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 uml_pt_regs *regs)
35 struct irq_fd *irq_fd;
38 if (smp_sigio_handler())
42 n = os_waiting_for_events(active_fds);
49 for (irq_fd = active_fds; irq_fd != NULL;
50 irq_fd = irq_fd->next) {
51 if (irq_fd->current_events != 0) {
52 irq_fd->current_events = 0;
53 do_IRQ(irq_fd->irq, regs);
61 static DEFINE_SPINLOCK(irq_lock);
63 static int activate_fd(int irq, int fd, int type, void *dev_id)
65 struct pollfd *tmp_pfd;
66 struct irq_fd *new_fd, *irq_fd;
70 err = os_set_fd_async(fd);
75 new_fd = kmalloc(sizeof(struct irq_fd), GFP_KERNEL);
80 events = UM_POLLIN | UM_POLLPRI;
81 else events = UM_POLLOUT;
82 *new_fd = ((struct irq_fd) { .next = NULL,
88 .current_events = 0 } );
91 spin_lock_irqsave(&irq_lock, flags);
92 for (irq_fd = active_fds; irq_fd != NULL; irq_fd = irq_fd->next) {
93 if ((irq_fd->fd == fd) && (irq_fd->type == type)) {
94 printk(KERN_ERR "Registering fd %d twice\n", fd);
95 printk(KERN_ERR "Irqs : %d, %d\n", irq_fd->irq, irq);
96 printk(KERN_ERR "Ids : 0x%p, 0x%p\n", irq_fd->id,
102 if (type == IRQ_WRITE)
109 n = os_create_pollfd(fd, events, tmp_pfd, n);
115 * It means we couldn't put new pollfd to current pollfds
116 * and tmp_fds is NULL or too small for new pollfds array.
117 * Needed size is equal to n as minimum.
119 * Here we have to drop the lock in order to call
120 * kmalloc, which might sleep.
121 * If something else came in and changed the pollfds array
122 * so we will not be able to put new pollfd struct to pollfds
123 * then we free the buffer tmp_fds and try again.
125 spin_unlock_irqrestore(&irq_lock, flags);
128 tmp_pfd = kmalloc(n, GFP_KERNEL);
132 spin_lock_irqsave(&irq_lock, flags);
135 *last_irq_ptr = new_fd;
136 last_irq_ptr = &new_fd->next;
138 spin_unlock_irqrestore(&irq_lock, flags);
141 * This calls activate_fd, so it has to be outside the critical
144 maybe_sigio_broken(fd, (type == IRQ_READ));
149 spin_unlock_irqrestore(&irq_lock, flags);
156 static void free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg)
160 spin_lock_irqsave(&irq_lock, flags);
161 os_free_irq_by_cb(test, arg, active_fds, &last_irq_ptr);
162 spin_unlock_irqrestore(&irq_lock, flags);
170 static int same_irq_and_dev(struct irq_fd *irq, void *d)
172 struct irq_and_dev *data = d;
174 return ((irq->irq == data->irq) && (irq->id == data->dev));
177 static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
179 struct irq_and_dev data = ((struct irq_and_dev) { .irq = irq,
182 free_irq_by_cb(same_irq_and_dev, &data);
185 static int same_fd(struct irq_fd *irq, void *fd)
187 return (irq->fd == *((int *)fd));
190 void free_irq_by_fd(int fd)
192 free_irq_by_cb(same_fd, &fd);
195 /* Must be called with irq_lock held */
196 static struct irq_fd *find_irq_by_fd(int fd, int irqnum, int *index_out)
202 for (irq = active_fds; irq != NULL; irq = irq->next) {
203 if ((irq->fd == fd) && (irq->irq == irqnum))
208 printk(KERN_ERR "find_irq_by_fd doesn't have descriptor %d\n",
212 fdi = os_get_pollfd(i);
213 if ((fdi != -1) && (fdi != fd)) {
214 printk(KERN_ERR "find_irq_by_fd - mismatch between active_fds "
215 "and pollfds, fd %d vs %d, need %d\n", irq->fd,
225 void reactivate_fd(int fd, int irqnum)
231 spin_lock_irqsave(&irq_lock, flags);
232 irq = find_irq_by_fd(fd, irqnum, &i);
234 spin_unlock_irqrestore(&irq_lock, flags);
237 os_set_pollfd(i, irq->fd);
238 spin_unlock_irqrestore(&irq_lock, flags);
243 void deactivate_fd(int fd, int irqnum)
249 spin_lock_irqsave(&irq_lock, flags);
250 irq = find_irq_by_fd(fd, irqnum, &i);
252 spin_unlock_irqrestore(&irq_lock, flags);
256 os_set_pollfd(i, -1);
257 spin_unlock_irqrestore(&irq_lock, flags);
261 EXPORT_SYMBOL(deactivate_fd);
264 * Called just before shutdown in order to provide a clean exec
265 * environment in case the system is rebooting. No locking because
266 * that would cause a pointless shutdown hang if something hadn't
269 int deactivate_all_fds(void)
274 for (irq = active_fds; irq != NULL; irq = irq->next) {
275 err = os_clear_fd_async(irq->fd);
279 /* If there is a signal already queued, after unblocking ignore it */
286 * do_IRQ handles all normal device IRQs (the special
287 * SMP cross-CPU interrupts have their own specific
290 unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
292 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
294 generic_handle_irq(irq);
296 set_irq_regs(old_regs);
300 int um_request_irq(unsigned int irq, int fd, int type,
301 irq_handler_t handler,
302 unsigned long irqflags, const char * devname,
308 err = activate_fd(irq, fd, type, dev_id);
313 return request_irq(irq, handler, irqflags, devname, dev_id);
316 EXPORT_SYMBOL(um_request_irq);
317 EXPORT_SYMBOL(reactivate_fd);
320 * irq_chip must define at least enable/disable and ack when
321 * the edge handler is used.
323 static void dummy(struct irq_data *d)
327 /* This is used for everything else than the timer. */
328 static struct irq_chip normal_irq_type = {
330 .release = free_irq_by_irq_and_dev,
331 .irq_disable = dummy,
336 static struct irq_chip SIGVTALRM_irq_type = {
338 .release = free_irq_by_irq_and_dev,
339 .irq_disable = dummy,
344 void __init init_IRQ(void)
348 irq_set_chip_and_handler(TIMER_IRQ, &SIGVTALRM_irq_type, handle_edge_irq);
350 for (i = 1; i < NR_IRQS; i++)
351 irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
355 * IRQ stack entry and exit:
357 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack
358 * and switch over to the IRQ stack after some preparation. We use
359 * sigaltstack to receive signals on a separate stack from the start.
360 * These two functions make sure the rest of the kernel won't be too
361 * upset by being on a different stack. The IRQ stack has a
362 * thread_info structure at the bottom so that current et al continue
365 * to_irq_stack copies the current task's thread_info to the IRQ stack
366 * thread_info and sets the tasks's stack to point to the IRQ stack.
368 * from_irq_stack copies the thread_info struct back (flags may have
369 * been modified) and resets the task's stack pointer.
373 * What happens when two signals race each other? UML doesn't block
374 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal
375 * could arrive while a previous one is still setting up the
378 * There are three cases -
379 * The first interrupt on the stack - sets up the thread_info and
380 * handles the interrupt
381 * A nested interrupt interrupting the copying of the thread_info -
382 * can't handle the interrupt, as the stack is in an unknown state
383 * A nested interrupt not interrupting the copying of the
384 * thread_info - doesn't do any setup, just handles the interrupt
386 * The first job is to figure out whether we interrupted stack setup.
387 * This is done by xchging the signal mask with thread_info->pending.
388 * If the value that comes back is zero, then there is no setup in
389 * progress, and the interrupt can be handled. If the value is
390 * non-zero, then there is stack setup in progress. In order to have
391 * the interrupt handled, we leave our signal in the mask, and it will
392 * be handled by the upper handler after it has set up the stack.
394 * Next is to figure out whether we are the outer handler or a nested
395 * one. As part of setting up the stack, thread_info->real_thread is
396 * set to non-NULL (and is reset to NULL on exit). This is the
397 * nesting indicator. If it is non-NULL, then the stack is already
398 * set up and the handler can run.
401 static unsigned long pending_mask;
403 unsigned long to_irq_stack(unsigned long *mask_out)
405 struct thread_info *ti;
406 unsigned long mask, old;
409 mask = xchg(&pending_mask, *mask_out);
412 * If any interrupts come in at this point, we want to
413 * make sure that their bits aren't lost by our
414 * putting our bit in. So, this loop accumulates bits
415 * until xchg returns the same value that we put in.
416 * When that happens, there were no new interrupts,
417 * and pending_mask contains a bit for each interrupt
423 mask = xchg(&pending_mask, old);
424 } while (mask != old);
428 ti = current_thread_info();
429 nested = (ti->real_thread != NULL);
431 struct task_struct *task;
432 struct thread_info *tti;
434 task = cpu_tasks[ti->cpu].task;
435 tti = task_thread_info(task);
438 ti->real_thread = tti;
442 mask = xchg(&pending_mask, 0);
443 *mask_out |= mask | nested;
447 unsigned long from_irq_stack(int nested)
449 struct thread_info *ti, *to;
452 ti = current_thread_info();
456 to = ti->real_thread;
458 ti->real_thread = NULL;
461 mask = xchg(&pending_mask, 0);