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 "um_malloc.h" 34 #include "misc_constants.h" 35 #include "as-layout.h" 36 37 /* 38 * Generic, controller-independent functions: 39 */ 40 41 int show_interrupts(struct seq_file *p, void *v) 42 { 43 int i = *(loff_t *) v, j; 44 struct irqaction * action; 45 unsigned long flags; 46 47 if (i == 0) { 48 seq_printf(p, " "); 49 for_each_online_cpu(j) 50 seq_printf(p, "CPU%d ",j); 51 seq_putc(p, '\n'); 52 } 53 54 if (i < NR_IRQS) { 55 spin_lock_irqsave(&irq_desc[i].lock, flags); 56 action = irq_desc[i].action; 57 if (!action) 58 goto skip; 59 seq_printf(p, "%3d: ",i); 60 #ifndef CONFIG_SMP 61 seq_printf(p, "%10u ", kstat_irqs(i)); 62 #else 63 for_each_online_cpu(j) 64 seq_printf(p, "%10u ", kstat_cpu(j).irqs[i]); 65 #endif 66 seq_printf(p, " %14s", irq_desc[i].chip->typename); 67 seq_printf(p, " %s", action->name); 68 69 for (action=action->next; action; action = action->next) 70 seq_printf(p, ", %s", action->name); 71 72 seq_putc(p, '\n'); 73 skip: 74 spin_unlock_irqrestore(&irq_desc[i].lock, flags); 75 } else if (i == NR_IRQS) { 76 seq_putc(p, '\n'); 77 } 78 79 return 0; 80 } 81 82 /* 83 * This list is accessed under irq_lock, except in sigio_handler, 84 * where it is safe from being modified. IRQ handlers won't change it - 85 * if an IRQ source has vanished, it will be freed by free_irqs just 86 * before returning from sigio_handler. That will process a separate 87 * list of irqs to free, with its own locking, coming back here to 88 * remove list elements, taking the irq_lock to do so. 89 */ 90 static struct irq_fd *active_fds = NULL; 91 static struct irq_fd **last_irq_ptr = &active_fds; 92 93 extern void free_irqs(void); 94 95 void sigio_handler(int sig, union uml_pt_regs *regs) 96 { 97 struct irq_fd *irq_fd; 98 int n; 99 100 if (smp_sigio_handler()) 101 return; 102 103 while (1) { 104 n = os_waiting_for_events(active_fds); 105 if (n <= 0) { 106 if(n == -EINTR) continue; 107 else break; 108 } 109 110 for (irq_fd = active_fds; irq_fd != NULL; irq_fd = irq_fd->next) { 111 if (irq_fd->current_events != 0) { 112 irq_fd->current_events = 0; 113 do_IRQ(irq_fd->irq, regs); 114 } 115 } 116 } 117 118 free_irqs(); 119 } 120 121 static DEFINE_SPINLOCK(irq_lock); 122 123 int activate_fd(int irq, int fd, int type, void *dev_id) 124 { 125 struct pollfd *tmp_pfd; 126 struct irq_fd *new_fd, *irq_fd; 127 unsigned long flags; 128 int pid, events, err, n; 129 130 pid = os_getpid(); 131 err = os_set_fd_async(fd, pid); 132 if (err < 0) 133 goto out; 134 135 err = -ENOMEM; 136 new_fd = kmalloc(sizeof(struct irq_fd), GFP_KERNEL); 137 if (new_fd == NULL) 138 goto out; 139 140 if (type == IRQ_READ) 141 events = UM_POLLIN | UM_POLLPRI; 142 else 143 events = UM_POLLOUT; 144 *new_fd = ((struct irq_fd) { .next = NULL, 145 .id = dev_id, 146 .fd = fd, 147 .type = type, 148 .irq = irq, 149 .pid = pid, 150 .events = events, 151 .current_events = 0 } ); 152 153 err = -EBUSY; 154 spin_lock_irqsave(&irq_lock, flags); 155 for (irq_fd = active_fds; irq_fd != NULL; irq_fd = irq_fd->next) { 156 if ((irq_fd->fd == fd) && (irq_fd->type == type)) { 157 printk("Registering fd %d twice\n", fd); 158 printk("Irqs : %d, %d\n", irq_fd->irq, irq); 159 printk("Ids : 0x%p, 0x%p\n", irq_fd->id, dev_id); 160 goto out_unlock; 161 } 162 } 163 164 if (type == IRQ_WRITE) 165 fd = -1; 166 167 tmp_pfd = NULL; 168 n = 0; 169 170 while (1) { 171 n = os_create_pollfd(fd, events, tmp_pfd, n); 172 if (n == 0) 173 break; 174 175 /* n > 0 176 * It means we couldn't put new pollfd to current pollfds 177 * and tmp_fds is NULL or too small for new pollfds array. 178 * Needed size is equal to n as minimum. 179 * 180 * Here we have to drop the lock in order to call 181 * kmalloc, which might sleep. 182 * If something else came in and changed the pollfds array 183 * so we will not be able to put new pollfd struct to pollfds 184 * then we free the buffer tmp_fds and try again. 185 */ 186 spin_unlock_irqrestore(&irq_lock, flags); 187 kfree(tmp_pfd); 188 189 tmp_pfd = kmalloc(n, GFP_KERNEL); 190 if (tmp_pfd == NULL) 191 goto out_kfree; 192 193 spin_lock_irqsave(&irq_lock, flags); 194 } 195 196 *last_irq_ptr = new_fd; 197 last_irq_ptr = &new_fd->next; 198 199 spin_unlock_irqrestore(&irq_lock, flags); 200 201 /* This calls activate_fd, so it has to be outside the critical 202 * section. 203 */ 204 maybe_sigio_broken(fd, (type == IRQ_READ)); 205 206 return 0; 207 208 out_unlock: 209 spin_unlock_irqrestore(&irq_lock, flags); 210 out_kfree: 211 kfree(new_fd); 212 out: 213 return err; 214 } 215 216 static void free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg) 217 { 218 unsigned long flags; 219 220 spin_lock_irqsave(&irq_lock, flags); 221 os_free_irq_by_cb(test, arg, active_fds, &last_irq_ptr); 222 spin_unlock_irqrestore(&irq_lock, flags); 223 } 224 225 struct irq_and_dev { 226 int irq; 227 void *dev; 228 }; 229 230 static int same_irq_and_dev(struct irq_fd *irq, void *d) 231 { 232 struct irq_and_dev *data = d; 233 234 return ((irq->irq == data->irq) && (irq->id == data->dev)); 235 } 236 237 void free_irq_by_irq_and_dev(unsigned int irq, void *dev) 238 { 239 struct irq_and_dev data = ((struct irq_and_dev) { .irq = irq, 240 .dev = dev }); 241 242 free_irq_by_cb(same_irq_and_dev, &data); 243 } 244 245 static int same_fd(struct irq_fd *irq, void *fd) 246 { 247 return (irq->fd == *((int *)fd)); 248 } 249 250 void free_irq_by_fd(int fd) 251 { 252 free_irq_by_cb(same_fd, &fd); 253 } 254 255 /* Must be called with irq_lock held */ 256 static struct irq_fd *find_irq_by_fd(int fd, int irqnum, int *index_out) 257 { 258 struct irq_fd *irq; 259 int i = 0; 260 int fdi; 261 262 for (irq = active_fds; irq != NULL; irq = irq->next) { 263 if ((irq->fd == fd) && (irq->irq == irqnum)) 264 break; 265 i++; 266 } 267 if (irq == NULL) { 268 printk("find_irq_by_fd doesn't have descriptor %d\n", fd); 269 goto out; 270 } 271 fdi = os_get_pollfd(i); 272 if ((fdi != -1) && (fdi != fd)) { 273 printk("find_irq_by_fd - mismatch between active_fds and " 274 "pollfds, fd %d vs %d, need %d\n", irq->fd, 275 fdi, fd); 276 irq = NULL; 277 goto out; 278 } 279 *index_out = i; 280 out: 281 return irq; 282 } 283 284 void reactivate_fd(int fd, int irqnum) 285 { 286 struct irq_fd *irq; 287 unsigned long flags; 288 int i; 289 290 spin_lock_irqsave(&irq_lock, flags); 291 irq = find_irq_by_fd(fd, irqnum, &i); 292 if (irq == NULL) { 293 spin_unlock_irqrestore(&irq_lock, flags); 294 return; 295 } 296 os_set_pollfd(i, irq->fd); 297 spin_unlock_irqrestore(&irq_lock, flags); 298 299 add_sigio_fd(fd); 300 } 301 302 void deactivate_fd(int fd, int irqnum) 303 { 304 struct irq_fd *irq; 305 unsigned long flags; 306 int i; 307 308 spin_lock_irqsave(&irq_lock, flags); 309 irq = find_irq_by_fd(fd, irqnum, &i); 310 if(irq == NULL){ 311 spin_unlock_irqrestore(&irq_lock, flags); 312 return; 313 } 314 315 os_set_pollfd(i, -1); 316 spin_unlock_irqrestore(&irq_lock, flags); 317 318 ignore_sigio_fd(fd); 319 } 320 321 /* 322 * Called just before shutdown in order to provide a clean exec 323 * environment in case the system is rebooting. No locking because 324 * that would cause a pointless shutdown hang if something hadn't 325 * released the lock. 326 */ 327 int deactivate_all_fds(void) 328 { 329 struct irq_fd *irq; 330 int err; 331 332 for (irq = active_fds; irq != NULL; irq = irq->next) { 333 err = os_clear_fd_async(irq->fd); 334 if (err) 335 return err; 336 } 337 /* If there is a signal already queued, after unblocking ignore it */ 338 os_set_ioignore(); 339 340 return 0; 341 } 342 343 #ifdef CONFIG_MODE_TT 344 void forward_interrupts(int pid) 345 { 346 struct irq_fd *irq; 347 unsigned long flags; 348 int err; 349 350 spin_lock_irqsave(&irq_lock, flags); 351 for (irq = active_fds; irq != NULL; irq = irq->next) { 352 err = os_set_owner(irq->fd, pid); 353 if (err < 0) { 354 /* XXX Just remove the irq rather than 355 * print out an infinite stream of these 356 */ 357 printk("Failed to forward %d to pid %d, err = %d\n", 358 irq->fd, pid, -err); 359 } 360 361 irq->pid = pid; 362 } 363 spin_unlock_irqrestore(&irq_lock, flags); 364 } 365 #endif 366 367 /* 368 * do_IRQ handles all normal device IRQ's (the special 369 * SMP cross-CPU interrupts have their own specific 370 * handlers). 371 */ 372 unsigned int do_IRQ(int irq, union uml_pt_regs *regs) 373 { 374 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs); 375 irq_enter(); 376 __do_IRQ(irq); 377 irq_exit(); 378 set_irq_regs(old_regs); 379 return 1; 380 } 381 382 int um_request_irq(unsigned int irq, int fd, int type, 383 irq_handler_t handler, 384 unsigned long irqflags, const char * devname, 385 void *dev_id) 386 { 387 int err; 388 389 err = request_irq(irq, handler, irqflags, devname, dev_id); 390 if (err) 391 return err; 392 393 if (fd != -1) 394 err = activate_fd(irq, fd, type, dev_id); 395 return err; 396 } 397 EXPORT_SYMBOL(um_request_irq); 398 EXPORT_SYMBOL(reactivate_fd); 399 400 /* hw_interrupt_type must define (startup || enable) && 401 * (shutdown || disable) && end */ 402 static void dummy(unsigned int irq) 403 { 404 } 405 406 /* This is used for everything else than the timer. */ 407 static struct hw_interrupt_type normal_irq_type = { 408 .typename = "SIGIO", 409 .release = free_irq_by_irq_and_dev, 410 .disable = dummy, 411 .enable = dummy, 412 .ack = dummy, 413 .end = dummy 414 }; 415 416 static struct hw_interrupt_type SIGVTALRM_irq_type = { 417 .typename = "SIGVTALRM", 418 .release = free_irq_by_irq_and_dev, 419 .shutdown = dummy, /* never called */ 420 .disable = dummy, 421 .enable = dummy, 422 .ack = dummy, 423 .end = dummy 424 }; 425 426 void __init init_IRQ(void) 427 { 428 int i; 429 430 irq_desc[TIMER_IRQ].status = IRQ_DISABLED; 431 irq_desc[TIMER_IRQ].action = NULL; 432 irq_desc[TIMER_IRQ].depth = 1; 433 irq_desc[TIMER_IRQ].chip = &SIGVTALRM_irq_type; 434 enable_irq(TIMER_IRQ); 435 for (i = 1; i < NR_IRQS; i++) { 436 irq_desc[i].status = IRQ_DISABLED; 437 irq_desc[i].action = NULL; 438 irq_desc[i].depth = 1; 439 irq_desc[i].chip = &normal_irq_type; 440 enable_irq(i); 441 } 442 } 443 444 int init_aio_irq(int irq, char *name, irq_handler_t handler) 445 { 446 int fds[2], err; 447 448 err = os_pipe(fds, 1, 1); 449 if (err) { 450 printk("init_aio_irq - os_pipe failed, err = %d\n", -err); 451 goto out; 452 } 453 454 err = um_request_irq(irq, fds[0], IRQ_READ, handler, 455 IRQF_DISABLED | IRQF_SAMPLE_RANDOM, name, 456 (void *) (long) fds[0]); 457 if (err) { 458 printk("init_aio_irq - : um_request_irq failed, err = %d\n", 459 err); 460 goto out_close; 461 } 462 463 err = fds[1]; 464 goto out; 465 466 out_close: 467 os_close_file(fds[0]); 468 os_close_file(fds[1]); 469 out: 470 return err; 471 } 472 473 /* 474 * IRQ stack entry and exit: 475 * 476 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack 477 * and switch over to the IRQ stack after some preparation. We use 478 * sigaltstack to receive signals on a separate stack from the start. 479 * These two functions make sure the rest of the kernel won't be too 480 * upset by being on a different stack. The IRQ stack has a 481 * thread_info structure at the bottom so that current et al continue 482 * to work. 483 * 484 * to_irq_stack copies the current task's thread_info to the IRQ stack 485 * thread_info and sets the tasks's stack to point to the IRQ stack. 486 * 487 * from_irq_stack copies the thread_info struct back (flags may have 488 * been modified) and resets the task's stack pointer. 489 * 490 * Tricky bits - 491 * 492 * What happens when two signals race each other? UML doesn't block 493 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal 494 * could arrive while a previous one is still setting up the 495 * thread_info. 496 * 497 * There are three cases - 498 * The first interrupt on the stack - sets up the thread_info and 499 * handles the interrupt 500 * A nested interrupt interrupting the copying of the thread_info - 501 * can't handle the interrupt, as the stack is in an unknown state 502 * A nested interrupt not interrupting the copying of the 503 * thread_info - doesn't do any setup, just handles the interrupt 504 * 505 * The first job is to figure out whether we interrupted stack setup. 506 * This is done by xchging the signal mask with thread_info->pending. 507 * If the value that comes back is zero, then there is no setup in 508 * progress, and the interrupt can be handled. If the value is 509 * non-zero, then there is stack setup in progress. In order to have 510 * the interrupt handled, we leave our signal in the mask, and it will 511 * be handled by the upper handler after it has set up the stack. 512 * 513 * Next is to figure out whether we are the outer handler or a nested 514 * one. As part of setting up the stack, thread_info->real_thread is 515 * set to non-NULL (and is reset to NULL on exit). This is the 516 * nesting indicator. If it is non-NULL, then the stack is already 517 * set up and the handler can run. 518 */ 519 520 static unsigned long pending_mask; 521 522 unsigned long to_irq_stack(int sig, unsigned long *mask_out) 523 { 524 struct thread_info *ti; 525 unsigned long mask, old; 526 int nested; 527 528 mask = xchg(&pending_mask, 1 << sig); 529 if(mask != 0){ 530 /* If any interrupts come in at this point, we want to 531 * make sure that their bits aren't lost by our 532 * putting our bit in. So, this loop accumulates bits 533 * until xchg returns the same value that we put in. 534 * When that happens, there were no new interrupts, 535 * and pending_mask contains a bit for each interrupt 536 * that came in. 537 */ 538 old = 1 << sig; 539 do { 540 old |= mask; 541 mask = xchg(&pending_mask, old); 542 } while(mask != old); 543 return 1; 544 } 545 546 ti = current_thread_info(); 547 nested = (ti->real_thread != NULL); 548 if(!nested){ 549 struct task_struct *task; 550 struct thread_info *tti; 551 552 task = cpu_tasks[ti->cpu].task; 553 tti = task_thread_info(task); 554 *ti = *tti; 555 ti->real_thread = tti; 556 task->stack = ti; 557 } 558 559 mask = xchg(&pending_mask, 0); 560 *mask_out |= mask | nested; 561 return 0; 562 } 563 564 unsigned long from_irq_stack(int nested) 565 { 566 struct thread_info *ti, *to; 567 unsigned long mask; 568 569 ti = current_thread_info(); 570 571 pending_mask = 1; 572 573 to = ti->real_thread; 574 current->stack = to; 575 ti->real_thread = NULL; 576 *to = *ti; 577 578 mask = xchg(&pending_mask, 0); 579 return mask & ~1; 580 } 581 582