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
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>
21 * This list is accessed under irq_lock, except in sigio_handler,
22 * where it is safe from being modified. IRQ handlers won't change it -
23 * if an IRQ source has vanished, it will be freed by free_irqs just
24 * before returning from sigio_handler. That will process a separate
25 * list of irqs to free, with its own locking, coming back here to
26 * remove list elements, taking the irq_lock to do so.
28 static struct irq_fd *active_fds = NULL;
29 static struct irq_fd **last_irq_ptr = &active_fds;
31 extern void free_irqs(void);
33 void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
35 struct irq_fd *irq_fd;
39 n = os_waiting_for_events(active_fds);
46 for (irq_fd = active_fds; irq_fd != NULL;
47 irq_fd = irq_fd->next) {
48 if (irq_fd->current_events != 0) {
49 irq_fd->current_events = 0;
50 do_IRQ(irq_fd->irq, regs);
58 static DEFINE_SPINLOCK(irq_lock);
60 static int activate_fd(int irq, int fd, int type, void *dev_id)
62 struct pollfd *tmp_pfd;
63 struct irq_fd *new_fd, *irq_fd;
67 err = os_set_fd_async(fd);
72 new_fd = kmalloc(sizeof(struct irq_fd), GFP_KERNEL);
77 events = UM_POLLIN | UM_POLLPRI;
78 else events = UM_POLLOUT;
79 *new_fd = ((struct irq_fd) { .next = NULL,
85 .current_events = 0 } );
88 spin_lock_irqsave(&irq_lock, flags);
89 for (irq_fd = active_fds; irq_fd != NULL; irq_fd = irq_fd->next) {
90 if ((irq_fd->fd == fd) && (irq_fd->type == type)) {
91 printk(KERN_ERR "Registering fd %d twice\n", fd);
92 printk(KERN_ERR "Irqs : %d, %d\n", irq_fd->irq, irq);
93 printk(KERN_ERR "Ids : 0x%p, 0x%p\n", irq_fd->id,
99 if (type == IRQ_WRITE)
106 n = os_create_pollfd(fd, events, tmp_pfd, n);
112 * It means we couldn't put new pollfd to current pollfds
113 * and tmp_fds is NULL or too small for new pollfds array.
114 * Needed size is equal to n as minimum.
116 * Here we have to drop the lock in order to call
117 * kmalloc, which might sleep.
118 * If something else came in and changed the pollfds array
119 * so we will not be able to put new pollfd struct to pollfds
120 * then we free the buffer tmp_fds and try again.
122 spin_unlock_irqrestore(&irq_lock, flags);
125 tmp_pfd = kmalloc(n, GFP_KERNEL);
129 spin_lock_irqsave(&irq_lock, flags);
132 *last_irq_ptr = new_fd;
133 last_irq_ptr = &new_fd->next;
135 spin_unlock_irqrestore(&irq_lock, flags);
138 * This calls activate_fd, so it has to be outside the critical
141 maybe_sigio_broken(fd, (type == IRQ_READ));
146 spin_unlock_irqrestore(&irq_lock, flags);
153 static void free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg)
157 spin_lock_irqsave(&irq_lock, flags);
158 os_free_irq_by_cb(test, arg, active_fds, &last_irq_ptr);
159 spin_unlock_irqrestore(&irq_lock, flags);
167 static int same_irq_and_dev(struct irq_fd *irq, void *d)
169 struct irq_and_dev *data = d;
171 return ((irq->irq == data->irq) && (irq->id == data->dev));
174 static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
176 struct irq_and_dev data = ((struct irq_and_dev) { .irq = irq,
179 free_irq_by_cb(same_irq_and_dev, &data);
182 static int same_fd(struct irq_fd *irq, void *fd)
184 return (irq->fd == *((int *)fd));
187 void free_irq_by_fd(int fd)
189 free_irq_by_cb(same_fd, &fd);
192 /* Must be called with irq_lock held */
193 static struct irq_fd *find_irq_by_fd(int fd, int irqnum, int *index_out)
199 for (irq = active_fds; irq != NULL; irq = irq->next) {
200 if ((irq->fd == fd) && (irq->irq == irqnum))
205 printk(KERN_ERR "find_irq_by_fd doesn't have descriptor %d\n",
209 fdi = os_get_pollfd(i);
210 if ((fdi != -1) && (fdi != fd)) {
211 printk(KERN_ERR "find_irq_by_fd - mismatch between active_fds "
212 "and pollfds, fd %d vs %d, need %d\n", irq->fd,
222 void reactivate_fd(int fd, int irqnum)
228 spin_lock_irqsave(&irq_lock, flags);
229 irq = find_irq_by_fd(fd, irqnum, &i);
231 spin_unlock_irqrestore(&irq_lock, flags);
234 os_set_pollfd(i, irq->fd);
235 spin_unlock_irqrestore(&irq_lock, flags);
240 void deactivate_fd(int fd, int irqnum)
246 spin_lock_irqsave(&irq_lock, flags);
247 irq = find_irq_by_fd(fd, irqnum, &i);
249 spin_unlock_irqrestore(&irq_lock, flags);
253 os_set_pollfd(i, -1);
254 spin_unlock_irqrestore(&irq_lock, flags);
258 EXPORT_SYMBOL(deactivate_fd);
261 * Called just before shutdown in order to provide a clean exec
262 * environment in case the system is rebooting. No locking because
263 * that would cause a pointless shutdown hang if something hadn't
266 int deactivate_all_fds(void)
271 for (irq = active_fds; irq != NULL; irq = irq->next) {
272 err = os_clear_fd_async(irq->fd);
276 /* If there is a signal already queued, after unblocking ignore it */
283 * do_IRQ handles all normal device IRQs (the special
284 * SMP cross-CPU interrupts have their own specific
287 unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
289 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
291 generic_handle_irq(irq);
293 set_irq_regs(old_regs);
297 void um_free_irq(unsigned int irq, void *dev)
299 free_irq_by_irq_and_dev(irq, dev);
302 EXPORT_SYMBOL(um_free_irq);
304 int um_request_irq(unsigned int irq, int fd, int type,
305 irq_handler_t handler,
306 unsigned long irqflags, const char * devname,
312 err = activate_fd(irq, fd, type, dev_id);
317 return request_irq(irq, handler, irqflags, devname, dev_id);
320 EXPORT_SYMBOL(um_request_irq);
321 EXPORT_SYMBOL(reactivate_fd);
324 * irq_chip must define at least enable/disable and ack when
325 * the edge handler is used.
327 static void dummy(struct irq_data *d)
331 /* This is used for everything else than the timer. */
332 static struct irq_chip normal_irq_type = {
334 .irq_disable = dummy,
341 static struct irq_chip SIGVTALRM_irq_type = {
343 .irq_disable = dummy,
350 void __init init_IRQ(void)
354 irq_set_chip_and_handler(TIMER_IRQ, &SIGVTALRM_irq_type, handle_edge_irq);
356 for (i = 1; i < NR_IRQS; i++)
357 irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
361 * IRQ stack entry and exit:
363 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack
364 * and switch over to the IRQ stack after some preparation. We use
365 * sigaltstack to receive signals on a separate stack from the start.
366 * These two functions make sure the rest of the kernel won't be too
367 * upset by being on a different stack. The IRQ stack has a
368 * thread_info structure at the bottom so that current et al continue
371 * to_irq_stack copies the current task's thread_info to the IRQ stack
372 * thread_info and sets the tasks's stack to point to the IRQ stack.
374 * from_irq_stack copies the thread_info struct back (flags may have
375 * been modified) and resets the task's stack pointer.
379 * What happens when two signals race each other? UML doesn't block
380 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal
381 * could arrive while a previous one is still setting up the
384 * There are three cases -
385 * The first interrupt on the stack - sets up the thread_info and
386 * handles the interrupt
387 * A nested interrupt interrupting the copying of the thread_info -
388 * can't handle the interrupt, as the stack is in an unknown state
389 * A nested interrupt not interrupting the copying of the
390 * thread_info - doesn't do any setup, just handles the interrupt
392 * The first job is to figure out whether we interrupted stack setup.
393 * This is done by xchging the signal mask with thread_info->pending.
394 * If the value that comes back is zero, then there is no setup in
395 * progress, and the interrupt can be handled. If the value is
396 * non-zero, then there is stack setup in progress. In order to have
397 * the interrupt handled, we leave our signal in the mask, and it will
398 * be handled by the upper handler after it has set up the stack.
400 * Next is to figure out whether we are the outer handler or a nested
401 * one. As part of setting up the stack, thread_info->real_thread is
402 * set to non-NULL (and is reset to NULL on exit). This is the
403 * nesting indicator. If it is non-NULL, then the stack is already
404 * set up and the handler can run.
407 static unsigned long pending_mask;
409 unsigned long to_irq_stack(unsigned long *mask_out)
411 struct thread_info *ti;
412 unsigned long mask, old;
415 mask = xchg(&pending_mask, *mask_out);
418 * If any interrupts come in at this point, we want to
419 * make sure that their bits aren't lost by our
420 * putting our bit in. So, this loop accumulates bits
421 * until xchg returns the same value that we put in.
422 * When that happens, there were no new interrupts,
423 * and pending_mask contains a bit for each interrupt
429 mask = xchg(&pending_mask, old);
430 } while (mask != old);
434 ti = current_thread_info();
435 nested = (ti->real_thread != NULL);
437 struct task_struct *task;
438 struct thread_info *tti;
440 task = cpu_tasks[ti->cpu].task;
441 tti = task_thread_info(task);
444 ti->real_thread = tti;
448 mask = xchg(&pending_mask, 0);
449 *mask_out |= mask | nested;
453 unsigned long from_irq_stack(int nested)
455 struct thread_info *ti, *to;
458 ti = current_thread_info();
462 to = ti->real_thread;
464 ti->real_thread = NULL;
467 mask = xchg(&pending_mask, 0);