2 * Copyright (C) 2017 - Cambridge Greys Ltd
3 * Copyright (C) 2011 - 2014 Cisco Systems Inc
4 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
5 * Licensed under the GPL
6 * Derived (i.e. mostly copied) from arch/i386/kernel/irq.c:
7 * Copyright (C) 1992, 1998 Linus Torvalds, Ingo Molnar
10 #include <linux/cpumask.h>
11 #include <linux/hardirq.h>
12 #include <linux/interrupt.h>
13 #include <linux/kernel_stat.h>
14 #include <linux/module.h>
15 #include <linux/sched.h>
16 #include <linux/seq_file.h>
17 #include <linux/slab.h>
18 #include <as-layout.h>
19 #include <kern_util.h>
24 /* When epoll triggers we do not know why it did so
25 * we can also have different IRQs for read and write.
26 * This is why we keep a small irq_fd array for each fd -
27 * one entry per IRQ type
31 struct irq_entry *next;
33 struct irq_fd *irq_array[MAX_IRQ_TYPE + 1];
36 static struct irq_entry *active_fds;
38 static DEFINE_SPINLOCK(irq_lock);
40 static void irq_io_loop(struct irq_fd *irq, struct uml_pt_regs *regs)
43 * irq->active guards against reentry
44 * irq->pending accumulates pending requests
45 * if pending is raised the irq_handler is re-run
46 * until pending is cleared
52 do_IRQ(irq->irq, regs);
53 } while (irq->pending && (!irq->purge));
61 void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
63 struct irq_entry *irq_entry;
69 /* This is now lockless - epoll keeps back-referencesto the irqs
70 * which have trigger it so there is no need to walk the irq
71 * list and lock it every time. We avoid locking by turning off
72 * IO for a specific fd by executing os_del_epoll_fd(fd) before
73 * we do any changes to the actual data structures
75 n = os_waiting_for_events_epoll();
84 for (i = 0; i < n ; i++) {
85 /* Epoll back reference is the entry with 3 irq_fd
86 * leaves - one for each irq type.
88 irq_entry = (struct irq_entry *)
89 os_epoll_get_data_pointer(i);
90 for (j = 0; j < MAX_IRQ_TYPE ; j++) {
91 irq = irq_entry->irq_array[j];
94 if (os_epoll_triggered(i, irq->events) > 0)
95 irq_io_loop(irq, regs);
97 irq_entry->irq_array[j] = NULL;
105 static int assign_epoll_events_to_irq(struct irq_entry *irq_entry)
111 for (i = 0; i < MAX_IRQ_TYPE ; i++) {
112 irq = irq_entry->irq_array[i];
114 events = irq->events | events;
117 /* os_add_epoll will call os_mod_epoll if this already exists */
118 return os_add_epoll_fd(events, irq_entry->fd, irq_entry);
120 /* No events - delete */
121 return os_del_epoll_fd(irq_entry->fd);
126 static int activate_fd(int irq, int fd, int type, void *dev_id)
128 struct irq_fd *new_fd;
129 struct irq_entry *irq_entry;
133 err = os_set_fd_async(fd);
137 spin_lock_irqsave(&irq_lock, flags);
139 /* Check if we have an entry for this fd */
142 for (irq_entry = active_fds;
143 irq_entry != NULL; irq_entry = irq_entry->next) {
144 if (irq_entry->fd == fd)
148 if (irq_entry == NULL) {
149 /* This needs to be atomic as it may be called from an
152 irq_entry = kmalloc(sizeof(struct irq_entry), GFP_ATOMIC);
153 if (irq_entry == NULL) {
155 "Failed to allocate new IRQ entry\n");
159 for (i = 0; i < MAX_IRQ_TYPE; i++)
160 irq_entry->irq_array[i] = NULL;
161 irq_entry->next = active_fds;
162 active_fds = irq_entry;
165 /* Check if we are trying to re-register an interrupt for a
169 if (irq_entry->irq_array[type] != NULL) {
171 "Trying to reregister IRQ %d FD %d TYPE %d ID %p\n",
172 irq, fd, type, dev_id
176 /* New entry for this fd */
179 new_fd = kmalloc(sizeof(struct irq_fd), GFP_ATOMIC);
183 events = os_event_mask(type);
185 *new_fd = ((struct irq_fd) {
194 /* Turn off any IO on this fd - allows us to
195 * avoid locking the IRQ loop
197 os_del_epoll_fd(irq_entry->fd);
198 irq_entry->irq_array[type] = new_fd;
201 /* Turn back IO on with the correct (new) IO event mask */
202 assign_epoll_events_to_irq(irq_entry);
203 spin_unlock_irqrestore(&irq_lock, flags);
204 maybe_sigio_broken(fd, (type != IRQ_NONE));
208 spin_unlock_irqrestore(&irq_lock, flags);
214 * Walk the IRQ list and dispose of any unused entries.
215 * Should be done under irq_lock.
218 static void garbage_collect_irq_entries(void)
222 struct irq_entry *walk;
223 struct irq_entry *previous = NULL;
224 struct irq_entry *to_free;
226 if (active_fds == NULL)
229 while (walk != NULL) {
231 for (i = 0; i < MAX_IRQ_TYPE ; i++) {
232 if (walk->irq_array[i] != NULL) {
238 if (previous == NULL)
239 active_fds = walk->next;
241 previous->next = walk->next;
252 * Walk the IRQ list and get the descriptor for our FD
255 static struct irq_entry *get_irq_entry_by_fd(int fd)
257 struct irq_entry *walk = active_fds;
259 while (walk != NULL) {
269 * Walk the IRQ list and dispose of an entry for a specific
270 * device, fd and number. Note - if sharing an IRQ for read
271 * and writefor the same FD it will be disposed in either case.
272 * If this behaviour is undesirable use different IRQ ids.
276 #define IGNORE_DEV (1<<1)
278 static void do_free_by_irq_and_dev(
279 struct irq_entry *irq_entry,
286 struct irq_fd *to_free;
288 for (i = 0; i < MAX_IRQ_TYPE ; i++) {
289 if (irq_entry->irq_array[i] != NULL) {
291 ((flags & IGNORE_IRQ) ||
292 (irq_entry->irq_array[i]->irq == irq)) &&
293 ((flags & IGNORE_DEV) ||
294 (irq_entry->irq_array[i]->id == dev))
296 /* Turn off any IO on this fd - allows us to
297 * avoid locking the IRQ loop
299 os_del_epoll_fd(irq_entry->fd);
300 to_free = irq_entry->irq_array[i];
301 irq_entry->irq_array[i] = NULL;
302 assign_epoll_events_to_irq(irq_entry);
304 to_free->purge = true;
312 void free_irq_by_fd(int fd)
314 struct irq_entry *to_free;
317 spin_lock_irqsave(&irq_lock, flags);
318 to_free = get_irq_entry_by_fd(fd);
319 if (to_free != NULL) {
320 do_free_by_irq_and_dev(
324 IGNORE_IRQ | IGNORE_DEV
327 garbage_collect_irq_entries();
328 spin_unlock_irqrestore(&irq_lock, flags);
330 EXPORT_SYMBOL(free_irq_by_fd);
332 static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
334 struct irq_entry *to_free;
337 spin_lock_irqsave(&irq_lock, flags);
338 to_free = active_fds;
339 while (to_free != NULL) {
340 do_free_by_irq_and_dev(
346 to_free = to_free->next;
348 garbage_collect_irq_entries();
349 spin_unlock_irqrestore(&irq_lock, flags);
353 void reactivate_fd(int fd, int irqnum)
355 /** NOP - we do auto-EOI now **/
358 void deactivate_fd(int fd, int irqnum)
360 struct irq_entry *to_free;
364 spin_lock_irqsave(&irq_lock, flags);
365 to_free = get_irq_entry_by_fd(fd);
366 if (to_free != NULL) {
367 do_free_by_irq_and_dev(
374 garbage_collect_irq_entries();
375 spin_unlock_irqrestore(&irq_lock, flags);
378 EXPORT_SYMBOL(deactivate_fd);
381 * Called just before shutdown in order to provide a clean exec
382 * environment in case the system is rebooting. No locking because
383 * that would cause a pointless shutdown hang if something hadn't
386 int deactivate_all_fds(void)
389 struct irq_entry *to_free;
391 spin_lock_irqsave(&irq_lock, flags);
392 /* Stop IO. The IRQ loop has no lock so this is our
393 * only way of making sure we are safe to dispose
394 * of all IRQ handlers
397 to_free = active_fds;
398 while (to_free != NULL) {
399 do_free_by_irq_and_dev(
403 IGNORE_IRQ | IGNORE_DEV
405 to_free = to_free->next;
407 garbage_collect_irq_entries();
408 spin_unlock_irqrestore(&irq_lock, flags);
414 * do_IRQ handles all normal device IRQs (the special
415 * SMP cross-CPU interrupts have their own specific
418 unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
420 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
422 generic_handle_irq(irq);
424 set_irq_regs(old_regs);
428 void um_free_irq(unsigned int irq, void *dev)
430 free_irq_by_irq_and_dev(irq, dev);
433 EXPORT_SYMBOL(um_free_irq);
435 int um_request_irq(unsigned int irq, int fd, int type,
436 irq_handler_t handler,
437 unsigned long irqflags, const char * devname,
443 err = activate_fd(irq, fd, type, dev_id);
448 return request_irq(irq, handler, irqflags, devname, dev_id);
451 EXPORT_SYMBOL(um_request_irq);
452 EXPORT_SYMBOL(reactivate_fd);
455 * irq_chip must define at least enable/disable and ack when
456 * the edge handler is used.
458 static void dummy(struct irq_data *d)
462 /* This is used for everything else than the timer. */
463 static struct irq_chip normal_irq_type = {
465 .irq_disable = dummy,
472 static struct irq_chip SIGVTALRM_irq_type = {
474 .irq_disable = dummy,
481 void __init init_IRQ(void)
485 irq_set_chip_and_handler(TIMER_IRQ, &SIGVTALRM_irq_type, handle_edge_irq);
488 for (i = 1; i < NR_IRQS; i++)
489 irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
490 /* Initialize EPOLL Loop */
495 * IRQ stack entry and exit:
497 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack
498 * and switch over to the IRQ stack after some preparation. We use
499 * sigaltstack to receive signals on a separate stack from the start.
500 * These two functions make sure the rest of the kernel won't be too
501 * upset by being on a different stack. The IRQ stack has a
502 * thread_info structure at the bottom so that current et al continue
505 * to_irq_stack copies the current task's thread_info to the IRQ stack
506 * thread_info and sets the tasks's stack to point to the IRQ stack.
508 * from_irq_stack copies the thread_info struct back (flags may have
509 * been modified) and resets the task's stack pointer.
513 * What happens when two signals race each other? UML doesn't block
514 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal
515 * could arrive while a previous one is still setting up the
518 * There are three cases -
519 * The first interrupt on the stack - sets up the thread_info and
520 * handles the interrupt
521 * A nested interrupt interrupting the copying of the thread_info -
522 * can't handle the interrupt, as the stack is in an unknown state
523 * A nested interrupt not interrupting the copying of the
524 * thread_info - doesn't do any setup, just handles the interrupt
526 * The first job is to figure out whether we interrupted stack setup.
527 * This is done by xchging the signal mask with thread_info->pending.
528 * If the value that comes back is zero, then there is no setup in
529 * progress, and the interrupt can be handled. If the value is
530 * non-zero, then there is stack setup in progress. In order to have
531 * the interrupt handled, we leave our signal in the mask, and it will
532 * be handled by the upper handler after it has set up the stack.
534 * Next is to figure out whether we are the outer handler or a nested
535 * one. As part of setting up the stack, thread_info->real_thread is
536 * set to non-NULL (and is reset to NULL on exit). This is the
537 * nesting indicator. If it is non-NULL, then the stack is already
538 * set up and the handler can run.
541 static unsigned long pending_mask;
543 unsigned long to_irq_stack(unsigned long *mask_out)
545 struct thread_info *ti;
546 unsigned long mask, old;
549 mask = xchg(&pending_mask, *mask_out);
552 * If any interrupts come in at this point, we want to
553 * make sure that their bits aren't lost by our
554 * putting our bit in. So, this loop accumulates bits
555 * until xchg returns the same value that we put in.
556 * When that happens, there were no new interrupts,
557 * and pending_mask contains a bit for each interrupt
563 mask = xchg(&pending_mask, old);
564 } while (mask != old);
568 ti = current_thread_info();
569 nested = (ti->real_thread != NULL);
571 struct task_struct *task;
572 struct thread_info *tti;
574 task = cpu_tasks[ti->cpu].task;
575 tti = task_thread_info(task);
578 ti->real_thread = tti;
582 mask = xchg(&pending_mask, 0);
583 *mask_out |= mask | nested;
587 unsigned long from_irq_stack(int nested)
589 struct thread_info *ti, *to;
592 ti = current_thread_info();
596 to = ti->real_thread;
598 ti->real_thread = NULL;
601 mask = xchg(&pending_mask, 0);