1 /* 2 * Copyright (C) 2000 Jeff Dike (jdike@karaya.com) 3 * Licensed under the GPL 4 * Derived (i.e. mostly copied) from arch/i386/kernel/irq.c: 5 * Copyright (C) 1992, 1998 Linus Torvalds, Ingo Molnar 6 */ 7 8 #include "linux/kernel.h" 9 #include "linux/module.h" 10 #include "linux/smp.h" 11 #include "linux/kernel_stat.h" 12 #include "linux/interrupt.h" 13 #include "linux/random.h" 14 #include "linux/slab.h" 15 #include "linux/file.h" 16 #include "linux/proc_fs.h" 17 #include "linux/init.h" 18 #include "linux/seq_file.h" 19 #include "linux/profile.h" 20 #include "linux/hardirq.h" 21 #include "asm/irq.h" 22 #include "asm/hw_irq.h" 23 #include "asm/atomic.h" 24 #include "asm/signal.h" 25 #include "asm/system.h" 26 #include "asm/errno.h" 27 #include "asm/uaccess.h" 28 #include "kern_util.h" 29 #include "irq_user.h" 30 #include "irq_kern.h" 31 #include "os.h" 32 #include "sigio.h" 33 #include "misc_constants.h" 34 #include "as-layout.h" 35 36 /* 37 * Generic, controller-independent functions: 38 */ 39 40 int show_interrupts(struct seq_file *p, void *v) 41 { 42 int i = *(loff_t *) v, j; 43 struct irqaction * action; 44 unsigned long flags; 45 46 if (i == 0) { 47 seq_printf(p, " "); 48 for_each_online_cpu(j) 49 seq_printf(p, "CPU%d ",j); 50 seq_putc(p, '\n'); 51 } 52 53 if (i < NR_IRQS) { 54 spin_lock_irqsave(&irq_desc[i].lock, flags); 55 action = irq_desc[i].action; 56 if (!action) 57 goto skip; 58 seq_printf(p, "%3d: ",i); 59 #ifndef CONFIG_SMP 60 seq_printf(p, "%10u ", kstat_irqs(i)); 61 #else 62 for_each_online_cpu(j) 63 seq_printf(p, "%10u ", kstat_cpu(j).irqs[i]); 64 #endif 65 seq_printf(p, " %14s", irq_desc[i].chip->typename); 66 seq_printf(p, " %s", action->name); 67 68 for (action=action->next; action; action = action->next) 69 seq_printf(p, ", %s", action->name); 70 71 seq_putc(p, '\n'); 72 skip: 73 spin_unlock_irqrestore(&irq_desc[i].lock, flags); 74 } else if (i == NR_IRQS) { 75 seq_putc(p, '\n'); 76 } 77 78 return 0; 79 } 80 81 /* 82 * This list is accessed under irq_lock, except in sigio_handler, 83 * where it is safe from being modified. IRQ handlers won't change it - 84 * if an IRQ source has vanished, it will be freed by free_irqs just 85 * before returning from sigio_handler. That will process a separate 86 * list of irqs to free, with its own locking, coming back here to 87 * remove list elements, taking the irq_lock to do so. 88 */ 89 static struct irq_fd *active_fds = NULL; 90 static struct irq_fd **last_irq_ptr = &active_fds; 91 92 extern void free_irqs(void); 93 94 void sigio_handler(int sig, union uml_pt_regs *regs) 95 { 96 struct irq_fd *irq_fd; 97 int n; 98 99 if (smp_sigio_handler()) 100 return; 101 102 while (1) { 103 n = os_waiting_for_events(active_fds); 104 if (n <= 0) { 105 if(n == -EINTR) continue; 106 else break; 107 } 108 109 for (irq_fd = active_fds; irq_fd != NULL; irq_fd = irq_fd->next) { 110 if (irq_fd->current_events != 0) { 111 irq_fd->current_events = 0; 112 do_IRQ(irq_fd->irq, regs); 113 } 114 } 115 } 116 117 free_irqs(); 118 } 119 120 static DEFINE_SPINLOCK(irq_lock); 121 122 int activate_fd(int irq, int fd, int type, void *dev_id) 123 { 124 struct pollfd *tmp_pfd; 125 struct irq_fd *new_fd, *irq_fd; 126 unsigned long flags; 127 int pid, events, err, n; 128 129 pid = os_getpid(); 130 err = os_set_fd_async(fd, pid); 131 if (err < 0) 132 goto out; 133 134 err = -ENOMEM; 135 new_fd = kmalloc(sizeof(struct irq_fd), GFP_KERNEL); 136 if (new_fd == NULL) 137 goto out; 138 139 if (type == IRQ_READ) 140 events = UM_POLLIN | UM_POLLPRI; 141 else 142 events = UM_POLLOUT; 143 *new_fd = ((struct irq_fd) { .next = NULL, 144 .id = dev_id, 145 .fd = fd, 146 .type = type, 147 .irq = irq, 148 .pid = pid, 149 .events = events, 150 .current_events = 0 } ); 151 152 err = -EBUSY; 153 spin_lock_irqsave(&irq_lock, flags); 154 for (irq_fd = active_fds; irq_fd != NULL; irq_fd = irq_fd->next) { 155 if ((irq_fd->fd == fd) && (irq_fd->type == type)) { 156 printk("Registering fd %d twice\n", fd); 157 printk("Irqs : %d, %d\n", irq_fd->irq, irq); 158 printk("Ids : 0x%p, 0x%p\n", irq_fd->id, dev_id); 159 goto out_unlock; 160 } 161 } 162 163 if (type == IRQ_WRITE) 164 fd = -1; 165 166 tmp_pfd = NULL; 167 n = 0; 168 169 while (1) { 170 n = os_create_pollfd(fd, events, tmp_pfd, n); 171 if (n == 0) 172 break; 173 174 /* n > 0 175 * It means we couldn't put new pollfd to current pollfds 176 * and tmp_fds is NULL or too small for new pollfds array. 177 * Needed size is equal to n as minimum. 178 * 179 * Here we have to drop the lock in order to call 180 * kmalloc, which might sleep. 181 * If something else came in and changed the pollfds array 182 * so we will not be able to put new pollfd struct to pollfds 183 * then we free the buffer tmp_fds and try again. 184 */ 185 spin_unlock_irqrestore(&irq_lock, flags); 186 kfree(tmp_pfd); 187 188 tmp_pfd = kmalloc(n, GFP_KERNEL); 189 if (tmp_pfd == NULL) 190 goto out_kfree; 191 192 spin_lock_irqsave(&irq_lock, flags); 193 } 194 195 *last_irq_ptr = new_fd; 196 last_irq_ptr = &new_fd->next; 197 198 spin_unlock_irqrestore(&irq_lock, flags); 199 200 /* This calls activate_fd, so it has to be outside the critical 201 * section. 202 */ 203 maybe_sigio_broken(fd, (type == IRQ_READ)); 204 205 return 0; 206 207 out_unlock: 208 spin_unlock_irqrestore(&irq_lock, flags); 209 out_kfree: 210 kfree(new_fd); 211 out: 212 return err; 213 } 214 215 static void free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg) 216 { 217 unsigned long flags; 218 219 spin_lock_irqsave(&irq_lock, flags); 220 os_free_irq_by_cb(test, arg, active_fds, &last_irq_ptr); 221 spin_unlock_irqrestore(&irq_lock, flags); 222 } 223 224 struct irq_and_dev { 225 int irq; 226 void *dev; 227 }; 228 229 static int same_irq_and_dev(struct irq_fd *irq, void *d) 230 { 231 struct irq_and_dev *data = d; 232 233 return ((irq->irq == data->irq) && (irq->id == data->dev)); 234 } 235 236 void free_irq_by_irq_and_dev(unsigned int irq, void *dev) 237 { 238 struct irq_and_dev data = ((struct irq_and_dev) { .irq = irq, 239 .dev = dev }); 240 241 free_irq_by_cb(same_irq_and_dev, &data); 242 } 243 244 static int same_fd(struct irq_fd *irq, void *fd) 245 { 246 return (irq->fd == *((int *)fd)); 247 } 248 249 void free_irq_by_fd(int fd) 250 { 251 free_irq_by_cb(same_fd, &fd); 252 } 253 254 /* Must be called with irq_lock held */ 255 static struct irq_fd *find_irq_by_fd(int fd, int irqnum, int *index_out) 256 { 257 struct irq_fd *irq; 258 int i = 0; 259 int fdi; 260 261 for (irq = active_fds; irq != NULL; irq = irq->next) { 262 if ((irq->fd == fd) && (irq->irq == irqnum)) 263 break; 264 i++; 265 } 266 if (irq == NULL) { 267 printk("find_irq_by_fd doesn't have descriptor %d\n", fd); 268 goto out; 269 } 270 fdi = os_get_pollfd(i); 271 if ((fdi != -1) && (fdi != fd)) { 272 printk("find_irq_by_fd - mismatch between active_fds and " 273 "pollfds, fd %d vs %d, need %d\n", irq->fd, 274 fdi, fd); 275 irq = NULL; 276 goto out; 277 } 278 *index_out = i; 279 out: 280 return irq; 281 } 282 283 void reactivate_fd(int fd, int irqnum) 284 { 285 struct irq_fd *irq; 286 unsigned long flags; 287 int i; 288 289 spin_lock_irqsave(&irq_lock, flags); 290 irq = find_irq_by_fd(fd, irqnum, &i); 291 if (irq == NULL) { 292 spin_unlock_irqrestore(&irq_lock, flags); 293 return; 294 } 295 os_set_pollfd(i, irq->fd); 296 spin_unlock_irqrestore(&irq_lock, flags); 297 298 add_sigio_fd(fd); 299 } 300 301 void deactivate_fd(int fd, int irqnum) 302 { 303 struct irq_fd *irq; 304 unsigned long flags; 305 int i; 306 307 spin_lock_irqsave(&irq_lock, flags); 308 irq = find_irq_by_fd(fd, irqnum, &i); 309 if(irq == NULL){ 310 spin_unlock_irqrestore(&irq_lock, flags); 311 return; 312 } 313 314 os_set_pollfd(i, -1); 315 spin_unlock_irqrestore(&irq_lock, flags); 316 317 ignore_sigio_fd(fd); 318 } 319 320 /* 321 * Called just before shutdown in order to provide a clean exec 322 * environment in case the system is rebooting. No locking because 323 * that would cause a pointless shutdown hang if something hadn't 324 * released the lock. 325 */ 326 int deactivate_all_fds(void) 327 { 328 struct irq_fd *irq; 329 int err; 330 331 for (irq = active_fds; irq != NULL; irq = irq->next) { 332 err = os_clear_fd_async(irq->fd); 333 if (err) 334 return err; 335 } 336 /* If there is a signal already queued, after unblocking ignore it */ 337 os_set_ioignore(); 338 339 return 0; 340 } 341 342 #ifdef CONFIG_MODE_TT 343 void forward_interrupts(int pid) 344 { 345 struct irq_fd *irq; 346 unsigned long flags; 347 int err; 348 349 spin_lock_irqsave(&irq_lock, flags); 350 for (irq = active_fds; irq != NULL; irq = irq->next) { 351 err = os_set_owner(irq->fd, pid); 352 if (err < 0) { 353 /* XXX Just remove the irq rather than 354 * print out an infinite stream of these 355 */ 356 printk("Failed to forward %d to pid %d, err = %d\n", 357 irq->fd, pid, -err); 358 } 359 360 irq->pid = pid; 361 } 362 spin_unlock_irqrestore(&irq_lock, flags); 363 } 364 #endif 365 366 /* 367 * do_IRQ handles all normal device IRQ's (the special 368 * SMP cross-CPU interrupts have their own specific 369 * handlers). 370 */ 371 unsigned int do_IRQ(int irq, union uml_pt_regs *regs) 372 { 373 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs); 374 irq_enter(); 375 __do_IRQ(irq); 376 irq_exit(); 377 set_irq_regs(old_regs); 378 return 1; 379 } 380 381 int um_request_irq(unsigned int irq, int fd, int type, 382 irq_handler_t handler, 383 unsigned long irqflags, const char * devname, 384 void *dev_id) 385 { 386 int err; 387 388 err = request_irq(irq, handler, irqflags, devname, dev_id); 389 if (err) 390 return err; 391 392 if (fd != -1) 393 err = activate_fd(irq, fd, type, dev_id); 394 return err; 395 } 396 EXPORT_SYMBOL(um_request_irq); 397 EXPORT_SYMBOL(reactivate_fd); 398 399 /* hw_interrupt_type must define (startup || enable) && 400 * (shutdown || disable) && end */ 401 static void dummy(unsigned int irq) 402 { 403 } 404 405 /* This is used for everything else than the timer. */ 406 static struct hw_interrupt_type normal_irq_type = { 407 .typename = "SIGIO", 408 .release = free_irq_by_irq_and_dev, 409 .disable = dummy, 410 .enable = dummy, 411 .ack = dummy, 412 .end = dummy 413 }; 414 415 static struct hw_interrupt_type SIGVTALRM_irq_type = { 416 .typename = "SIGVTALRM", 417 .release = free_irq_by_irq_and_dev, 418 .shutdown = dummy, /* never called */ 419 .disable = dummy, 420 .enable = dummy, 421 .ack = dummy, 422 .end = dummy 423 }; 424 425 void __init init_IRQ(void) 426 { 427 int i; 428 429 irq_desc[TIMER_IRQ].status = IRQ_DISABLED; 430 irq_desc[TIMER_IRQ].action = NULL; 431 irq_desc[TIMER_IRQ].depth = 1; 432 irq_desc[TIMER_IRQ].chip = &SIGVTALRM_irq_type; 433 enable_irq(TIMER_IRQ); 434 for (i = 1; i < NR_IRQS; i++) { 435 irq_desc[i].status = IRQ_DISABLED; 436 irq_desc[i].action = NULL; 437 irq_desc[i].depth = 1; 438 irq_desc[i].chip = &normal_irq_type; 439 enable_irq(i); 440 } 441 } 442 443 int init_aio_irq(int irq, char *name, irq_handler_t handler) 444 { 445 int fds[2], err; 446 447 err = os_pipe(fds, 1, 1); 448 if (err) { 449 printk("init_aio_irq - os_pipe failed, err = %d\n", -err); 450 goto out; 451 } 452 453 err = um_request_irq(irq, fds[0], IRQ_READ, handler, 454 IRQF_DISABLED | IRQF_SAMPLE_RANDOM, name, 455 (void *) (long) fds[0]); 456 if (err) { 457 printk("init_aio_irq - : um_request_irq failed, err = %d\n", 458 err); 459 goto out_close; 460 } 461 462 err = fds[1]; 463 goto out; 464 465 out_close: 466 os_close_file(fds[0]); 467 os_close_file(fds[1]); 468 out: 469 return err; 470 } 471 472 /* 473 * IRQ stack entry and exit: 474 * 475 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack 476 * and switch over to the IRQ stack after some preparation. We use 477 * sigaltstack to receive signals on a separate stack from the start. 478 * These two functions make sure the rest of the kernel won't be too 479 * upset by being on a different stack. The IRQ stack has a 480 * thread_info structure at the bottom so that current et al continue 481 * to work. 482 * 483 * to_irq_stack copies the current task's thread_info to the IRQ stack 484 * thread_info and sets the tasks's stack to point to the IRQ stack. 485 * 486 * from_irq_stack copies the thread_info struct back (flags may have 487 * been modified) and resets the task's stack pointer. 488 * 489 * Tricky bits - 490 * 491 * What happens when two signals race each other? UML doesn't block 492 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal 493 * could arrive while a previous one is still setting up the 494 * thread_info. 495 * 496 * There are three cases - 497 * The first interrupt on the stack - sets up the thread_info and 498 * handles the interrupt 499 * A nested interrupt interrupting the copying of the thread_info - 500 * can't handle the interrupt, as the stack is in an unknown state 501 * A nested interrupt not interrupting the copying of the 502 * thread_info - doesn't do any setup, just handles the interrupt 503 * 504 * The first job is to figure out whether we interrupted stack setup. 505 * This is done by xchging the signal mask with thread_info->pending. 506 * If the value that comes back is zero, then there is no setup in 507 * progress, and the interrupt can be handled. If the value is 508 * non-zero, then there is stack setup in progress. In order to have 509 * the interrupt handled, we leave our signal in the mask, and it will 510 * be handled by the upper handler after it has set up the stack. 511 * 512 * Next is to figure out whether we are the outer handler or a nested 513 * one. As part of setting up the stack, thread_info->real_thread is 514 * set to non-NULL (and is reset to NULL on exit). This is the 515 * nesting indicator. If it is non-NULL, then the stack is already 516 * set up and the handler can run. 517 */ 518 519 static unsigned long pending_mask; 520 521 unsigned long to_irq_stack(int sig, unsigned long *mask_out) 522 { 523 struct thread_info *ti; 524 unsigned long mask, old; 525 int nested; 526 527 mask = xchg(&pending_mask, 1 << sig); 528 if(mask != 0){ 529 /* If any interrupts come in at this point, we want to 530 * make sure that their bits aren't lost by our 531 * putting our bit in. So, this loop accumulates bits 532 * until xchg returns the same value that we put in. 533 * When that happens, there were no new interrupts, 534 * and pending_mask contains a bit for each interrupt 535 * that came in. 536 */ 537 old = 1 << sig; 538 do { 539 old |= mask; 540 mask = xchg(&pending_mask, old); 541 } while(mask != old); 542 return 1; 543 } 544 545 ti = current_thread_info(); 546 nested = (ti->real_thread != NULL); 547 if(!nested){ 548 struct task_struct *task; 549 struct thread_info *tti; 550 551 task = cpu_tasks[ti->cpu].task; 552 tti = task_thread_info(task); 553 *ti = *tti; 554 ti->real_thread = tti; 555 task->stack = ti; 556 } 557 558 mask = xchg(&pending_mask, 0); 559 *mask_out |= mask | nested; 560 return 0; 561 } 562 563 unsigned long from_irq_stack(int nested) 564 { 565 struct thread_info *ti, *to; 566 unsigned long mask; 567 568 ti = current_thread_info(); 569 570 pending_mask = 1; 571 572 to = ti->real_thread; 573 current->stack = to; 574 ti->real_thread = NULL; 575 *to = *ti; 576 577 mask = xchg(&pending_mask, 0); 578 return mask & ~1; 579 } 580 581