arch/um/kernel/irq.c

   1 /*
   2  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.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/cpumask.h"
   9 #include "linux/hardirq.h"
  10 #include "linux/interrupt.h"
  11 #include "linux/kernel_stat.h"
  12 #include "linux/module.h"
  13 #include "linux/sched.h"
  14 #include "linux/seq_file.h"
  15 #include "linux/slab.h"
  16 #include "as-layout.h"
  17 #include "kern_util.h"
  18 #include "os.h"
  19
  20 /*
  21  * Generic, controller-independent functions:
  22  */
  23
  24 int show_interrupts(struct seq_file *p, void *v)
  25 {
  26         int i = *(loff_t *) v, j;
  27         struct irqaction * action;
  28         unsigned long flags;
  29
  30         if (i == 0) {
  31                 seq_printf(p, "           ");
  32                 for_each_online_cpu(j)
  33                         seq_printf(p, "CPU%d       ",j);
  34                 seq_putc(p, '\n');
  35         }
  36
  37         if (i < NR_IRQS) {
  38                 raw_spin_lock_irqsave(&irq_desc[i].lock, flags);
  39                 action = irq_desc[i].action;
  40                 if (!action)
  41                         goto skip;
  42                 seq_printf(p, "%3d: ",i);
  43 #ifndef CONFIG_SMP
  44                 seq_printf(p, "%10u ", kstat_irqs(i));
  45 #else
  46                 for_each_online_cpu(j)
  47                         seq_printf(p, "%10u ", kstat_irqs_cpu(i, j));
  48 #endif
  49                 seq_printf(p, " %14s", irq_desc[i].chip->typename);
  50                 seq_printf(p, "  %s", action->name);
  51
  52                 for (action=action->next; action; action = action->next)
  53                         seq_printf(p, ", %s", action->name);
  54
  55                 seq_putc(p, '\n');
  56 skip:
  57                 raw_spin_unlock_irqrestore(&irq_desc[i].lock, flags);
  58         } else if (i == NR_IRQS)
  59                 seq_putc(p, '\n');
  60
  61         return 0;
  62 }
  63
  64 /*
  65  * This list is accessed under irq_lock, except in sigio_handler,
  66  * where it is safe from being modified.  IRQ handlers won't change it -
  67  * if an IRQ source has vanished, it will be freed by free_irqs just
  68  * before returning from sigio_handler.  That will process a separate
  69  * list of irqs to free, with its own locking, coming back here to
  70  * remove list elements, taking the irq_lock to do so.
  71  */
  72 static struct irq_fd *active_fds = NULL;
  73 static struct irq_fd **last_irq_ptr = &active_fds;
  74
  75 extern void free_irqs(void);
  76
  77 void sigio_handler(int sig, struct uml_pt_regs *regs)
  78 {
  79         struct irq_fd *irq_fd;
  80         int n;
  81
  82         if (smp_sigio_handler())
  83                 return;
  84
  85         while (1) {
  86                 n = os_waiting_for_events(active_fds);
  87                 if (n <= 0) {
  88                         if (n == -EINTR)
  89                                 continue;
  90                         else break;
  91                 }
  92
  93                 for (irq_fd = active_fds; irq_fd != NULL;
  94                      irq_fd = irq_fd->next) {
  95                         if (irq_fd->current_events != 0) {
  96                                 irq_fd->current_events = 0;
  97                                 do_IRQ(irq_fd->irq, regs);
  98                         }
  99                 }
 100         }
 101
 102         free_irqs();
 103 }
 104
 105 static DEFINE_SPINLOCK(irq_lock);
 106
 107 static int activate_fd(int irq, int fd, int type, void *dev_id)
 108 {
 109         struct pollfd *tmp_pfd;
 110         struct irq_fd *new_fd, *irq_fd;
 111         unsigned long flags;
 112         int events, err, n;
 113
 114         err = os_set_fd_async(fd);
 115         if (err < 0)
 116                 goto out;
 117
 118         err = -ENOMEM;
 119         new_fd = kmalloc(sizeof(struct irq_fd), GFP_KERNEL);
 120         if (new_fd == NULL)
 121                 goto out;
 122
 123         if (type == IRQ_READ)
 124                 events = UM_POLLIN | UM_POLLPRI;
 125         else events = UM_POLLOUT;
 126         *new_fd = ((struct irq_fd) { .next              = NULL,
 127                                      .id                = dev_id,
 128                                      .fd                = fd,
 129                                      .type              = type,
 130                                      .irq               = irq,
 131                                      .events            = events,
 132                                      .current_events    = 0 } );
 133
 134         err = -EBUSY;
 135         spin_lock_irqsave(&irq_lock, flags);
 136         for (irq_fd = active_fds; irq_fd != NULL; irq_fd = irq_fd->next) {
 137                 if ((irq_fd->fd == fd) && (irq_fd->type == type)) {
 138                         printk(KERN_ERR "Registering fd %d twice\n", fd);
 139                         printk(KERN_ERR "Irqs : %d, %d\n", irq_fd->irq, irq);
 140                         printk(KERN_ERR "Ids : 0x%p, 0x%p\n", irq_fd->id,
 141                                dev_id);
 142                         goto out_unlock;
 143                 }
 144         }
 145
 146         if (type == IRQ_WRITE)
 147                 fd = -1;
 148
 149         tmp_pfd = NULL;
 150         n = 0;
 151
 152         while (1) {
 153                 n = os_create_pollfd(fd, events, tmp_pfd, n);
 154                 if (n == 0)
 155                         break;
 156
 157                 /*
 158                  * n > 0
 159                  * It means we couldn't put new pollfd to current pollfds
 160                  * and tmp_fds is NULL or too small for new pollfds array.
 161                  * Needed size is equal to n as minimum.
 162                  *
 163                  * Here we have to drop the lock in order to call
 164                  * kmalloc, which might sleep.
 165                  * If something else came in and changed the pollfds array
 166                  * so we will not be able to put new pollfd struct to pollfds
 167                  * then we free the buffer tmp_fds and try again.
 168                  */
 169                 spin_unlock_irqrestore(&irq_lock, flags);
 170                 kfree(tmp_pfd);
 171
 172                 tmp_pfd = kmalloc(n, GFP_KERNEL);
 173                 if (tmp_pfd == NULL)
 174                         goto out_kfree;
 175
 176                 spin_lock_irqsave(&irq_lock, flags);
 177         }
 178
 179         *last_irq_ptr = new_fd;
 180         last_irq_ptr = &new_fd->next;
 181
 182         spin_unlock_irqrestore(&irq_lock, flags);
 183
 184         /*
 185          * This calls activate_fd, so it has to be outside the critical
 186          * section.
 187          */
 188         maybe_sigio_broken(fd, (type == IRQ_READ));
 189
 190         return 0;
 191
 192  out_unlock:
 193         spin_unlock_irqrestore(&irq_lock, flags);
 194  out_kfree:
 195         kfree(new_fd);
 196  out:
 197         return err;
 198 }
 199
 200 static void free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg)
 201 {
 202         unsigned long flags;
 203
 204         spin_lock_irqsave(&irq_lock, flags);
 205         os_free_irq_by_cb(test, arg, active_fds, &last_irq_ptr);
 206         spin_unlock_irqrestore(&irq_lock, flags);
 207 }
 208
 209 struct irq_and_dev {
 210         int irq;
 211         void *dev;
 212 };
 213
 214 static int same_irq_and_dev(struct irq_fd *irq, void *d)
 215 {
 216         struct irq_and_dev *data = d;
 217
 218         return ((irq->irq == data->irq) && (irq->id == data->dev));
 219 }
 220
 221 static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
 222 {
 223         struct irq_and_dev data = ((struct irq_and_dev) { .irq  = irq,
 224                                                           .dev  = dev });
 225
 226         free_irq_by_cb(same_irq_and_dev, &data);
 227 }
 228
 229 static int same_fd(struct irq_fd *irq, void *fd)
 230 {
 231         return (irq->fd == *((int *)fd));
 232 }
 233
 234 void free_irq_by_fd(int fd)
 235 {
 236         free_irq_by_cb(same_fd, &fd);
 237 }
 238
 239 /* Must be called with irq_lock held */
 240 static struct irq_fd *find_irq_by_fd(int fd, int irqnum, int *index_out)
 241 {
 242         struct irq_fd *irq;
 243         int i = 0;
 244         int fdi;
 245
 246         for (irq = active_fds; irq != NULL; irq = irq->next) {
 247                 if ((irq->fd == fd) && (irq->irq == irqnum))
 248                         break;
 249                 i++;
 250         }
 251         if (irq == NULL) {
 252                 printk(KERN_ERR "find_irq_by_fd doesn't have descriptor %d\n",
 253                        fd);
 254                 goto out;
 255         }
 256         fdi = os_get_pollfd(i);
 257         if ((fdi != -1) && (fdi != fd)) {
 258                 printk(KERN_ERR "find_irq_by_fd - mismatch between active_fds "
 259                        "and pollfds, fd %d vs %d, need %d\n", irq->fd,
 260                        fdi, fd);
 261                 irq = NULL;
 262                 goto out;
 263         }
 264         *index_out = i;
 265  out:
 266         return irq;
 267 }
 268
 269 void reactivate_fd(int fd, int irqnum)
 270 {
 271         struct irq_fd *irq;
 272         unsigned long flags;
 273         int i;
 274
 275         spin_lock_irqsave(&irq_lock, flags);
 276         irq = find_irq_by_fd(fd, irqnum, &i);
 277         if (irq == NULL) {
 278                 spin_unlock_irqrestore(&irq_lock, flags);
 279                 return;
 280         }
 281         os_set_pollfd(i, irq->fd);
 282         spin_unlock_irqrestore(&irq_lock, flags);
 283
 284         add_sigio_fd(fd);
 285 }
 286
 287 void deactivate_fd(int fd, int irqnum)
 288 {
 289         struct irq_fd *irq;
 290         unsigned long flags;
 291         int i;
 292
 293         spin_lock_irqsave(&irq_lock, flags);
 294         irq = find_irq_by_fd(fd, irqnum, &i);
 295         if (irq == NULL) {
 296                 spin_unlock_irqrestore(&irq_lock, flags);
 297                 return;
 298         }
 299
 300         os_set_pollfd(i, -1);
 301         spin_unlock_irqrestore(&irq_lock, flags);
 302
 303         ignore_sigio_fd(fd);
 304 }
 305
 306 /*
 307  * Called just before shutdown in order to provide a clean exec
 308  * environment in case the system is rebooting.  No locking because
 309  * that would cause a pointless shutdown hang if something hadn't
 310  * released the lock.
 311  */
 312 int deactivate_all_fds(void)
 313 {
 314         struct irq_fd *irq;
 315         int err;
 316
 317         for (irq = active_fds; irq != NULL; irq = irq->next) {
 318                 err = os_clear_fd_async(irq->fd);
 319                 if (err)
 320                         return err;
 321         }
 322         /* If there is a signal already queued, after unblocking ignore it */
 323         os_set_ioignore();
 324
 325         return 0;
 326 }
 327
 328 /*
 329  * do_IRQ handles all normal device IRQs (the special
 330  * SMP cross-CPU interrupts have their own specific
 331  * handlers).
 332  */
 333 unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
 334 {
 335         struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
 336         irq_enter();
 337         __do_IRQ(irq);
 338         irq_exit();
 339         set_irq_regs(old_regs);
 340         return 1;
 341 }
 342
 343 int um_request_irq(unsigned int irq, int fd, int type,
 344                    irq_handler_t handler,
 345                    unsigned long irqflags, const char * devname,
 346                    void *dev_id)
 347 {
 348         int err;
 349
 350         if (fd != -1) {
 351                 err = activate_fd(irq, fd, type, dev_id);
 352                 if (err)
 353                         return err;
 354         }
 355
 356         return request_irq(irq, handler, irqflags, devname, dev_id);
 357 }
 358
 359 EXPORT_SYMBOL(um_request_irq);
 360 EXPORT_SYMBOL(reactivate_fd);
 361
 362 /*
 363  * irq_chip must define (startup || enable) &&
 364  * (shutdown || disable) && end
 365  */
 366 static void dummy(unsigned int irq)
 367 {
 368 }
 369
 370 /* This is used for everything else than the timer. */
 371 static struct irq_chip normal_irq_type = {
 372         .typename = "SIGIO",
 373         .release = free_irq_by_irq_and_dev,
 374         .disable = dummy,
 375         .enable = dummy,
 376         .ack = dummy,
 377         .end = dummy
 378 };
 379
 380 static struct irq_chip SIGVTALRM_irq_type = {
 381         .typename = "SIGVTALRM",
 382         .release = free_irq_by_irq_and_dev,
 383         .shutdown = dummy, /* never called */
 384         .disable = dummy,
 385         .enable = dummy,
 386         .ack = dummy,
 387         .end = dummy
 388 };
 389
 390 void __init init_IRQ(void)
 391 {
 392         int i;
 393
 394         irq_desc[TIMER_IRQ].status = IRQ_DISABLED;
 395         irq_desc[TIMER_IRQ].action = NULL;
 396         irq_desc[TIMER_IRQ].depth = 1;
 397         irq_desc[TIMER_IRQ].chip = &SIGVTALRM_irq_type;
 398         enable_irq(TIMER_IRQ);
 399         for (i = 1; i < NR_IRQS; i++) {
 400                 irq_desc[i].status = IRQ_DISABLED;
 401                 irq_desc[i].action = NULL;
 402                 irq_desc[i].depth = 1;
 403                 irq_desc[i].chip = &normal_irq_type;
 404                 enable_irq(i);
 405         }
 406 }
 407
 408 /*
 409  * IRQ stack entry and exit:
 410  *
 411  * Unlike i386, UML doesn't receive IRQs on the normal kernel stack
 412  * and switch over to the IRQ stack after some preparation.  We use
 413  * sigaltstack to receive signals on a separate stack from the start.
 414  * These two functions make sure the rest of the kernel won't be too
 415  * upset by being on a different stack.  The IRQ stack has a
 416  * thread_info structure at the bottom so that current et al continue
 417  * to work.
 418  *
 419  * to_irq_stack copies the current task's thread_info to the IRQ stack
 420  * thread_info and sets the tasks's stack to point to the IRQ stack.
 421  *
 422  * from_irq_stack copies the thread_info struct back (flags may have
 423  * been modified) and resets the task's stack pointer.
 424  *
 425  * Tricky bits -
 426  *
 427  * What happens when two signals race each other?  UML doesn't block
 428  * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal
 429  * could arrive while a previous one is still setting up the
 430  * thread_info.
 431  *
 432  * There are three cases -
 433  *     The first interrupt on the stack - sets up the thread_info and
 434  * handles the interrupt
 435  *     A nested interrupt interrupting the copying of the thread_info -
 436  * can't handle the interrupt, as the stack is in an unknown state
 437  *     A nested interrupt not interrupting the copying of the
 438  * thread_info - doesn't do any setup, just handles the interrupt
 439  *
 440  * The first job is to figure out whether we interrupted stack setup.
 441  * This is done by xchging the signal mask with thread_info->pending.
 442  * If the value that comes back is zero, then there is no setup in
 443  * progress, and the interrupt can be handled.  If the value is
 444  * non-zero, then there is stack setup in progress.  In order to have
 445  * the interrupt handled, we leave our signal in the mask, and it will
 446  * be handled by the upper handler after it has set up the stack.
 447  *
 448  * Next is to figure out whether we are the outer handler or a nested
 449  * one.  As part of setting up the stack, thread_info->real_thread is
 450  * set to non-NULL (and is reset to NULL on exit).  This is the
 451  * nesting indicator.  If it is non-NULL, then the stack is already
 452  * set up and the handler can run.
 453  */
 454
 455 static unsigned long pending_mask;
 456
 457 unsigned long to_irq_stack(unsigned long *mask_out)
 458 {
 459         struct thread_info *ti;
 460         unsigned long mask, old;
 461         int nested;
 462
 463         mask = xchg(&pending_mask, *mask_out);
 464         if (mask != 0) {
 465                 /*
 466                  * If any interrupts come in at this point, we want to
 467                  * make sure that their bits aren't lost by our
 468                  * putting our bit in.  So, this loop accumulates bits
 469                  * until xchg returns the same value that we put in.
 470                  * When that happens, there were no new interrupts,
 471                  * and pending_mask contains a bit for each interrupt
 472                  * that came in.
 473                  */
 474                 old = *mask_out;
 475                 do {
 476                         old |= mask;
 477                         mask = xchg(&pending_mask, old);
 478                 } while (mask != old);
 479                 return 1;
 480         }
 481
 482         ti = current_thread_info();
 483         nested = (ti->real_thread != NULL);
 484         if (!nested) {
 485                 struct task_struct *task;
 486                 struct thread_info *tti;
 487
 488                 task = cpu_tasks[ti->cpu].task;
 489                 tti = task_thread_info(task);
 490
 491                 *ti = *tti;
 492                 ti->real_thread = tti;
 493                 task->stack = ti;
 494         }
 495
 496         mask = xchg(&pending_mask, 0);
 497         *mask_out |= mask | nested;
 498         return 0;
 499 }
 500
 501 unsigned long from_irq_stack(int nested)
 502 {
 503         struct thread_info *ti, *to;
 504         unsigned long mask;
 505
 506         ti = current_thread_info();
 507
 508         pending_mask = 1;
 509
 510         to = ti->real_thread;
 511         current->stack = to;
 512         ti->real_thread = NULL;
 513         *to = *ti;
 514
 515         mask = xchg(&pending_mask, 0);
 516         return mask & ~1;
 517 }
 518