1 /* This file contains essentially all of the process and message handling. 2 * Together with "mpx.s" it forms the lowest layer of the MINIX kernel. 3 * There is one entry point from the outside: 4 * 5 * sys_call: a system call, i.e., the kernel is trapped with an INT 6 * 7 * Changes: 8 * Aug 19, 2005 rewrote scheduling code (Jorrit N. Herder) 9 * Jul 25, 2005 rewrote system call handling (Jorrit N. Herder) 10 * May 26, 2005 rewrote message passing functions (Jorrit N. Herder) 11 * May 24, 2005 new notification system call (Jorrit N. Herder) 12 * Oct 28, 2004 nonblocking send and receive calls (Jorrit N. Herder) 13 * 14 * The code here is critical to make everything work and is important for the 15 * overall performance of the system. A large fraction of the code deals with 16 * list manipulation. To make this both easy to understand and fast to execute 17 * pointer pointers are used throughout the code. Pointer pointers prevent 18 * exceptions for the head or tail of a linked list. 19 * 20 * node_t *queue, *new_node; // assume these as global variables 21 * node_t **xpp = &queue; // get pointer pointer to head of queue 22 * while (*xpp != NULL) // find last pointer of the linked list 23 * xpp = &(*xpp)->next; // get pointer to next pointer 24 * *xpp = new_node; // now replace the end (the NULL pointer) 25 * new_node->next = NULL; // and mark the new end of the list 26 * 27 * For example, when adding a new node to the end of the list, one normally 28 * makes an exception for an empty list and looks up the end of the list for 29 * nonempty lists. As shown above, this is not required with pointer pointers. 30 */ 31 32 #include <minix/com.h> 33 #include <minix/ipcconst.h> 34 #include <stddef.h> 35 #include <signal.h> 36 #include <assert.h> 37 #include <string.h> 38 39 #include "kernel/kernel.h" 40 #include "vm.h" 41 #include "clock.h" 42 #include "spinlock.h" 43 #include "arch_proto.h" 44 45 #include <minix/syslib.h> 46 47 /* Scheduling and message passing functions */ 48 static void idle(void); 49 /** 50 * Made public for use in clock.c (for user-space scheduling) 51 static int mini_send(struct proc *caller_ptr, endpoint_t dst_e, message 52 *m_ptr, int flags); 53 */ 54 static int mini_receive(struct proc *caller_ptr, endpoint_t src, 55 message *m_buff_usr, int flags); 56 static int mini_senda(struct proc *caller_ptr, asynmsg_t *table, size_t 57 size); 58 static int deadlock(int function, register struct proc *caller, 59 endpoint_t src_dst_e); 60 static int try_async(struct proc *caller_ptr); 61 static int try_one(endpoint_t receive_e, struct proc *src_ptr, 62 struct proc *dst_ptr); 63 static struct proc * pick_proc(void); 64 static void enqueue_head(struct proc *rp); 65 66 /* all idles share the same idle_priv structure */ 67 static struct priv idle_priv; 68 69 static void set_idle_name(char * name, int n) 70 { 71 int i, c; 72 int p_z = 0; 73 74 if (n > 999) 75 n = 999; 76 77 name[0] = 'i'; 78 name[1] = 'd'; 79 name[2] = 'l'; 80 name[3] = 'e'; 81 82 for (i = 4, c = 100; c > 0; c /= 10) { 83 int digit; 84 85 digit = n / c; 86 n -= digit * c; 87 88 if (p_z || digit != 0 || c == 1) { 89 p_z = 1; 90 name[i++] = '0' + digit; 91 } 92 } 93 94 name[i] = '\0'; 95 96 } 97 98 99 #define PICK_ANY 1 100 #define PICK_HIGHERONLY 2 101 102 #define BuildNotifyMessage(m_ptr, src, dst_ptr) \ 103 memset((m_ptr), 0, sizeof(*(m_ptr))); \ 104 (m_ptr)->m_type = NOTIFY_MESSAGE; \ 105 (m_ptr)->m_notify.timestamp = get_monotonic(); \ 106 switch (src) { \ 107 case HARDWARE: \ 108 (m_ptr)->m_notify.interrupts = \ 109 priv(dst_ptr)->s_int_pending; \ 110 priv(dst_ptr)->s_int_pending = 0; \ 111 break; \ 112 case SYSTEM: \ 113 memcpy(&(m_ptr)->m_notify.sigset, \ 114 &priv(dst_ptr)->s_sig_pending, \ 115 sizeof(sigset_t)); \ 116 sigemptyset(&priv(dst_ptr)->s_sig_pending); \ 117 break; \ 118 } 119 120 static message m_notify_buff = { 0, NOTIFY_MESSAGE }; 121 122 void proc_init(void) 123 { 124 struct proc * rp; 125 struct priv *sp; 126 int i; 127 128 /* Clear the process table. Announce each slot as empty and set up 129 * mappings for proc_addr() and proc_nr() macros. Do the same for the 130 * table with privilege structures for the system processes. 131 */ 132 for (rp = BEG_PROC_ADDR, i = -NR_TASKS; rp < END_PROC_ADDR; ++rp, ++i) { 133 rp->p_rts_flags = RTS_SLOT_FREE;/* initialize free slot */ 134 rp->p_magic = PMAGIC; 135 rp->p_nr = i; /* proc number from ptr */ 136 rp->p_endpoint = _ENDPOINT(0, rp->p_nr); /* generation no. 0 */ 137 rp->p_scheduler = NULL; /* no user space scheduler */ 138 rp->p_priority = 0; /* no priority */ 139 rp->p_quantum_size_ms = 0; /* no quantum size */ 140 141 /* arch-specific initialization */ 142 arch_proc_reset(rp); 143 } 144 for (sp = BEG_PRIV_ADDR, i = 0; sp < END_PRIV_ADDR; ++sp, ++i) { 145 sp->s_proc_nr = NONE; /* initialize as free */ 146 sp->s_id = (sys_id_t) i; /* priv structure index */ 147 ppriv_addr[i] = sp; /* priv ptr from number */ 148 sp->s_sig_mgr = NONE; /* clear signal managers */ 149 sp->s_bak_sig_mgr = NONE; 150 } 151 152 idle_priv.s_flags = IDL_F; 153 /* initialize IDLE structures for every CPU */ 154 for (i = 0; i < CONFIG_MAX_CPUS; i++) { 155 struct proc * ip = get_cpu_var_ptr(i, idle_proc); 156 ip->p_endpoint = IDLE; 157 ip->p_priv = &idle_priv; 158 /* must not let idle ever get scheduled */ 159 ip->p_rts_flags |= RTS_PROC_STOP; 160 set_idle_name(ip->p_name, i); 161 } 162 } 163 164 static void switch_address_space_idle(void) 165 { 166 #ifdef CONFIG_SMP 167 /* 168 * currently we bet that VM is always alive and its pages available so 169 * when the CPU wakes up the kernel is mapped and no surprises happen. 170 * This is only a problem if more than 1 cpus are available 171 */ 172 switch_address_space(proc_addr(VM_PROC_NR)); 173 #endif 174 } 175 176 /*===========================================================================* 177 * idle * 178 *===========================================================================*/ 179 static void idle(void) 180 { 181 struct proc * p; 182 183 /* This function is called whenever there is no work to do. 184 * Halt the CPU, and measure how many timestamp counter ticks are 185 * spent not doing anything. This allows test setups to measure 186 * the CPU utilization of certain workloads with high precision. 187 */ 188 189 p = get_cpulocal_var(proc_ptr) = get_cpulocal_var_ptr(idle_proc); 190 if (priv(p)->s_flags & BILLABLE) 191 get_cpulocal_var(bill_ptr) = p; 192 193 switch_address_space_idle(); 194 195 #ifdef CONFIG_SMP 196 get_cpulocal_var(cpu_is_idle) = 1; 197 /* we don't need to keep time on APs as it is handled on the BSP */ 198 if (cpuid != bsp_cpu_id) 199 stop_local_timer(); 200 else 201 #endif 202 { 203 /* 204 * If the timer has expired while in kernel we must 205 * rearm it before we go to sleep 206 */ 207 restart_local_timer(); 208 } 209 210 /* start accounting for the idle time */ 211 context_stop(proc_addr(KERNEL)); 212 #if !SPROFILE 213 halt_cpu(); 214 #else 215 if (!sprofiling) 216 halt_cpu(); 217 else { 218 volatile int * v; 219 220 v = get_cpulocal_var_ptr(idle_interrupted); 221 interrupts_enable(); 222 while (!*v) 223 arch_pause(); 224 interrupts_disable(); 225 *v = 0; 226 } 227 #endif 228 /* 229 * end of accounting for the idle task does not happen here, the kernel 230 * is handling stuff for quite a while before it gets back here! 231 */ 232 } 233 234 /*===========================================================================* 235 * vm_suspend * 236 *===========================================================================*/ 237 void vm_suspend(struct proc *caller, const struct proc *target, 238 const vir_bytes linaddr, const vir_bytes len, const int type, 239 const int writeflag) 240 { 241 /* This range is not OK for this process. Set parameters 242 * of the request and notify VM about the pending request. 243 */ 244 assert(!RTS_ISSET(caller, RTS_VMREQUEST)); 245 assert(!RTS_ISSET(target, RTS_VMREQUEST)); 246 247 RTS_SET(caller, RTS_VMREQUEST); 248 249 caller->p_vmrequest.req_type = VMPTYPE_CHECK; 250 caller->p_vmrequest.target = target->p_endpoint; 251 caller->p_vmrequest.params.check.start = linaddr; 252 caller->p_vmrequest.params.check.length = len; 253 caller->p_vmrequest.params.check.writeflag = writeflag; 254 caller->p_vmrequest.type = type; 255 256 /* Connect caller on vmrequest wait queue. */ 257 if(!(caller->p_vmrequest.nextrequestor = vmrequest)) 258 if(OK != send_sig(VM_PROC_NR, SIGKMEM)) 259 panic("send_sig failed"); 260 vmrequest = caller; 261 } 262 263 /*===========================================================================* 264 * delivermsg * 265 *===========================================================================*/ 266 static void delivermsg(struct proc *rp) 267 { 268 assert(!RTS_ISSET(rp, RTS_VMREQUEST)); 269 assert(rp->p_misc_flags & MF_DELIVERMSG); 270 assert(rp->p_delivermsg.m_source != NONE); 271 272 if (copy_msg_to_user(&rp->p_delivermsg, 273 (message *) rp->p_delivermsg_vir)) { 274 if(rp->p_misc_flags & MF_MSGFAILED) { 275 /* 2nd consecutive failure means this won't succeed */ 276 printf("WARNING wrong user pointer 0x%08lx from " 277 "process %s / %d\n", 278 rp->p_delivermsg_vir, 279 rp->p_name, 280 rp->p_endpoint); 281 cause_sig(rp->p_nr, SIGSEGV); 282 } else { 283 /* 1st failure means we have to ask VM to handle it */ 284 vm_suspend(rp, rp, rp->p_delivermsg_vir, 285 sizeof(message), VMSTYPE_DELIVERMSG, 1); 286 rp->p_misc_flags |= MF_MSGFAILED; 287 } 288 } else { 289 /* Indicate message has been delivered; address is 'used'. */ 290 rp->p_delivermsg.m_source = NONE; 291 rp->p_misc_flags &= ~(MF_DELIVERMSG|MF_MSGFAILED); 292 293 if(!(rp->p_misc_flags & MF_CONTEXT_SET)) { 294 rp->p_reg.retreg = OK; 295 } 296 } 297 } 298 299 /*===========================================================================* 300 * switch_to_user * 301 *===========================================================================*/ 302 void switch_to_user(void) 303 { 304 /* This function is called an instant before proc_ptr is 305 * to be scheduled again. 306 */ 307 struct proc * p; 308 #ifdef CONFIG_SMP 309 int tlb_must_refresh = 0; 310 #endif 311 312 p = get_cpulocal_var(proc_ptr); 313 /* 314 * if the current process is still runnable check the misc flags and let 315 * it run unless it becomes not runnable in the meantime 316 */ 317 if (proc_is_runnable(p)) 318 goto check_misc_flags; 319 /* 320 * if a process becomes not runnable while handling the misc flags, we 321 * need to pick a new one here and start from scratch. Also if the 322 * current process wasn't runnable, we pick a new one here 323 */ 324 not_runnable_pick_new: 325 if (proc_is_preempted(p)) { 326 p->p_rts_flags &= ~RTS_PREEMPTED; 327 if (proc_is_runnable(p)) { 328 if (p->p_cpu_time_left) 329 enqueue_head(p); 330 else 331 enqueue(p); 332 } 333 } 334 335 /* 336 * if we have no process to run, set IDLE as the current process for 337 * time accounting and put the cpu in an idle state. After the next 338 * timer interrupt the execution resumes here and we can pick another 339 * process. If there is still nothing runnable we "schedule" IDLE again 340 */ 341 while (!(p = pick_proc())) { 342 idle(); 343 } 344 345 /* update the global variable */ 346 get_cpulocal_var(proc_ptr) = p; 347 348 #ifdef CONFIG_SMP 349 if (p->p_misc_flags & MF_FLUSH_TLB && get_cpulocal_var(ptproc) == p) 350 tlb_must_refresh = 1; 351 #endif 352 switch_address_space(p); 353 354 check_misc_flags: 355 356 assert(p); 357 assert(proc_is_runnable(p)); 358 while (p->p_misc_flags & 359 (MF_KCALL_RESUME | MF_DELIVERMSG | 360 MF_SC_DEFER | MF_SC_TRACE | MF_SC_ACTIVE)) { 361 362 assert(proc_is_runnable(p)); 363 if (p->p_misc_flags & MF_KCALL_RESUME) { 364 kernel_call_resume(p); 365 } 366 else if (p->p_misc_flags & MF_DELIVERMSG) { 367 TRACE(VF_SCHEDULING, printf("delivering to %s / %d\n", 368 p->p_name, p->p_endpoint);); 369 delivermsg(p); 370 } 371 else if (p->p_misc_flags & MF_SC_DEFER) { 372 /* Perform the system call that we deferred earlier. */ 373 374 assert (!(p->p_misc_flags & MF_SC_ACTIVE)); 375 376 arch_do_syscall(p); 377 378 /* If the process is stopped for signal delivery, and 379 * not blocked sending a message after the system call, 380 * inform PM. 381 */ 382 if ((p->p_misc_flags & MF_SIG_DELAY) && 383 !RTS_ISSET(p, RTS_SENDING)) 384 sig_delay_done(p); 385 } 386 else if (p->p_misc_flags & MF_SC_TRACE) { 387 /* Trigger a system call leave event if this was a 388 * system call. We must do this after processing the 389 * other flags above, both for tracing correctness and 390 * to be able to use 'break'. 391 */ 392 if (!(p->p_misc_flags & MF_SC_ACTIVE)) 393 break; 394 395 p->p_misc_flags &= 396 ~(MF_SC_TRACE | MF_SC_ACTIVE); 397 398 /* Signal the "leave system call" event. 399 * Block the process. 400 */ 401 cause_sig(proc_nr(p), SIGTRAP); 402 } 403 else if (p->p_misc_flags & MF_SC_ACTIVE) { 404 /* If MF_SC_ACTIVE was set, remove it now: 405 * we're leaving the system call. 406 */ 407 p->p_misc_flags &= ~MF_SC_ACTIVE; 408 409 break; 410 } 411 412 /* 413 * the selected process might not be runnable anymore. We have 414 * to checkit and schedule another one 415 */ 416 if (!proc_is_runnable(p)) 417 goto not_runnable_pick_new; 418 } 419 /* 420 * check the quantum left before it runs again. We must do it only here 421 * as we are sure that a possible out-of-quantum message to the 422 * scheduler will not collide with the regular ipc 423 */ 424 if (!p->p_cpu_time_left) 425 proc_no_time(p); 426 /* 427 * After handling the misc flags the selected process might not be 428 * runnable anymore. We have to checkit and schedule another one 429 */ 430 if (!proc_is_runnable(p)) 431 goto not_runnable_pick_new; 432 433 TRACE(VF_SCHEDULING, printf("cpu %d starting %s / %d " 434 "pc 0x%08x\n", 435 cpuid, p->p_name, p->p_endpoint, p->p_reg.pc);); 436 #if DEBUG_TRACE 437 p->p_schedules++; 438 #endif 439 440 p = arch_finish_switch_to_user(); 441 assert(p->p_cpu_time_left); 442 443 context_stop(proc_addr(KERNEL)); 444 445 /* If the process isn't the owner of FPU, enable the FPU exception */ 446 if (get_cpulocal_var(fpu_owner) != p) 447 enable_fpu_exception(); 448 else 449 disable_fpu_exception(); 450 451 /* If MF_CONTEXT_SET is set, don't clobber process state within 452 * the kernel. The next kernel entry is OK again though. 453 */ 454 p->p_misc_flags &= ~MF_CONTEXT_SET; 455 456 #if defined(__i386__) 457 assert(p->p_seg.p_cr3 != 0); 458 #elif defined(__arm__) 459 assert(p->p_seg.p_ttbr != 0); 460 #endif 461 #ifdef CONFIG_SMP 462 if (p->p_misc_flags & MF_FLUSH_TLB) { 463 if (tlb_must_refresh) 464 refresh_tlb(); 465 p->p_misc_flags &= ~MF_FLUSH_TLB; 466 } 467 #endif 468 469 restart_local_timer(); 470 471 /* 472 * restore_user_context() carries out the actual mode switch from kernel 473 * to userspace. This function does not return 474 */ 475 restore_user_context(p); 476 NOT_REACHABLE; 477 } 478 479 /* 480 * handler for all synchronous IPC calls 481 */ 482 static int do_sync_ipc(struct proc * caller_ptr, /* who made the call */ 483 int call_nr, /* system call number and flags */ 484 endpoint_t src_dst_e, /* src or dst of the call */ 485 message *m_ptr) /* users pointer to a message */ 486 { 487 int result; /* the system call's result */ 488 int src_dst_p; /* Process slot number */ 489 char *callname; 490 491 /* Check destination. RECEIVE is the only call that accepts ANY (in addition 492 * to a real endpoint). The other calls (SEND, SENDREC, and NOTIFY) require an 493 * endpoint to corresponds to a process. In addition, it is necessary to check 494 * whether a process is allowed to send to a given destination. 495 */ 496 assert(call_nr != SENDA); 497 498 /* Only allow non-negative call_nr values less than 32 */ 499 if (call_nr < 0 || call_nr > IPCNO_HIGHEST || call_nr >= 32 500 || !(callname = ipc_call_names[call_nr])) { 501 #if DEBUG_ENABLE_IPC_WARNINGS 502 printf("sys_call: trap %d not allowed, caller %d, src_dst %d\n", 503 call_nr, proc_nr(caller_ptr), src_dst_e); 504 #endif 505 return(ETRAPDENIED); /* trap denied by mask or kernel */ 506 } 507 508 if (src_dst_e == ANY) 509 { 510 if (call_nr != RECEIVE) 511 { 512 #if 0 513 printf("sys_call: %s by %d with bad endpoint %d\n", 514 callname, 515 proc_nr(caller_ptr), src_dst_e); 516 #endif 517 return EINVAL; 518 } 519 src_dst_p = (int) src_dst_e; 520 } 521 else 522 { 523 /* Require a valid source and/or destination process. */ 524 if(!isokendpt(src_dst_e, &src_dst_p)) { 525 #if 0 526 printf("sys_call: %s by %d with bad endpoint %d\n", 527 callname, 528 proc_nr(caller_ptr), src_dst_e); 529 #endif 530 return EDEADSRCDST; 531 } 532 533 /* If the call is to send to a process, i.e., for SEND, SENDNB, 534 * SENDREC or NOTIFY, verify that the caller is allowed to send to 535 * the given destination. 536 */ 537 if (call_nr != RECEIVE) 538 { 539 if (!may_send_to(caller_ptr, src_dst_p)) { 540 #if DEBUG_ENABLE_IPC_WARNINGS 541 printf( 542 "sys_call: ipc mask denied %s from %d to %d\n", 543 callname, 544 caller_ptr->p_endpoint, src_dst_e); 545 #endif 546 return(ECALLDENIED); /* call denied by ipc mask */ 547 } 548 } 549 } 550 551 /* Check if the process has privileges for the requested call. Calls to the 552 * kernel may only be SENDREC, because tasks always reply and may not block 553 * if the caller doesn't do receive(). 554 */ 555 if (!(priv(caller_ptr)->s_trap_mask & (1 << call_nr))) { 556 #if DEBUG_ENABLE_IPC_WARNINGS 557 printf("sys_call: %s not allowed, caller %d, src_dst %d\n", 558 callname, proc_nr(caller_ptr), src_dst_p); 559 #endif 560 return(ETRAPDENIED); /* trap denied by mask or kernel */ 561 } 562 563 if (call_nr != SENDREC && call_nr != RECEIVE && iskerneln(src_dst_p)) { 564 #if DEBUG_ENABLE_IPC_WARNINGS 565 printf("sys_call: trap %s not allowed, caller %d, src_dst %d\n", 566 callname, proc_nr(caller_ptr), src_dst_e); 567 #endif 568 return(ETRAPDENIED); /* trap denied by mask or kernel */ 569 } 570 571 switch(call_nr) { 572 case SENDREC: 573 /* A flag is set so that notifications cannot interrupt SENDREC. */ 574 caller_ptr->p_misc_flags |= MF_REPLY_PEND; 575 /* fall through */ 576 case SEND: 577 result = mini_send(caller_ptr, src_dst_e, m_ptr, 0); 578 if (call_nr == SEND || result != OK) 579 break; /* done, or SEND failed */ 580 /* fall through for SENDREC */ 581 case RECEIVE: 582 if (call_nr == RECEIVE) { 583 caller_ptr->p_misc_flags &= ~MF_REPLY_PEND; 584 IPC_STATUS_CLEAR(caller_ptr); /* clear IPC status code */ 585 } 586 result = mini_receive(caller_ptr, src_dst_e, m_ptr, 0); 587 break; 588 case NOTIFY: 589 result = mini_notify(caller_ptr, src_dst_e); 590 break; 591 case SENDNB: 592 result = mini_send(caller_ptr, src_dst_e, m_ptr, NON_BLOCKING); 593 break; 594 default: 595 result = EBADCALL; /* illegal system call */ 596 } 597 598 /* Now, return the result of the system call to the caller. */ 599 return(result); 600 } 601 602 int do_ipc(reg_t r1, reg_t r2, reg_t r3) 603 { 604 struct proc *const caller_ptr = get_cpulocal_var(proc_ptr); /* get pointer to caller */ 605 int call_nr = (int) r1; 606 607 assert(!RTS_ISSET(caller_ptr, RTS_SLOT_FREE)); 608 609 /* bill kernel time to this process. */ 610 kbill_ipc = caller_ptr; 611 612 /* If this process is subject to system call tracing, handle that first. */ 613 if (caller_ptr->p_misc_flags & (MF_SC_TRACE | MF_SC_DEFER)) { 614 /* Are we tracing this process, and is it the first sys_call entry? */ 615 if ((caller_ptr->p_misc_flags & (MF_SC_TRACE | MF_SC_DEFER)) == 616 MF_SC_TRACE) { 617 /* We must notify the tracer before processing the actual 618 * system call. If we don't, the tracer could not obtain the 619 * input message. Postpone the entire system call. 620 */ 621 caller_ptr->p_misc_flags &= ~MF_SC_TRACE; 622 assert(!(caller_ptr->p_misc_flags & MF_SC_DEFER)); 623 caller_ptr->p_misc_flags |= MF_SC_DEFER; 624 caller_ptr->p_defer.r1 = r1; 625 caller_ptr->p_defer.r2 = r2; 626 caller_ptr->p_defer.r3 = r3; 627 628 /* Signal the "enter system call" event. Block the process. */ 629 cause_sig(proc_nr(caller_ptr), SIGTRAP); 630 631 /* Preserve the return register's value. */ 632 return caller_ptr->p_reg.retreg; 633 } 634 635 /* If the MF_SC_DEFER flag is set, the syscall is now being resumed. */ 636 caller_ptr->p_misc_flags &= ~MF_SC_DEFER; 637 638 assert (!(caller_ptr->p_misc_flags & MF_SC_ACTIVE)); 639 640 /* Set a flag to allow reliable tracing of leaving the system call. */ 641 caller_ptr->p_misc_flags |= MF_SC_ACTIVE; 642 } 643 644 if(caller_ptr->p_misc_flags & MF_DELIVERMSG) { 645 panic("sys_call: MF_DELIVERMSG on for %s / %d\n", 646 caller_ptr->p_name, caller_ptr->p_endpoint); 647 } 648 649 /* Now check if the call is known and try to perform the request. The only 650 * system calls that exist in MINIX are sending and receiving messages. 651 * - SENDREC: combines SEND and RECEIVE in a single system call 652 * - SEND: sender blocks until its message has been delivered 653 * - RECEIVE: receiver blocks until an acceptable message has arrived 654 * - NOTIFY: asynchronous call; deliver notification or mark pending 655 * - SENDA: list of asynchronous send requests 656 */ 657 switch(call_nr) { 658 case SENDREC: 659 case SEND: 660 case RECEIVE: 661 case NOTIFY: 662 case SENDNB: 663 { 664 /* Process accounting for scheduling */ 665 caller_ptr->p_accounting.ipc_sync++; 666 667 return do_sync_ipc(caller_ptr, call_nr, (endpoint_t) r2, 668 (message *) r3); 669 } 670 case SENDA: 671 { 672 /* 673 * Get and check the size of the argument in bytes as it is a 674 * table 675 */ 676 size_t msg_size = (size_t) r2; 677 678 /* Process accounting for scheduling */ 679 caller_ptr->p_accounting.ipc_async++; 680 681 /* Limit size to something reasonable. An arbitrary choice is 16 682 * times the number of process table entries. 683 */ 684 if (msg_size > 16*(NR_TASKS + NR_PROCS)) 685 return EDOM; 686 return mini_senda(caller_ptr, (asynmsg_t *) r3, msg_size); 687 } 688 case MINIX_KERNINFO: 689 { 690 /* It might not be initialized yet. */ 691 if(!minix_kerninfo_user) { 692 return EBADCALL; 693 } 694 695 arch_set_secondary_ipc_return(caller_ptr, minix_kerninfo_user); 696 return OK; 697 } 698 default: 699 return EBADCALL; /* illegal system call */ 700 } 701 } 702 703 /*===========================================================================* 704 * deadlock * 705 *===========================================================================*/ 706 static int deadlock(function, cp, src_dst_e) 707 int function; /* trap number */ 708 register struct proc *cp; /* pointer to caller */ 709 endpoint_t src_dst_e; /* src or dst process */ 710 { 711 /* Check for deadlock. This can happen if 'caller_ptr' and 'src_dst' have 712 * a cyclic dependency of blocking send and receive calls. The only cyclic 713 * dependency that is not fatal is if the caller and target directly SEND(REC) 714 * and RECEIVE to each other. If a deadlock is found, the group size is 715 * returned. Otherwise zero is returned. 716 */ 717 register struct proc *xp; /* process pointer */ 718 int group_size = 1; /* start with only caller */ 719 #if DEBUG_ENABLE_IPC_WARNINGS 720 static struct proc *processes[NR_PROCS + NR_TASKS]; 721 processes[0] = cp; 722 #endif 723 724 while (src_dst_e != ANY) { /* check while process nr */ 725 int src_dst_slot; 726 okendpt(src_dst_e, &src_dst_slot); 727 xp = proc_addr(src_dst_slot); /* follow chain of processes */ 728 assert(proc_ptr_ok(xp)); 729 assert(!RTS_ISSET(xp, RTS_SLOT_FREE)); 730 #if DEBUG_ENABLE_IPC_WARNINGS 731 processes[group_size] = xp; 732 #endif 733 group_size ++; /* extra process in group */ 734 735 /* Check whether the last process in the chain has a dependency. If it 736 * has not, the cycle cannot be closed and we are done. 737 */ 738 if((src_dst_e = P_BLOCKEDON(xp)) == NONE) 739 return 0; 740 741 /* Now check if there is a cyclic dependency. For group sizes of two, 742 * a combination of SEND(REC) and RECEIVE is not fatal. Larger groups 743 * or other combinations indicate a deadlock. 744 */ 745 if (src_dst_e == cp->p_endpoint) { /* possible deadlock */ 746 if (group_size == 2) { /* caller and src_dst */ 747 /* The function number is magically converted to flags. */ 748 if ((xp->p_rts_flags ^ (function << 2)) & RTS_SENDING) { 749 return(0); /* not a deadlock */ 750 } 751 } 752 #if DEBUG_ENABLE_IPC_WARNINGS 753 { 754 int i; 755 printf("deadlock between these processes:\n"); 756 for(i = 0; i < group_size; i++) { 757 printf(" %10s ", processes[i]->p_name); 758 } 759 printf("\n\n"); 760 for(i = 0; i < group_size; i++) { 761 print_proc(processes[i]); 762 proc_stacktrace(processes[i]); 763 } 764 } 765 #endif 766 return(group_size); /* deadlock found */ 767 } 768 } 769 return(0); /* not a deadlock */ 770 } 771 772 /*===========================================================================* 773 * has_pending * 774 *===========================================================================*/ 775 static int has_pending(sys_map_t *map, int src_p, int asynm) 776 { 777 /* Check to see if there is a pending message from the desired source 778 * available. 779 */ 780 781 int src_id; 782 sys_id_t id = NULL_PRIV_ID; 783 #ifdef CONFIG_SMP 784 struct proc * p; 785 #endif 786 787 /* Either check a specific bit in the mask map, or find the first bit set in 788 * it (if any), depending on whether the receive was called on a specific 789 * source endpoint. 790 */ 791 if (src_p != ANY) { 792 src_id = nr_to_id(src_p); 793 if (get_sys_bit(*map, src_id)) { 794 #ifdef CONFIG_SMP 795 p = proc_addr(id_to_nr(src_id)); 796 if (asynm && RTS_ISSET(p, RTS_VMINHIBIT)) 797 p->p_misc_flags |= MF_SENDA_VM_MISS; 798 else 799 #endif 800 id = src_id; 801 } 802 } else { 803 /* Find a source with a pending message */ 804 for (src_id = 0; src_id < NR_SYS_PROCS; src_id += BITCHUNK_BITS) { 805 if (get_sys_bits(*map, src_id) != 0) { 806 #ifdef CONFIG_SMP 807 while (src_id < NR_SYS_PROCS) { 808 while (!get_sys_bit(*map, src_id)) { 809 if (src_id == NR_SYS_PROCS) 810 goto quit_search; 811 src_id++; 812 } 813 p = proc_addr(id_to_nr(src_id)); 814 /* 815 * We must not let kernel fiddle with pages of a 816 * process which are currently being changed by 817 * VM. It is dangerous! So do not report such a 818 * process as having pending async messages. 819 * Skip it. 820 */ 821 if (asynm && RTS_ISSET(p, RTS_VMINHIBIT)) { 822 p->p_misc_flags |= MF_SENDA_VM_MISS; 823 src_id++; 824 } else 825 goto quit_search; 826 } 827 #else 828 while (!get_sys_bit(*map, src_id)) src_id++; 829 goto quit_search; 830 #endif 831 } 832 } 833 834 quit_search: 835 if (src_id < NR_SYS_PROCS) /* Found one */ 836 id = src_id; 837 } 838 839 return(id); 840 } 841 842 /*===========================================================================* 843 * has_pending_notify * 844 *===========================================================================*/ 845 int has_pending_notify(struct proc * caller, int src_p) 846 { 847 sys_map_t * map = &priv(caller)->s_notify_pending; 848 return has_pending(map, src_p, 0); 849 } 850 851 /*===========================================================================* 852 * has_pending_asend * 853 *===========================================================================*/ 854 int has_pending_asend(struct proc * caller, int src_p) 855 { 856 sys_map_t * map = &priv(caller)->s_asyn_pending; 857 return has_pending(map, src_p, 1); 858 } 859 860 /*===========================================================================* 861 * unset_notify_pending * 862 *===========================================================================*/ 863 void unset_notify_pending(struct proc * caller, int src_p) 864 { 865 sys_map_t * map = &priv(caller)->s_notify_pending; 866 unset_sys_bit(*map, src_p); 867 } 868 869 /*===========================================================================* 870 * mini_send * 871 *===========================================================================*/ 872 int mini_send( 873 register struct proc *caller_ptr, /* who is trying to send a message? */ 874 endpoint_t dst_e, /* to whom is message being sent? */ 875 message *m_ptr, /* pointer to message buffer */ 876 const int flags 877 ) 878 { 879 /* Send a message from 'caller_ptr' to 'dst'. If 'dst' is blocked waiting 880 * for this message, copy the message to it and unblock 'dst'. If 'dst' is 881 * not waiting at all, or is waiting for another source, queue 'caller_ptr'. 882 */ 883 register struct proc *dst_ptr; 884 register struct proc **xpp; 885 int dst_p; 886 dst_p = _ENDPOINT_P(dst_e); 887 dst_ptr = proc_addr(dst_p); 888 889 if (RTS_ISSET(dst_ptr, RTS_NO_ENDPOINT)) 890 { 891 return EDEADSRCDST; 892 } 893 894 /* Check if 'dst' is blocked waiting for this message. The destination's 895 * RTS_SENDING flag may be set when its SENDREC call blocked while sending. 896 */ 897 if (WILLRECEIVE(caller_ptr->p_endpoint, dst_ptr, (vir_bytes)m_ptr, NULL)) { 898 int call; 899 /* Destination is indeed waiting for this message. */ 900 assert(!(dst_ptr->p_misc_flags & MF_DELIVERMSG)); 901 902 if (!(flags & FROM_KERNEL)) { 903 if(copy_msg_from_user(m_ptr, &dst_ptr->p_delivermsg)) 904 return EFAULT; 905 } else { 906 dst_ptr->p_delivermsg = *m_ptr; 907 IPC_STATUS_ADD_FLAGS(dst_ptr, IPC_FLG_MSG_FROM_KERNEL); 908 } 909 910 dst_ptr->p_delivermsg.m_source = caller_ptr->p_endpoint; 911 dst_ptr->p_misc_flags |= MF_DELIVERMSG; 912 913 call = (caller_ptr->p_misc_flags & MF_REPLY_PEND ? SENDREC 914 : (flags & NON_BLOCKING ? SENDNB : SEND)); 915 IPC_STATUS_ADD_CALL(dst_ptr, call); 916 917 if (dst_ptr->p_misc_flags & MF_REPLY_PEND) 918 dst_ptr->p_misc_flags &= ~MF_REPLY_PEND; 919 920 RTS_UNSET(dst_ptr, RTS_RECEIVING); 921 922 #if DEBUG_IPC_HOOK 923 hook_ipc_msgsend(&dst_ptr->p_delivermsg, caller_ptr, dst_ptr); 924 hook_ipc_msgrecv(&dst_ptr->p_delivermsg, caller_ptr, dst_ptr); 925 #endif 926 } else { 927 if(flags & NON_BLOCKING) { 928 return(ENOTREADY); 929 } 930 931 /* Check for a possible deadlock before actually blocking. */ 932 if (deadlock(SEND, caller_ptr, dst_e)) { 933 return(ELOCKED); 934 } 935 936 /* Destination is not waiting. Block and dequeue caller. */ 937 if (!(flags & FROM_KERNEL)) { 938 if(copy_msg_from_user(m_ptr, &caller_ptr->p_sendmsg)) 939 return EFAULT; 940 } else { 941 caller_ptr->p_sendmsg = *m_ptr; 942 /* 943 * we need to remember that this message is from kernel so we 944 * can set the delivery status flags when the message is 945 * actually delivered 946 */ 947 caller_ptr->p_misc_flags |= MF_SENDING_FROM_KERNEL; 948 } 949 950 RTS_SET(caller_ptr, RTS_SENDING); 951 caller_ptr->p_sendto_e = dst_e; 952 953 /* Process is now blocked. Put in on the destination's queue. */ 954 assert(caller_ptr->p_q_link == NULL); 955 xpp = &dst_ptr->p_caller_q; /* find end of list */ 956 while (*xpp) xpp = &(*xpp)->p_q_link; 957 *xpp = caller_ptr; /* add caller to end */ 958 959 #if DEBUG_IPC_HOOK 960 hook_ipc_msgsend(&caller_ptr->p_sendmsg, caller_ptr, dst_ptr); 961 #endif 962 } 963 return(OK); 964 } 965 966 /*===========================================================================* 967 * mini_receive * 968 *===========================================================================*/ 969 static int mini_receive(struct proc * caller_ptr, 970 endpoint_t src_e, /* which message source is wanted */ 971 message * m_buff_usr, /* pointer to message buffer */ 972 const int flags) 973 { 974 /* A process or task wants to get a message. If a message is already queued, 975 * acquire it and deblock the sender. If no message from the desired source 976 * is available block the caller. 977 */ 978 register struct proc **xpp; 979 int r, src_id, found, src_proc_nr, src_p; 980 endpoint_t sender_e; 981 982 assert(!(caller_ptr->p_misc_flags & MF_DELIVERMSG)); 983 984 /* This is where we want our message. */ 985 caller_ptr->p_delivermsg_vir = (vir_bytes) m_buff_usr; 986 987 if(src_e == ANY) src_p = ANY; 988 else 989 { 990 okendpt(src_e, &src_p); 991 if (RTS_ISSET(proc_addr(src_p), RTS_NO_ENDPOINT)) 992 { 993 return EDEADSRCDST; 994 } 995 } 996 997 998 /* Check to see if a message from desired source is already available. The 999 * caller's RTS_SENDING flag may be set if SENDREC couldn't send. If it is 1000 * set, the process should be blocked. 1001 */ 1002 if (!RTS_ISSET(caller_ptr, RTS_SENDING)) { 1003 1004 /* Check if there are pending notifications, except for SENDREC. */ 1005 if (! (caller_ptr->p_misc_flags & MF_REPLY_PEND)) { 1006 1007 /* Check for pending notifications */ 1008 src_id = has_pending_notify(caller_ptr, src_p); 1009 found = src_id != NULL_PRIV_ID; 1010 if(found) { 1011 src_proc_nr = id_to_nr(src_id); /* get source proc */ 1012 sender_e = proc_addr(src_proc_nr)->p_endpoint; 1013 } 1014 1015 if (found && CANRECEIVE(src_e, sender_e, caller_ptr, 0, 1016 &m_notify_buff)) { 1017 1018 #if DEBUG_ENABLE_IPC_WARNINGS 1019 if(src_proc_nr == NONE) { 1020 printf("mini_receive: sending notify from NONE\n"); 1021 } 1022 #endif 1023 assert(src_proc_nr != NONE); 1024 unset_notify_pending(caller_ptr, src_id); /* no longer pending */ 1025 1026 /* Found a suitable source, deliver the notification message. */ 1027 assert(!(caller_ptr->p_misc_flags & MF_DELIVERMSG)); 1028 assert(src_e == ANY || sender_e == src_e); 1029 1030 /* assemble message */ 1031 BuildNotifyMessage(&caller_ptr->p_delivermsg, src_proc_nr, caller_ptr); 1032 caller_ptr->p_delivermsg.m_source = sender_e; 1033 caller_ptr->p_misc_flags |= MF_DELIVERMSG; 1034 1035 IPC_STATUS_ADD_CALL(caller_ptr, NOTIFY); 1036 1037 goto receive_done; 1038 } 1039 } 1040 1041 /* Check for pending asynchronous messages */ 1042 if (has_pending_asend(caller_ptr, src_p) != NULL_PRIV_ID) { 1043 if (src_p != ANY) 1044 r = try_one(src_e, proc_addr(src_p), caller_ptr); 1045 else 1046 r = try_async(caller_ptr); 1047 1048 if (r == OK) { 1049 IPC_STATUS_ADD_CALL(caller_ptr, SENDA); 1050 goto receive_done; 1051 } 1052 } 1053 1054 /* Check caller queue. Use pointer pointers to keep code simple. */ 1055 xpp = &caller_ptr->p_caller_q; 1056 while (*xpp) { 1057 struct proc * sender = *xpp; 1058 endpoint_t sender_e = sender->p_endpoint; 1059 1060 if (CANRECEIVE(src_e, sender_e, caller_ptr, 0, &sender->p_sendmsg)) { 1061 int call; 1062 assert(!RTS_ISSET(sender, RTS_SLOT_FREE)); 1063 assert(!RTS_ISSET(sender, RTS_NO_ENDPOINT)); 1064 1065 /* Found acceptable message. Copy it and update status. */ 1066 assert(!(caller_ptr->p_misc_flags & MF_DELIVERMSG)); 1067 caller_ptr->p_delivermsg = sender->p_sendmsg; 1068 caller_ptr->p_delivermsg.m_source = sender->p_endpoint; 1069 caller_ptr->p_misc_flags |= MF_DELIVERMSG; 1070 RTS_UNSET(sender, RTS_SENDING); 1071 1072 call = (sender->p_misc_flags & MF_REPLY_PEND ? SENDREC : SEND); 1073 IPC_STATUS_ADD_CALL(caller_ptr, call); 1074 1075 /* 1076 * if the message is originally from the kernel on behalf of this 1077 * process, we must send the status flags accordingly 1078 */ 1079 if (sender->p_misc_flags & MF_SENDING_FROM_KERNEL) { 1080 IPC_STATUS_ADD_FLAGS(caller_ptr, IPC_FLG_MSG_FROM_KERNEL); 1081 /* we can clean the flag now, not need anymore */ 1082 sender->p_misc_flags &= ~MF_SENDING_FROM_KERNEL; 1083 } 1084 if (sender->p_misc_flags & MF_SIG_DELAY) 1085 sig_delay_done(sender); 1086 1087 #if DEBUG_IPC_HOOK 1088 hook_ipc_msgrecv(&caller_ptr->p_delivermsg, *xpp, caller_ptr); 1089 #endif 1090 1091 *xpp = sender->p_q_link; /* remove from queue */ 1092 sender->p_q_link = NULL; 1093 goto receive_done; 1094 } 1095 xpp = &sender->p_q_link; /* proceed to next */ 1096 } 1097 } 1098 1099 /* No suitable message is available or the caller couldn't send in SENDREC. 1100 * Block the process trying to receive, unless the flags tell otherwise. 1101 */ 1102 if ( ! (flags & NON_BLOCKING)) { 1103 /* Check for a possible deadlock before actually blocking. */ 1104 if (deadlock(RECEIVE, caller_ptr, src_e)) { 1105 return(ELOCKED); 1106 } 1107 1108 caller_ptr->p_getfrom_e = src_e; 1109 RTS_SET(caller_ptr, RTS_RECEIVING); 1110 return(OK); 1111 } else { 1112 return(ENOTREADY); 1113 } 1114 1115 receive_done: 1116 if (caller_ptr->p_misc_flags & MF_REPLY_PEND) 1117 caller_ptr->p_misc_flags &= ~MF_REPLY_PEND; 1118 return OK; 1119 } 1120 1121 /*===========================================================================* 1122 * mini_notify * 1123 *===========================================================================*/ 1124 int mini_notify( 1125 const struct proc *caller_ptr, /* sender of the notification */ 1126 endpoint_t dst_e /* which process to notify */ 1127 ) 1128 { 1129 register struct proc *dst_ptr; 1130 int src_id; /* source id for late delivery */ 1131 int dst_p; 1132 1133 if (!isokendpt(dst_e, &dst_p)) { 1134 util_stacktrace(); 1135 printf("mini_notify: bogus endpoint %d\n", dst_e); 1136 return EDEADSRCDST; 1137 } 1138 1139 dst_ptr = proc_addr(dst_p); 1140 1141 /* Check to see if target is blocked waiting for this message. A process 1142 * can be both sending and receiving during a SENDREC system call. 1143 */ 1144 if (WILLRECEIVE(caller_ptr->p_endpoint, dst_ptr, 0, &m_notify_buff) && 1145 !(dst_ptr->p_misc_flags & MF_REPLY_PEND)) { 1146 /* Destination is indeed waiting for a message. Assemble a notification 1147 * message and deliver it. Copy from pseudo-source HARDWARE, since the 1148 * message is in the kernel's address space. 1149 */ 1150 assert(!(dst_ptr->p_misc_flags & MF_DELIVERMSG)); 1151 1152 BuildNotifyMessage(&dst_ptr->p_delivermsg, proc_nr(caller_ptr), dst_ptr); 1153 dst_ptr->p_delivermsg.m_source = caller_ptr->p_endpoint; 1154 dst_ptr->p_misc_flags |= MF_DELIVERMSG; 1155 1156 IPC_STATUS_ADD_CALL(dst_ptr, NOTIFY); 1157 RTS_UNSET(dst_ptr, RTS_RECEIVING); 1158 1159 return(OK); 1160 } 1161 1162 /* Destination is not ready to receive the notification. Add it to the 1163 * bit map with pending notifications. Note the indirectness: the privilege id 1164 * instead of the process number is used in the pending bit map. 1165 */ 1166 src_id = priv(caller_ptr)->s_id; 1167 set_sys_bit(priv(dst_ptr)->s_notify_pending, src_id); 1168 return(OK); 1169 } 1170 1171 #define ASCOMPLAIN(caller, entry, field) \ 1172 printf("kernel:%s:%d: asyn failed for %s in %s " \ 1173 "(%d/%zu, tab 0x%lx)\n",__FILE__,__LINE__, \ 1174 field, caller->p_name, entry, priv(caller)->s_asynsize, priv(caller)->s_asyntab) 1175 1176 #define A_RETR(entry) do { \ 1177 if (data_copy( \ 1178 caller_ptr->p_endpoint, table_v + (entry)*sizeof(asynmsg_t),\ 1179 KERNEL, (vir_bytes) &tabent, \ 1180 sizeof(tabent)) != OK) { \ 1181 ASCOMPLAIN(caller_ptr, entry, "message entry"); \ 1182 r = EFAULT; \ 1183 goto asyn_error; \ 1184 } \ 1185 else if(tabent.dst == SELF) { \ 1186 tabent.dst = caller_ptr->p_endpoint; \ 1187 } \ 1188 } while(0) 1189 1190 #define A_INSRT(entry) do { \ 1191 if (data_copy(KERNEL, (vir_bytes) &tabent, \ 1192 caller_ptr->p_endpoint, table_v + (entry)*sizeof(asynmsg_t),\ 1193 sizeof(tabent)) != OK) { \ 1194 ASCOMPLAIN(caller_ptr, entry, "message entry"); \ 1195 /* Do NOT set r or goto asyn_error here! */ \ 1196 } \ 1197 } while(0) 1198 1199 /*===========================================================================* 1200 * try_deliver_senda * 1201 *===========================================================================*/ 1202 int try_deliver_senda(struct proc *caller_ptr, 1203 asynmsg_t *table, 1204 size_t size) 1205 { 1206 int r, dst_p, done, do_notify; 1207 unsigned int i; 1208 unsigned flags; 1209 endpoint_t dst; 1210 struct proc *dst_ptr; 1211 struct priv *privp; 1212 asynmsg_t tabent; 1213 const vir_bytes table_v = (vir_bytes) table; 1214 message *m_ptr = NULL; 1215 1216 privp = priv(caller_ptr); 1217 1218 /* Clear table */ 1219 privp->s_asyntab = -1; 1220 privp->s_asynsize = 0; 1221 privp->s_asynendpoint = caller_ptr->p_endpoint; 1222 1223 if (size == 0) return(OK); /* Nothing to do, just return */ 1224 1225 /* Scan the table */ 1226 do_notify = FALSE; 1227 done = TRUE; 1228 1229 /* Limit size to something reasonable. An arbitrary choice is 16 1230 * times the number of process table entries. 1231 * 1232 * (this check has been duplicated in sys_call but is left here 1233 * as a sanity check) 1234 */ 1235 if (size > 16*(NR_TASKS + NR_PROCS)) { 1236 r = EDOM; 1237 return r; 1238 } 1239 1240 for (i = 0; i < size; i++) { 1241 /* Process each entry in the table and store the result in the table. 1242 * If we're done handling a message, copy the result to the sender. */ 1243 1244 dst = NONE; 1245 /* Copy message to kernel */ 1246 A_RETR(i); 1247 flags = tabent.flags; 1248 dst = tabent.dst; 1249 1250 if (flags == 0) continue; /* Skip empty entries */ 1251 1252 /* 'flags' field must contain only valid bits */ 1253 if(flags & ~(AMF_VALID|AMF_DONE|AMF_NOTIFY|AMF_NOREPLY|AMF_NOTIFY_ERR)) { 1254 r = EINVAL; 1255 goto asyn_error; 1256 } 1257 if (!(flags & AMF_VALID)) { /* Must contain message */ 1258 r = EINVAL; 1259 goto asyn_error; 1260 } 1261 if (flags & AMF_DONE) continue; /* Already done processing */ 1262 1263 r = OK; 1264 if (!isokendpt(tabent.dst, &dst_p)) 1265 r = EDEADSRCDST; /* Bad destination, report the error */ 1266 else if (iskerneln(dst_p)) 1267 r = ECALLDENIED; /* Asyn sends to the kernel are not allowed */ 1268 else if (!may_asynsend_to(caller_ptr, dst_p)) 1269 r = ECALLDENIED; /* Send denied by IPC mask */ 1270 else /* r == OK */ 1271 dst_ptr = proc_addr(dst_p); 1272 1273 /* XXX: RTS_NO_ENDPOINT should be removed */ 1274 if (r == OK && RTS_ISSET(dst_ptr, RTS_NO_ENDPOINT)) { 1275 r = EDEADSRCDST; 1276 } 1277 1278 /* Check if 'dst' is blocked waiting for this message. 1279 * If AMF_NOREPLY is set, do not satisfy the receiving part of 1280 * a SENDREC. 1281 */ 1282 if (r == OK && WILLRECEIVE(caller_ptr->p_endpoint, dst_ptr, 1283 (vir_bytes)&table[i].msg, NULL) && 1284 (!(flags&AMF_NOREPLY) || !(dst_ptr->p_misc_flags&MF_REPLY_PEND))) { 1285 /* Destination is indeed waiting for this message. */ 1286 dst_ptr->p_delivermsg = tabent.msg; 1287 dst_ptr->p_delivermsg.m_source = caller_ptr->p_endpoint; 1288 dst_ptr->p_misc_flags |= MF_DELIVERMSG; 1289 IPC_STATUS_ADD_CALL(dst_ptr, SENDA); 1290 RTS_UNSET(dst_ptr, RTS_RECEIVING); 1291 #if DEBUG_IPC_HOOK 1292 hook_ipc_msgrecv(&dst_ptr->p_delivermsg, caller_ptr, dst_ptr); 1293 #endif 1294 } else if (r == OK) { 1295 /* Inform receiver that something is pending */ 1296 set_sys_bit(priv(dst_ptr)->s_asyn_pending, 1297 priv(caller_ptr)->s_id); 1298 done = FALSE; 1299 continue; 1300 } 1301 1302 /* Store results */ 1303 tabent.result = r; 1304 tabent.flags = flags | AMF_DONE; 1305 if (flags & AMF_NOTIFY) 1306 do_notify = TRUE; 1307 else if (r != OK && (flags & AMF_NOTIFY_ERR)) 1308 do_notify = TRUE; 1309 A_INSRT(i); /* Copy results to caller; ignore errors */ 1310 continue; 1311 1312 asyn_error: 1313 if (dst != NONE) 1314 printf("KERNEL senda error %d to %d\n", r, dst); 1315 else 1316 printf("KERNEL senda error %d\n", r); 1317 } 1318 1319 if (do_notify) 1320 mini_notify(proc_addr(ASYNCM), caller_ptr->p_endpoint); 1321 1322 if (!done) { 1323 privp->s_asyntab = (vir_bytes) table; 1324 privp->s_asynsize = size; 1325 } 1326 1327 return(OK); 1328 } 1329 1330 /*===========================================================================* 1331 * mini_senda * 1332 *===========================================================================*/ 1333 static int mini_senda(struct proc *caller_ptr, asynmsg_t *table, size_t size) 1334 { 1335 struct priv *privp; 1336 1337 privp = priv(caller_ptr); 1338 if (!(privp->s_flags & SYS_PROC)) { 1339 printf( "mini_senda: warning caller has no privilege structure\n"); 1340 return(EPERM); 1341 } 1342 1343 return try_deliver_senda(caller_ptr, table, size); 1344 } 1345 1346 1347 /*===========================================================================* 1348 * try_async * 1349 *===========================================================================*/ 1350 static int try_async(caller_ptr) 1351 struct proc *caller_ptr; 1352 { 1353 int r; 1354 struct priv *privp; 1355 struct proc *src_ptr; 1356 sys_map_t *map; 1357 1358 map = &priv(caller_ptr)->s_asyn_pending; 1359 1360 /* Try all privilege structures */ 1361 for (privp = BEG_PRIV_ADDR; privp < END_PRIV_ADDR; ++privp) { 1362 if (privp->s_proc_nr == NONE) 1363 continue; 1364 1365 if (!get_sys_bit(*map, privp->s_id)) 1366 continue; 1367 1368 src_ptr = proc_addr(privp->s_proc_nr); 1369 1370 #ifdef CONFIG_SMP 1371 /* 1372 * Do not copy from a process which does not have a stable address space 1373 * due to VM fiddling with it 1374 */ 1375 if (RTS_ISSET(src_ptr, RTS_VMINHIBIT)) { 1376 src_ptr->p_misc_flags |= MF_SENDA_VM_MISS; 1377 continue; 1378 } 1379 #endif 1380 1381 assert(!(caller_ptr->p_misc_flags & MF_DELIVERMSG)); 1382 if ((r = try_one(ANY, src_ptr, caller_ptr)) == OK) 1383 return(r); 1384 } 1385 1386 return(ESRCH); 1387 } 1388 1389 1390 /*===========================================================================* 1391 * try_one * 1392 *===========================================================================*/ 1393 static int try_one(endpoint_t receive_e, struct proc *src_ptr, 1394 struct proc *dst_ptr) 1395 { 1396 /* Try to receive an asynchronous message from 'src_ptr' */ 1397 int r = EAGAIN, done, do_notify; 1398 unsigned int flags, i; 1399 size_t size; 1400 endpoint_t dst, src_e; 1401 struct proc *caller_ptr; 1402 struct priv *privp; 1403 asynmsg_t tabent; 1404 vir_bytes table_v; 1405 1406 privp = priv(src_ptr); 1407 if (!(privp->s_flags & SYS_PROC)) return(EPERM); 1408 size = privp->s_asynsize; 1409 table_v = privp->s_asyntab; 1410 1411 /* Clear table pending message flag. We're done unless we're not. */ 1412 unset_sys_bit(priv(dst_ptr)->s_asyn_pending, privp->s_id); 1413 1414 if (size == 0) return(EAGAIN); 1415 if (privp->s_asynendpoint != src_ptr->p_endpoint) return EAGAIN; 1416 if (!may_asynsend_to(src_ptr, proc_nr(dst_ptr))) return (ECALLDENIED); 1417 1418 caller_ptr = src_ptr; /* Needed for A_ macros later on */ 1419 src_e = src_ptr->p_endpoint; 1420 1421 /* Scan the table */ 1422 do_notify = FALSE; 1423 done = TRUE; 1424 1425 for (i = 0; i < size; i++) { 1426 /* Process each entry in the table and store the result in the table. 1427 * If we're done handling a message, copy the result to the sender. 1428 * Some checks done in mini_senda are duplicated here, as the sender 1429 * could've altered the contents of the table in the meantime. 1430 */ 1431 1432 /* Copy message to kernel */ 1433 A_RETR(i); 1434 flags = tabent.flags; 1435 dst = tabent.dst; 1436 1437 if (flags == 0) continue; /* Skip empty entries */ 1438 1439 /* 'flags' field must contain only valid bits */ 1440 if(flags & ~(AMF_VALID|AMF_DONE|AMF_NOTIFY|AMF_NOREPLY|AMF_NOTIFY_ERR)) 1441 r = EINVAL; 1442 else if (!(flags & AMF_VALID)) /* Must contain message */ 1443 r = EINVAL; 1444 else if (flags & AMF_DONE) continue; /* Already done processing */ 1445 1446 /* Clear done flag. The sender is done sending when all messages in the 1447 * table are marked done or empty. However, we will know that only 1448 * the next time we enter this function or when the sender decides to 1449 * send additional asynchronous messages and manages to deliver them 1450 * all. 1451 */ 1452 done = FALSE; 1453 1454 if (r == EINVAL) 1455 goto store_result; 1456 1457 /* Message must be directed at receiving end */ 1458 if (dst != dst_ptr->p_endpoint) continue; 1459 1460 if (!CANRECEIVE(receive_e, src_e, dst_ptr, 1461 table_v + i*sizeof(asynmsg_t) + offsetof(struct asynmsg,msg), 1462 NULL)) { 1463 continue; 1464 } 1465 1466 /* If AMF_NOREPLY is set, then this message is not a reply to a 1467 * SENDREC and thus should not satisfy the receiving part of the 1468 * SENDREC. This message is to be delivered later. 1469 */ 1470 if ((flags & AMF_NOREPLY) && (dst_ptr->p_misc_flags & MF_REPLY_PEND)) 1471 continue; 1472 1473 /* Destination is ready to receive the message; deliver it */ 1474 r = OK; 1475 dst_ptr->p_delivermsg = tabent.msg; 1476 dst_ptr->p_delivermsg.m_source = src_ptr->p_endpoint; 1477 dst_ptr->p_misc_flags |= MF_DELIVERMSG; 1478 #if DEBUG_IPC_HOOK 1479 hook_ipc_msgrecv(&dst_ptr->p_delivermsg, src_ptr, dst_ptr); 1480 #endif 1481 1482 store_result: 1483 /* Store results for sender. We may just have started delivering a 1484 * message, so we must not return an error to the caller in the case 1485 * that storing the results triggers an error! 1486 */ 1487 tabent.result = r; 1488 tabent.flags = flags | AMF_DONE; 1489 if (flags & AMF_NOTIFY) do_notify = TRUE; 1490 else if (r != OK && (flags & AMF_NOTIFY_ERR)) do_notify = TRUE; 1491 A_INSRT(i); /* Copy results to sender; ignore errors */ 1492 1493 break; 1494 } 1495 1496 if (do_notify) 1497 mini_notify(proc_addr(ASYNCM), src_ptr->p_endpoint); 1498 1499 if (done) { 1500 privp->s_asyntab = -1; 1501 privp->s_asynsize = 0; 1502 } else { 1503 set_sys_bit(priv(dst_ptr)->s_asyn_pending, privp->s_id); 1504 } 1505 1506 asyn_error: 1507 return(r); 1508 } 1509 1510 /*===========================================================================* 1511 * cancel_async * 1512 *===========================================================================*/ 1513 int cancel_async(struct proc *src_ptr, struct proc *dst_ptr) 1514 { 1515 /* Cancel asynchronous messages from src to dst, because dst is not interested 1516 * in them (e.g., dst has been restarted) */ 1517 int done, do_notify; 1518 unsigned int flags, i; 1519 size_t size; 1520 endpoint_t dst; 1521 struct proc *caller_ptr; 1522 struct priv *privp; 1523 asynmsg_t tabent; 1524 vir_bytes table_v; 1525 1526 privp = priv(src_ptr); 1527 if (!(privp->s_flags & SYS_PROC)) return(EPERM); 1528 size = privp->s_asynsize; 1529 table_v = privp->s_asyntab; 1530 1531 /* Clear table pending message flag. We're done unless we're not. */ 1532 privp->s_asyntab = -1; 1533 privp->s_asynsize = 0; 1534 unset_sys_bit(priv(dst_ptr)->s_asyn_pending, privp->s_id); 1535 1536 if (size == 0) return(EAGAIN); 1537 if (!may_send_to(src_ptr, proc_nr(dst_ptr))) return(ECALLDENIED); 1538 1539 caller_ptr = src_ptr; /* Needed for A_ macros later on */ 1540 1541 /* Scan the table */ 1542 do_notify = FALSE; 1543 done = TRUE; 1544 1545 1546 for (i = 0; i < size; i++) { 1547 /* Process each entry in the table and store the result in the table. 1548 * If we're done handling a message, copy the result to the sender. 1549 * Some checks done in mini_senda are duplicated here, as the sender 1550 * could've altered the contents of the table in the mean time. 1551 */ 1552 1553 int r = EDEADSRCDST; /* Cancel delivery due to dead dst */ 1554 1555 /* Copy message to kernel */ 1556 A_RETR(i); 1557 flags = tabent.flags; 1558 dst = tabent.dst; 1559 1560 if (flags == 0) continue; /* Skip empty entries */ 1561 1562 /* 'flags' field must contain only valid bits */ 1563 if(flags & ~(AMF_VALID|AMF_DONE|AMF_NOTIFY|AMF_NOREPLY|AMF_NOTIFY_ERR)) 1564 r = EINVAL; 1565 else if (!(flags & AMF_VALID)) /* Must contain message */ 1566 r = EINVAL; 1567 else if (flags & AMF_DONE) continue; /* Already done processing */ 1568 1569 /* Message must be directed at receiving end */ 1570 if (dst != dst_ptr->p_endpoint) { 1571 done = FALSE; 1572 continue; 1573 } 1574 1575 /* Store results for sender */ 1576 tabent.result = r; 1577 tabent.flags = flags | AMF_DONE; 1578 if (flags & AMF_NOTIFY) do_notify = TRUE; 1579 else if (r != OK && (flags & AMF_NOTIFY_ERR)) do_notify = TRUE; 1580 A_INSRT(i); /* Copy results to sender; ignore errors */ 1581 } 1582 1583 if (do_notify) 1584 mini_notify(proc_addr(ASYNCM), src_ptr->p_endpoint); 1585 1586 if (!done) { 1587 privp->s_asyntab = table_v; 1588 privp->s_asynsize = size; 1589 } 1590 1591 asyn_error: 1592 return(OK); 1593 } 1594 1595 /*===========================================================================* 1596 * enqueue * 1597 *===========================================================================*/ 1598 void enqueue( 1599 register struct proc *rp /* this process is now runnable */ 1600 ) 1601 { 1602 /* Add 'rp' to one of the queues of runnable processes. This function is 1603 * responsible for inserting a process into one of the scheduling queues. 1604 * The mechanism is implemented here. The actual scheduling policy is 1605 * defined in sched() and pick_proc(). 1606 * 1607 * This function can be used x-cpu as it always uses the queues of the cpu the 1608 * process is assigned to. 1609 */ 1610 int q = rp->p_priority; /* scheduling queue to use */ 1611 struct proc **rdy_head, **rdy_tail; 1612 1613 assert(proc_is_runnable(rp)); 1614 1615 assert(q >= 0); 1616 1617 rdy_head = get_cpu_var(rp->p_cpu, run_q_head); 1618 rdy_tail = get_cpu_var(rp->p_cpu, run_q_tail); 1619 1620 /* Now add the process to the queue. */ 1621 if (!rdy_head[q]) { /* add to empty queue */ 1622 rdy_head[q] = rdy_tail[q] = rp; /* create a new queue */ 1623 rp->p_nextready = NULL; /* mark new end */ 1624 } 1625 else { /* add to tail of queue */ 1626 rdy_tail[q]->p_nextready = rp; /* chain tail of queue */ 1627 rdy_tail[q] = rp; /* set new queue tail */ 1628 rp->p_nextready = NULL; /* mark new end */ 1629 } 1630 1631 if (cpuid == rp->p_cpu) { 1632 /* 1633 * enqueueing a process with a higher priority than the current one, 1634 * it gets preempted. The current process must be preemptible. Testing 1635 * the priority also makes sure that a process does not preempt itself 1636 */ 1637 struct proc * p; 1638 p = get_cpulocal_var(proc_ptr); 1639 assert(p); 1640 if((p->p_priority > rp->p_priority) && 1641 (priv(p)->s_flags & PREEMPTIBLE)) 1642 RTS_SET(p, RTS_PREEMPTED); /* calls dequeue() */ 1643 } 1644 #ifdef CONFIG_SMP 1645 /* 1646 * if the process was enqueued on a different cpu and the cpu is idle, i.e. 1647 * the time is off, we need to wake up that cpu and let it schedule this new 1648 * process 1649 */ 1650 else if (get_cpu_var(rp->p_cpu, cpu_is_idle)) { 1651 smp_schedule(rp->p_cpu); 1652 } 1653 #endif 1654 1655 /* Make note of when this process was added to queue */ 1656 read_tsc_64(&(get_cpulocal_var(proc_ptr)->p_accounting.enter_queue)); 1657 1658 1659 #if DEBUG_SANITYCHECKS 1660 assert(runqueues_ok_local()); 1661 #endif 1662 } 1663 1664 /*===========================================================================* 1665 * enqueue_head * 1666 *===========================================================================*/ 1667 /* 1668 * put a process at the front of its run queue. It comes handy when a process is 1669 * preempted and removed from run queue to not to have a currently not-runnable 1670 * process on a run queue. We have to put this process back at the fron to be 1671 * fair 1672 */ 1673 static void enqueue_head(struct proc *rp) 1674 { 1675 const int q = rp->p_priority; /* scheduling queue to use */ 1676 1677 struct proc **rdy_head, **rdy_tail; 1678 1679 assert(proc_ptr_ok(rp)); 1680 assert(proc_is_runnable(rp)); 1681 1682 /* 1683 * the process was runnable without its quantum expired when dequeued. A 1684 * process with no time left should have been handled else and differently 1685 */ 1686 assert(rp->p_cpu_time_left); 1687 1688 assert(q >= 0); 1689 1690 1691 rdy_head = get_cpu_var(rp->p_cpu, run_q_head); 1692 rdy_tail = get_cpu_var(rp->p_cpu, run_q_tail); 1693 1694 /* Now add the process to the queue. */ 1695 if (!rdy_head[q]) { /* add to empty queue */ 1696 rdy_head[q] = rdy_tail[q] = rp; /* create a new queue */ 1697 rp->p_nextready = NULL; /* mark new end */ 1698 } else { /* add to head of queue */ 1699 rp->p_nextready = rdy_head[q]; /* chain head of queue */ 1700 rdy_head[q] = rp; /* set new queue head */ 1701 } 1702 1703 /* Make note of when this process was added to queue */ 1704 read_tsc_64(&(get_cpulocal_var(proc_ptr->p_accounting.enter_queue))); 1705 1706 1707 /* Process accounting for scheduling */ 1708 rp->p_accounting.dequeues--; 1709 rp->p_accounting.preempted++; 1710 1711 #if DEBUG_SANITYCHECKS 1712 assert(runqueues_ok_local()); 1713 #endif 1714 } 1715 1716 /*===========================================================================* 1717 * dequeue * 1718 *===========================================================================*/ 1719 void dequeue(struct proc *rp) 1720 /* this process is no longer runnable */ 1721 { 1722 /* A process must be removed from the scheduling queues, for example, because 1723 * it has blocked. If the currently active process is removed, a new process 1724 * is picked to run by calling pick_proc(). 1725 * 1726 * This function can operate x-cpu as it always removes the process from the 1727 * queue of the cpu the process is currently assigned to. 1728 */ 1729 int q = rp->p_priority; /* queue to use */ 1730 struct proc **xpp; /* iterate over queue */ 1731 struct proc *prev_xp; 1732 u64_t tsc, tsc_delta; 1733 1734 struct proc **rdy_tail; 1735 1736 assert(proc_ptr_ok(rp)); 1737 assert(!proc_is_runnable(rp)); 1738 1739 /* Side-effect for kernel: check if the task's stack still is ok? */ 1740 assert (!iskernelp(rp) || *priv(rp)->s_stack_guard == STACK_GUARD); 1741 1742 rdy_tail = get_cpu_var(rp->p_cpu, run_q_tail); 1743 1744 /* Now make sure that the process is not in its ready queue. Remove the 1745 * process if it is found. A process can be made unready even if it is not 1746 * running by being sent a signal that kills it. 1747 */ 1748 prev_xp = NULL; 1749 for (xpp = get_cpu_var_ptr(rp->p_cpu, run_q_head[q]); *xpp; 1750 xpp = &(*xpp)->p_nextready) { 1751 if (*xpp == rp) { /* found process to remove */ 1752 *xpp = (*xpp)->p_nextready; /* replace with next chain */ 1753 if (rp == rdy_tail[q]) { /* queue tail removed */ 1754 rdy_tail[q] = prev_xp; /* set new tail */ 1755 } 1756 1757 break; 1758 } 1759 prev_xp = *xpp; /* save previous in chain */ 1760 } 1761 1762 1763 /* Process accounting for scheduling */ 1764 rp->p_accounting.dequeues++; 1765 1766 /* this is not all that accurate on virtual machines, especially with 1767 IO bound processes that only spend a short amount of time in the queue 1768 at a time. */ 1769 if (rp->p_accounting.enter_queue) { 1770 read_tsc_64(&tsc); 1771 tsc_delta = tsc - rp->p_accounting.enter_queue; 1772 rp->p_accounting.time_in_queue = rp->p_accounting.time_in_queue + 1773 tsc_delta; 1774 rp->p_accounting.enter_queue = 0; 1775 } 1776 1777 /* For ps(1), remember when the process was last dequeued. */ 1778 rp->p_dequeued = get_monotonic(); 1779 1780 #if DEBUG_SANITYCHECKS 1781 assert(runqueues_ok_local()); 1782 #endif 1783 } 1784 1785 /*===========================================================================* 1786 * pick_proc * 1787 *===========================================================================*/ 1788 static struct proc * pick_proc(void) 1789 { 1790 /* Decide who to run now. A new process is selected an returned. 1791 * When a billable process is selected, record it in 'bill_ptr', so that the 1792 * clock task can tell who to bill for system time. 1793 * 1794 * This function always uses the run queues of the local cpu! 1795 */ 1796 register struct proc *rp; /* process to run */ 1797 struct proc **rdy_head; 1798 int q; /* iterate over queues */ 1799 1800 /* Check each of the scheduling queues for ready processes. The number of 1801 * queues is defined in proc.h, and priorities are set in the task table. 1802 * If there are no processes ready to run, return NULL. 1803 */ 1804 rdy_head = get_cpulocal_var(run_q_head); 1805 for (q=0; q < NR_SCHED_QUEUES; q++) { 1806 if(!(rp = rdy_head[q])) { 1807 TRACE(VF_PICKPROC, printf("cpu %d queue %d empty\n", cpuid, q);); 1808 continue; 1809 } 1810 assert(proc_is_runnable(rp)); 1811 if (priv(rp)->s_flags & BILLABLE) 1812 get_cpulocal_var(bill_ptr) = rp; /* bill for system time */ 1813 return rp; 1814 } 1815 return NULL; 1816 } 1817 1818 /*===========================================================================* 1819 * endpoint_lookup * 1820 *===========================================================================*/ 1821 struct proc *endpoint_lookup(endpoint_t e) 1822 { 1823 int n; 1824 1825 if(!isokendpt(e, &n)) return NULL; 1826 1827 return proc_addr(n); 1828 } 1829 1830 /*===========================================================================* 1831 * isokendpt_f * 1832 *===========================================================================*/ 1833 #if DEBUG_ENABLE_IPC_WARNINGS 1834 int isokendpt_f(file, line, e, p, fatalflag) 1835 const char *file; 1836 int line; 1837 #else 1838 int isokendpt_f(e, p, fatalflag) 1839 #endif 1840 endpoint_t e; 1841 int *p; 1842 const int fatalflag; 1843 { 1844 int ok = 0; 1845 /* Convert an endpoint number into a process number. 1846 * Return nonzero if the process is alive with the corresponding 1847 * generation number, zero otherwise. 1848 * 1849 * This function is called with file and line number by the 1850 * isokendpt_d macro if DEBUG_ENABLE_IPC_WARNINGS is defined, 1851 * otherwise without. This allows us to print the where the 1852 * conversion was attempted, making the errors verbose without 1853 * adding code for that at every call. 1854 * 1855 * If fatalflag is nonzero, we must panic if the conversion doesn't 1856 * succeed. 1857 */ 1858 *p = _ENDPOINT_P(e); 1859 ok = 0; 1860 if(isokprocn(*p) && !isemptyn(*p) && proc_addr(*p)->p_endpoint == e) 1861 ok = 1; 1862 if(!ok && fatalflag) 1863 panic("invalid endpoint: %d", e); 1864 return ok; 1865 } 1866 1867 static void notify_scheduler(struct proc *p) 1868 { 1869 message m_no_quantum; 1870 int err; 1871 1872 assert(!proc_kernel_scheduler(p)); 1873 1874 /* dequeue the process */ 1875 RTS_SET(p, RTS_NO_QUANTUM); 1876 /* 1877 * Notify the process's scheduler that it has run out of 1878 * quantum. This is done by sending a message to the scheduler 1879 * on the process's behalf 1880 */ 1881 m_no_quantum.m_source = p->p_endpoint; 1882 m_no_quantum.m_type = SCHEDULING_NO_QUANTUM; 1883 m_no_quantum.m_krn_lsys_schedule.acnt_queue = cpu_time_2_ms(p->p_accounting.time_in_queue); 1884 m_no_quantum.m_krn_lsys_schedule.acnt_deqs = p->p_accounting.dequeues; 1885 m_no_quantum.m_krn_lsys_schedule.acnt_ipc_sync = p->p_accounting.ipc_sync; 1886 m_no_quantum.m_krn_lsys_schedule.acnt_ipc_async = p->p_accounting.ipc_async; 1887 m_no_quantum.m_krn_lsys_schedule.acnt_preempt = p->p_accounting.preempted; 1888 m_no_quantum.m_krn_lsys_schedule.acnt_cpu = cpuid; 1889 m_no_quantum.m_krn_lsys_schedule.acnt_cpu_load = cpu_load(); 1890 1891 /* Reset accounting */ 1892 reset_proc_accounting(p); 1893 1894 if ((err = mini_send(p, p->p_scheduler->p_endpoint, 1895 &m_no_quantum, FROM_KERNEL))) { 1896 panic("WARNING: Scheduling: mini_send returned %d\n", err); 1897 } 1898 } 1899 1900 void proc_no_time(struct proc * p) 1901 { 1902 if (!proc_kernel_scheduler(p) && priv(p)->s_flags & PREEMPTIBLE) { 1903 /* this dequeues the process */ 1904 notify_scheduler(p); 1905 } 1906 else { 1907 /* 1908 * non-preemptible processes only need their quantum to 1909 * be renewed. In fact, they by pass scheduling 1910 */ 1911 p->p_cpu_time_left = ms_2_cpu_time(p->p_quantum_size_ms); 1912 #if DEBUG_RACE 1913 RTS_SET(p, RTS_PREEMPTED); 1914 RTS_UNSET(p, RTS_PREEMPTED); 1915 #endif 1916 } 1917 } 1918 1919 void reset_proc_accounting(struct proc *p) 1920 { 1921 p->p_accounting.preempted = 0; 1922 p->p_accounting.ipc_sync = 0; 1923 p->p_accounting.ipc_async = 0; 1924 p->p_accounting.dequeues = 0; 1925 p->p_accounting.time_in_queue = 0; 1926 p->p_accounting.enter_queue = 0; 1927 } 1928 1929 void copr_not_available_handler(void) 1930 { 1931 struct proc * p; 1932 struct proc ** local_fpu_owner; 1933 /* 1934 * Disable the FPU exception (both for the kernel and for the process 1935 * once it's scheduled), and initialize or restore the FPU state. 1936 */ 1937 1938 disable_fpu_exception(); 1939 1940 p = get_cpulocal_var(proc_ptr); 1941 1942 /* if FPU is not owned by anyone, do not store anything */ 1943 local_fpu_owner = get_cpulocal_var_ptr(fpu_owner); 1944 if (*local_fpu_owner != NULL) { 1945 assert(*local_fpu_owner != p); 1946 save_local_fpu(*local_fpu_owner, FALSE /*retain*/); 1947 } 1948 1949 /* 1950 * restore the current process' state and let it run again, do not 1951 * schedule! 1952 */ 1953 if (restore_fpu(p) != OK) { 1954 /* Restoring FPU state failed. This is always the process's own 1955 * fault. Send a signal, and schedule another process instead. 1956 */ 1957 *local_fpu_owner = NULL; /* release FPU */ 1958 cause_sig(proc_nr(p), SIGFPE); 1959 return; 1960 } 1961 1962 *local_fpu_owner = p; 1963 context_stop(proc_addr(KERNEL)); 1964 restore_user_context(p); 1965 NOT_REACHABLE; 1966 } 1967 1968 void release_fpu(struct proc * p) { 1969 struct proc ** fpu_owner_ptr; 1970 1971 fpu_owner_ptr = get_cpu_var_ptr(p->p_cpu, fpu_owner); 1972 1973 if (*fpu_owner_ptr == p) 1974 *fpu_owner_ptr = NULL; 1975 } 1976 1977 void ser_dump_proc() 1978 { 1979 struct proc *pp; 1980 1981 for (pp= BEG_PROC_ADDR; pp < END_PROC_ADDR; pp++) 1982 { 1983 if (isemptyp(pp)) 1984 continue; 1985 print_proc_recursive(pp); 1986 } 1987 } 1988