_thread=(union per_thread_data *)
  95         (spaces+
  96          THREAD_CONTROL_STACK_SIZE+
  97          BINDING_STACK_SIZE+
  98          ALIEN_STACK_SIZE);
  99
 100     th=&per_thread->thread;
 101     if(all_threads) {
 102         memcpy(per_thread,arch_os_get_current_thread(),
 103                dynamic_values_bytes);
 104     } else {
 105 #ifdef LISP_FEATURE_SB_THREAD
 106         int i;
 107         for(i=0;i<(dynamic_values_bytes/sizeof(lispobj));i++)
 108             per_thread->dynamic_values[i]=UNBOUND_MARKER_WIDETAG;
 109         if(SymbolValue(FREE_TLS_INDEX,0)==UNBOUND_MARKER_WIDETAG)
 110             SetSymbolValue
 111                 (FREE_TLS_INDEX,
 112                  make_fixnum(MAX_INTERRUPTS+
 113                              sizeof(struct thread)/sizeof(lispobj)),
 114                  0);
 115 #define STATIC_TLS_INIT(sym,field) \
 116   ((struct symbol *)(sym-OTHER_POINTER_LOWTAG))->tls_index= \
 117   make_fixnum(THREAD_SLOT_OFFSET_WORDS(field))
 118
 119         STATIC_TLS_INIT(BINDING_STACK_START,binding_stack_start);
 120         STATIC_TLS_INIT(BINDING_STACK_POINTER,binding_stack_pointer);
 121         STATIC_TLS_INIT(CONTROL_STACK_START,control_stack_start);
 122         STATIC_TLS_INIT(CONTROL_STACK_END,control_stack_end);
 123         STATIC_TLS_INIT(ALIEN_STACK,alien_stack_pointer);
 124 #ifdef LISP_FEATURE_X86
 125         STATIC_TLS_INIT(PSEUDO_ATOMIC_ATOMIC,pseudo_atomic_atomic);
 126         STATIC_TLS_INIT(PSEUDO_ATOMIC_INTERRUPTED,pseudo_atomic_interrupted);
 127 #endif
 128 #undef STATIC_TLS_INIT
 129 #endif
 130     }
 131
 132     th->control_stack_start = spaces;
 133     th->binding_stack_start=
 134         (lispobj*)((void*)th->control_stack_start+THREAD_CONTROL_STACK_SIZE);
 135     th->control_stack_end = th->binding_stack_start;
 136     th->alien_stack_start=
 137         (lispobj*)((void*)th->binding_stack_start+BINDING_STACK_SIZE);
 138     th->binding_stack_pointer=th->binding_stack_start;
 139     th->this=th;
 140     th->pid=0;
 141     th->state=STATE_STOPPED;
 142 #ifdef LISP_FEATURE_STACK_GROWS_DOWNWARD_NOT_UPWARD
 143     th->alien_stack_pointer=((void *)th->alien_stack_start
 144                              + ALIEN_STACK_SIZE-4); /* naked 4.  FIXME */
 145 #else
 146     th->alien_stack_pointer=((void *)th->alien_stack_start);
 147 #endif
 148 #ifdef LISP_FEATURE_X86
 149     th->pseudo_atomic_interrupted=0;
 150     th->pseudo_atomic_atomic=0;
 151 #endif
 152 #ifdef LISP_FEATURE_GENCGC
 153     gc_set_region_empty(&th->alloc_region);
 154 #endif
 155
 156 #ifndef LISP_FEATURE_SB_THREAD
 157     /* the tls-points-into-struct-thread trick is only good for threaded
 158      * sbcl, because unithread sbcl doesn't have tls.  So, we copy the
 159      * appropriate values from struct thread here, and make sure that
 160      * we use the appropriate SymbolValue macros to access any of the
 161      * variable quantities from the C runtime.  It's not quite OAOOM,
 162      * it just feels like it */
 163     SetSymbolValue(BINDING_STACK_START,(lispobj)th->binding_stack_start,th);
 164     SetSymbolValue(CONTROL_STACK_START,(lispobj)th->control_stack_start,th);
 165     SetSymbolValue(CONTROL_STACK_END,(lispobj)th->control_stack_end,th);
 166 #ifdef LISP_FEATURE_X86
 167     SetSymbolValue(BINDING_STACK_POINTER,(lispobj)th->binding_stack_pointer,th);
 168     SetSymbolValue(ALIEN_STACK,(lispobj)th->alien_stack_pointer,th);
 169     SetSymbolValue(PSEUDO_ATOMIC_ATOMIC,(lispobj)th->pseudo_atomic_atomic,th);
 170     SetSymbolValue(PSEUDO_ATOMIC_INTERRUPTED,th->pseudo_atomic_interrupted,th);
 171 #else
 172     current_binding_stack_pointer=th->binding_stack_pointer;
 173     current_control_stack_pointer=th->control_stack_start;
 174 #endif
 175 #endif
 176     bind_variable(CURRENT_CATCH_BLOCK,make_fixnum(0),th);
 177     bind_variable(CURRENT_UNWIND_PROTECT_BLOCK,make_fixnum(0),th);
 178     bind_variable(FREE_INTERRUPT_CONTEXT_INDEX,make_fixnum(0),th);
 179     bind_variable(INTERRUPT_PENDING, NIL,th);
 180     bind_variable(INTERRUPTS_ENABLED,T,th);
 181
 182     th->interrupt_data=os_validate(0,(sizeof (struct interrupt_data)));
 183     if(all_threads)
 184         memcpy(th->interrupt_data,
 185                arch_os_get_current_thread()->interrupt_data,
 186                sizeof (struct interrupt_data));
 187     else
 188         memcpy(th->interrupt_data,global_interrupt_data,
 189                sizeof (struct interrupt_data));
 190
 191     th->unbound_marker=initial_function;
 192     return th;
 193  cleanup:
 194     /* if(th && th->tls_cookie>=0) os_free_tls_pointer(th); */
 195     if(spaces) os_invalidate(spaces,
 196                              THREAD_CONTROL_STACK_SIZE+BINDING_STACK_SIZE+
 197                              ALIEN_STACK_SIZE+dynamic_values_bytes);
 198     return 0;
 199 }
 200
 201 void link_thread(struct thread *th,pid_t kid_pid)
 202 {
 203     sigset_t newset,oldset;
 204     sigemptyset(&newset);
 205     sigaddset_blockable(&newset);
 206     sigprocmask(SIG_BLOCK, &newset, &oldset);
 207
 208     get_spinlock(&all_threads_lock,kid_pid);
 209     th->next=all_threads;
 210     all_threads=th;
 211     /* note that th->pid is 0 at this time.  We rely on all_threads_lock
 212      * to ensure that we don't have >1 thread with pid=0 on the list at once
 213      */
 214     protect_control_stack_guard_page(th->pid,1);
 215     release_spinlock(&all_threads_lock);
 216
 217     sigprocmask(SIG_SETMASK,&oldset,0);
 218     th->pid=kid_pid;            /* child will not start until this is set */
 219 }
 220
 221 void create_initial_thread(lispobj initial_function) {
 222     struct thread *th=create_thread_struct(initial_function);
 223     pid_t kid_pid=getpid();
 224     if(th && kid_pid>0) {
 225         link_thread(th,kid_pid);
 226         initial_thread_trampoline(all_threads); /* no return */
 227     } else lose("can't create initial thread");
 228 }
 229
 230 #ifdef LISP_FEATURE_SB_THREAD
 231 pid_t create_thread(lispobj initial_function) {
 232     struct thread *th=create_thread_struct(initial_function);
 233     pid_t kid_pid=0;
 234
 235     if(th==0) return 0;
 236     kid_pid=clone(new_thread_trampoline,
 237                   (((void*)th->control_stack_start)+
 238                    THREAD_CONTROL_STACK_SIZE-4),
 239                   CLONE_FILES|SIG_THREAD_EXIT|CLONE_VM,th);
 240
 241     if(kid_pid>0) {
 242         link_thread(th,kid_pid);
 243         return th->pid;
 244     } else {
 245         os_invalidate((os_vm_address_t) th->control_stack_start,
 246                       ((sizeof (lispobj))
 247                        * (th->control_stack_end-th->control_stack_start)) +
 248                       BINDING_STACK_SIZE+ALIEN_STACK_SIZE+dynamic_values_bytes+
 249                       32*SIGSTKSZ);
 250         return 0;
 251     }
 252 }
 253 #endif
 254
 255 /* unused */
 256 void destroy_thread (struct thread *th)
 257 {
 258     /* precondition: the unix task has already been killed and exited.
 259      * This is called by the parent or some other thread */
 260 #ifdef LISP_FEATURE_GENCGC
 261     gc_alloc_update_page_tables(0, &th->alloc_region);
 262 #endif
 263     get_spinlock(&all_threads_lock,th->pid);
 264     th->unbound_marker=0;       /* for debugging */
 265     if(th==all_threads)
 266         all_threads=th->next;
 267     else {
 268         struct thread *th1=all_threads;
 269         while(th1 && th1->next!=th) th1=th1->next;
 270         if(th1) th1->next=th->next;     /* unlink */
 271     }
 272     release_spinlock(&all_threads_lock);
 273     if(th && th->tls_cookie>=0) arch_os_thread_cleanup(th);
 274     os_invalidate((os_vm_address_t) th->control_stack_start,
 275                   ((sizeof (lispobj))
 276                    * (th->control_stack_end-th->control_stack_start)) +
 277                   BINDING_STACK_SIZE+ALIEN_STACK_SIZE+dynamic_values_bytes+
 278                   32*SIGSTKSZ);
 279 }
 280
 281 struct thread *find_thread_by_pid(pid_t pid)
 282 {
 283     struct thread *th;
 284     for_each_thread(th)
 285         if(th->pid==pid) return th;
 286     return 0;
 287 }
 288
 289 #if defined LISP_FEATURE_SB_THREAD
 290 /* This is not needed unless #+SB-THREAD, as there's a trivial null
 291  * unithread definition. */
 292
 293 void mark_dead_threads()
 294 {
 295     pid_t kid;
 296     int status;
 297     while(1) {
 298         kid=waitpid(-1,&status,__WALL|WNOHANG);
 299         if(kid<=0) break;
 300         if(WIFEXITED(status) || WIFSIGNALED(status)) {
 301             struct thread *th=find_thread_by_pid(kid);
 302             if(th) th->state=STATE_DEAD;
 303         }
 304     }
 305 }
 306
 307 void reap_dead_threads()
 308 {
 309     struct thread *th,*next,*prev=0;
 310     th=all_threads;
 311     while(th) {
 312         next=th->next;
 313         if(th->state==STATE_DEAD) {
 314             funcall1(SymbolFunction(HANDLE_THREAD_EXIT),make_fixnum(th->pid));
 315 #ifdef LISP_FEATURE_GENCGC
 316             gc_alloc_update_page_tables(0, &th->alloc_region);
 317 #endif
 318             get_spinlock(&all_threads_lock,th->pid);
 319             if(prev) prev->next=next;
 320             else all_threads=next;
 321             release_spinlock(&all_threads_lock);
 322             if(th->tls_cookie>=0) arch_os_thread_cleanup(th);
 323             os_invalidate((os_vm_address_t) th->control_stack_start,
 324                           ((sizeof (lispobj))
 325                            * (th->control_stack_end-th->control_stack_start)) +
 326                           BINDING_STACK_SIZE+ALIEN_STACK_SIZE+dynamic_values_bytes+
 327                           32*SIGSTKSZ);
 328         } else
 329             prev=th;
 330         th=next;
 331     }
 332 }
 333
 334 /* These are not needed unless #+SB-THREAD, and since sigwaitinfo()
 335  * doesn't seem to be easily available everywhere (OpenBSD...) it's
 336  * more trouble than it's worth to compile it when not needed. */
 337 void block_sigcont(void)
 338 {
 339     /* don't allow ourselves to receive SIGCONT while we're in the
 340      * "ambiguous" state of being on the queue but not actually stopped.
 341      */
 342     sigset_t newset;
 343     sigemptyset(&newset);
 344     sigaddset(&newset,SIG_DEQUEUE);
 345     sigprocmask(SIG_BLOCK, &newset, 0);
 346 }
 347
 348 void unblock_sigcont_and_sleep(void)
 349 {
 350     sigset_t set;
 351     sigemptyset(&set);
 352     sigaddset(&set,SIG_DEQUEUE);
 353     do {
 354         errno=0;
 355         sigwaitinfo(&set,0);
 356     }while(errno==EINTR);
 357     sigprocmask(SIG_UNBLOCK,&set,0);
 358 }
 359
 360 int interrupt_thread(pid_t pid, lispobj function)
 361 {
 362     union sigval sigval;
 363     struct thread *th;
 364     sigval.sival_int=function;
 365     for_each_thread(th)
 366         if((th->pid==pid) && (th->state != STATE_DEAD))
 367             return sigqueue(pid, SIG_INTERRUPT_THREAD, sigval);
 368     errno=EPERM; return -1;
 369 }
 370
 371 int signal_thread_to_dequeue (pid_t pid)
 372 {
 373     return kill (pid, SIG_DEQUEUE);
 374 }
 375
 376
 377 /* stopping the world is a two-stage process.  From this thread we signal
 378  * all the others with SIG_STOP_FOR_GC.  The handler for this signal does
 379  * the usual pseudo-atomic checks (we don't want to stop a thread while
 380  * it's in the middle of allocation) then kills _itself_ with SIGSTOP.
 381  */
 382
 383 void gc_stop_the_world()
 384 {
 385     /* stop all other threads by sending them SIG_STOP_FOR_GC */
 386     struct thread *p,*th=arch_os_get_current_thread();
 387     pid_t old_pid;
 388     int finished;
 389     do {
 390         finished=1;
 391         for(p=all_threads,old_pid=p->pid; p; p=p->next) {
 392             if(p==th) continue;
 393             if(p->state==STATE_RUNNING) {
 394                 p->state=STATE_STOPPING;
 395                 if(kill(p->pid,SIG_STOP_FOR_GC)==-1) {
 396                     /* we can't kill the process; assume because it
 397                      * died already (and its parent is dead so never
 398                      * saw the SIGCHLD) */
 399                     p->state=STATE_DEAD;
 400                 }
 401             }
 402             if((p->state!=STATE_STOPPED) &&
 403                (p->state!=STATE_DEAD)) {
 404                 finished=0;
 405             }
 406         }
 407         if(old_pid!=all_threads->pid) {
 408             finished=0;
 409         }
 410     } while(!finished);
 411 }
 412
 413 void gc_start_the_world()
 414 {
 415     struct thread *p,*th=arch_os_get_current_thread();
 416     /* if a resumed thread creates a new thread before we're done with
 417      * this loop, the new thread will get consed on the front of *
 418      * all_threads_lock, but it won't have been stopped so won't need
 419      * restarting */
 420     for(p=all_threads;p;p=p->next) {
 421         if((p==th) || (p->state==STATE_DEAD)) continue;
 422         p->state=STATE_RUNNING;
 423         kill(p->pid,SIG_STOP_FOR_GC);
 424     }
 425 }
 426 #endif