Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2017 - Cambridge Greys Ltd
4 * Copyright (C) 2011 - 2014 Cisco Systems Inc
5 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
6 * Derived (i.e. mostly copied) from arch/i386/kernel/irq.c:
7 * Copyright (C) 1992, 1998 Linus Torvalds, Ingo Molnar
8 */
9
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>
20#include <os.h>
21#include <irq_user.h>
22#include <irq_kern.h>
23#include <linux/time-internal.h>
24
25
26/* When epoll triggers we do not know why it did so
27 * we can also have different IRQs for read and write.
28 * This is why we keep a small irq_reg array for each fd -
29 * one entry per IRQ type
30 */
31struct irq_reg {
32 void *id;
33 int irq;
34 /* it's cheaper to store this than to query it */
35 int events;
36 bool active;
37 bool pending;
38 bool wakeup;
39#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
40 bool pending_event;
41 void (*timetravel_handler)(int, int, void *,
42 struct time_travel_event *);
43 struct time_travel_event event;
44#endif
45};
46
47struct irq_entry {
48 struct list_head list;
49 int fd;
50 struct irq_reg reg[NUM_IRQ_TYPES];
51 bool suspended;
52 bool sigio_workaround;
53};
54
55static DEFINE_SPINLOCK(irq_lock);
56static LIST_HEAD(active_fds);
57static DECLARE_BITMAP(irqs_allocated, UM_LAST_SIGNAL_IRQ);
58static bool irqs_suspended;
59#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
60static bool irqs_pending;
61#endif
62
63static void irq_io_loop(struct irq_reg *irq, struct uml_pt_regs *regs)
64{
65/*
66 * irq->active guards against reentry
67 * irq->pending accumulates pending requests
68 * if pending is raised the irq_handler is re-run
69 * until pending is cleared
70 */
71 if (irq->active) {
72 irq->active = false;
73
74 do {
75 irq->pending = false;
76 do_IRQ(irq->irq, regs);
77 } while (irq->pending);
78
79 irq->active = true;
80 } else {
81 irq->pending = true;
82 }
83}
84
85#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
86static void irq_event_handler(struct time_travel_event *ev)
87{
88 struct irq_reg *reg = container_of(ev, struct irq_reg, event);
89
90 /* do nothing if suspended; just cause a wakeup and mark as pending */
91 if (irqs_suspended) {
92 irqs_pending = true;
93 reg->pending_event = true;
94 return;
95 }
96
97 generic_handle_irq(reg->irq);
98}
99
100static bool irq_do_timetravel_handler(struct irq_entry *entry,
101 enum um_irq_type t)
102{
103 struct irq_reg *reg = &entry->reg[t];
104
105 if (!reg->timetravel_handler)
106 return false;
107
108 /*
109 * Handle all messages - we might get multiple even while
110 * interrupts are already suspended, due to suspend order
111 * etc. Note that time_travel_add_irq_event() will not add
112 * an event twice, if it's pending already "first wins".
113 */
114 reg->timetravel_handler(reg->irq, entry->fd, reg->id, ®->event);
115
116 if (!reg->event.pending)
117 return false;
118
119 return true;
120}
121
122static void irq_do_pending_events(bool timetravel_handlers_only)
123{
124 struct irq_entry *entry;
125
126 if (!irqs_pending || timetravel_handlers_only)
127 return;
128
129 irqs_pending = false;
130
131 list_for_each_entry(entry, &active_fds, list) {
132 enum um_irq_type t;
133
134 for (t = 0; t < NUM_IRQ_TYPES; t++) {
135 struct irq_reg *reg = &entry->reg[t];
136
137 /*
138 * Any timetravel_handler was invoked already, just
139 * directly run the IRQ.
140 */
141 if (reg->pending_event) {
142 irq_enter();
143 generic_handle_irq(reg->irq);
144 irq_exit();
145 reg->pending_event = false;
146 }
147 }
148 }
149}
150#else
151static bool irq_do_timetravel_handler(struct irq_entry *entry,
152 enum um_irq_type t)
153{
154 return false;
155}
156
157static void irq_do_pending_events(bool timetravel_handlers_only)
158{
159}
160#endif
161
162static void sigio_reg_handler(int idx, struct irq_entry *entry, enum um_irq_type t,
163 struct uml_pt_regs *regs,
164 bool timetravel_handlers_only)
165{
166 struct irq_reg *reg = &entry->reg[t];
167
168 if (!reg->events)
169 return;
170
171 if (os_epoll_triggered(idx, reg->events) <= 0)
172 return;
173
174 if (irq_do_timetravel_handler(entry, t))
175 return;
176
177 /*
178 * If we're called to only run time-travel handlers then don't
179 * actually proceed but mark sigio as pending (if applicable).
180 * For suspend/resume, timetravel_handlers_only may be true
181 * despite time-travel not being configured and used.
182 */
183 if (timetravel_handlers_only) {
184#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
185 reg->pending_event = true;
186 irqs_pending = true;
187 mark_sigio_pending();
188#endif
189 return;
190 }
191
192 irq_io_loop(reg, regs);
193}
194
195static void _sigio_handler(struct uml_pt_regs *regs,
196 bool timetravel_handlers_only)
197{
198 struct irq_entry *irq_entry;
199 int n, i;
200
201 if (timetravel_handlers_only && !um_irq_timetravel_handler_used())
202 return;
203
204 /* Flush out pending events that were ignored due to time-travel. */
205 if (!irqs_suspended)
206 irq_do_pending_events(timetravel_handlers_only);
207
208 while (1) {
209 /* This is now lockless - epoll keeps back-referencesto the irqs
210 * which have trigger it so there is no need to walk the irq
211 * list and lock it every time. We avoid locking by turning off
212 * IO for a specific fd by executing os_del_epoll_fd(fd) before
213 * we do any changes to the actual data structures
214 */
215 n = os_waiting_for_events_epoll();
216
217 if (n <= 0) {
218 if (n == -EINTR)
219 continue;
220 else
221 break;
222 }
223
224 for (i = 0; i < n ; i++) {
225 enum um_irq_type t;
226
227 irq_entry = os_epoll_get_data_pointer(i);
228
229 for (t = 0; t < NUM_IRQ_TYPES; t++)
230 sigio_reg_handler(i, irq_entry, t, regs,
231 timetravel_handlers_only);
232 }
233 }
234
235 if (!timetravel_handlers_only)
236 free_irqs();
237}
238
239void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
240{
241 preempt_disable();
242 _sigio_handler(regs, irqs_suspended);
243 preempt_enable();
244}
245
246static struct irq_entry *get_irq_entry_by_fd(int fd)
247{
248 struct irq_entry *walk;
249
250 lockdep_assert_held(&irq_lock);
251
252 list_for_each_entry(walk, &active_fds, list) {
253 if (walk->fd == fd)
254 return walk;
255 }
256
257 return NULL;
258}
259
260static void free_irq_entry(struct irq_entry *to_free, bool remove)
261{
262 if (!to_free)
263 return;
264
265 if (remove)
266 os_del_epoll_fd(to_free->fd);
267 list_del(&to_free->list);
268 kfree(to_free);
269}
270
271static bool update_irq_entry(struct irq_entry *entry)
272{
273 enum um_irq_type i;
274 int events = 0;
275
276 for (i = 0; i < NUM_IRQ_TYPES; i++)
277 events |= entry->reg[i].events;
278
279 if (events) {
280 /* will modify (instead of add) if needed */
281 os_add_epoll_fd(events, entry->fd, entry);
282 return true;
283 }
284
285 os_del_epoll_fd(entry->fd);
286 return false;
287}
288
289static void update_or_free_irq_entry(struct irq_entry *entry)
290{
291 if (!update_irq_entry(entry))
292 free_irq_entry(entry, false);
293}
294
295static int activate_fd(int irq, int fd, enum um_irq_type type, void *dev_id,
296 void (*timetravel_handler)(int, int, void *,
297 struct time_travel_event *))
298{
299 struct irq_entry *irq_entry;
300 int err, events = os_event_mask(type);
301 unsigned long flags;
302
303 err = os_set_fd_async(fd);
304 if (err < 0)
305 goto out;
306
307 spin_lock_irqsave(&irq_lock, flags);
308 irq_entry = get_irq_entry_by_fd(fd);
309 if (irq_entry) {
310 /* cannot register the same FD twice with the same type */
311 if (WARN_ON(irq_entry->reg[type].events)) {
312 err = -EALREADY;
313 goto out_unlock;
314 }
315
316 /* temporarily disable to avoid IRQ-side locking */
317 os_del_epoll_fd(fd);
318 } else {
319 irq_entry = kzalloc(sizeof(*irq_entry), GFP_ATOMIC);
320 if (!irq_entry) {
321 err = -ENOMEM;
322 goto out_unlock;
323 }
324 irq_entry->fd = fd;
325 list_add_tail(&irq_entry->list, &active_fds);
326 maybe_sigio_broken(fd);
327 }
328
329 irq_entry->reg[type].id = dev_id;
330 irq_entry->reg[type].irq = irq;
331 irq_entry->reg[type].active = true;
332 irq_entry->reg[type].events = events;
333
334#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
335 if (um_irq_timetravel_handler_used()) {
336 irq_entry->reg[type].timetravel_handler = timetravel_handler;
337 irq_entry->reg[type].event.fn = irq_event_handler;
338 }
339#endif
340
341 WARN_ON(!update_irq_entry(irq_entry));
342 spin_unlock_irqrestore(&irq_lock, flags);
343
344 return 0;
345out_unlock:
346 spin_unlock_irqrestore(&irq_lock, flags);
347out:
348 return err;
349}
350
351/*
352 * Remove the entry or entries for a specific FD, if you
353 * don't want to remove all the possible entries then use
354 * um_free_irq() or deactivate_fd() instead.
355 */
356void free_irq_by_fd(int fd)
357{
358 struct irq_entry *to_free;
359 unsigned long flags;
360
361 spin_lock_irqsave(&irq_lock, flags);
362 to_free = get_irq_entry_by_fd(fd);
363 free_irq_entry(to_free, true);
364 spin_unlock_irqrestore(&irq_lock, flags);
365}
366EXPORT_SYMBOL(free_irq_by_fd);
367
368static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
369{
370 struct irq_entry *entry;
371 unsigned long flags;
372
373 spin_lock_irqsave(&irq_lock, flags);
374 list_for_each_entry(entry, &active_fds, list) {
375 enum um_irq_type i;
376
377 for (i = 0; i < NUM_IRQ_TYPES; i++) {
378 struct irq_reg *reg = &entry->reg[i];
379
380 if (!reg->events)
381 continue;
382 if (reg->irq != irq)
383 continue;
384 if (reg->id != dev)
385 continue;
386
387 os_del_epoll_fd(entry->fd);
388 reg->events = 0;
389 update_or_free_irq_entry(entry);
390 goto out;
391 }
392 }
393out:
394 spin_unlock_irqrestore(&irq_lock, flags);
395}
396
397void deactivate_fd(int fd, int irqnum)
398{
399 struct irq_entry *entry;
400 unsigned long flags;
401 enum um_irq_type i;
402
403 os_del_epoll_fd(fd);
404
405 spin_lock_irqsave(&irq_lock, flags);
406 entry = get_irq_entry_by_fd(fd);
407 if (!entry)
408 goto out;
409
410 for (i = 0; i < NUM_IRQ_TYPES; i++) {
411 if (!entry->reg[i].events)
412 continue;
413 if (entry->reg[i].irq == irqnum)
414 entry->reg[i].events = 0;
415 }
416
417 update_or_free_irq_entry(entry);
418out:
419 spin_unlock_irqrestore(&irq_lock, flags);
420
421 ignore_sigio_fd(fd);
422}
423EXPORT_SYMBOL(deactivate_fd);
424
425/*
426 * Called just before shutdown in order to provide a clean exec
427 * environment in case the system is rebooting. No locking because
428 * that would cause a pointless shutdown hang if something hadn't
429 * released the lock.
430 */
431int deactivate_all_fds(void)
432{
433 struct irq_entry *entry;
434
435 /* Stop IO. The IRQ loop has no lock so this is our
436 * only way of making sure we are safe to dispose
437 * of all IRQ handlers
438 */
439 os_set_ioignore();
440
441 /* we can no longer call kfree() here so just deactivate */
442 list_for_each_entry(entry, &active_fds, list)
443 os_del_epoll_fd(entry->fd);
444 os_close_epoll_fd();
445 return 0;
446}
447
448/*
449 * do_IRQ handles all normal device IRQs (the special
450 * SMP cross-CPU interrupts have their own specific
451 * handlers).
452 */
453unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
454{
455 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
456 irq_enter();
457 generic_handle_irq(irq);
458 irq_exit();
459 set_irq_regs(old_regs);
460 return 1;
461}
462
463void um_free_irq(int irq, void *dev)
464{
465 if (WARN(irq < 0 || irq > UM_LAST_SIGNAL_IRQ,
466 "freeing invalid irq %d", irq))
467 return;
468
469 free_irq_by_irq_and_dev(irq, dev);
470 free_irq(irq, dev);
471 clear_bit(irq, irqs_allocated);
472}
473EXPORT_SYMBOL(um_free_irq);
474
475static int
476_um_request_irq(int irq, int fd, enum um_irq_type type,
477 irq_handler_t handler, unsigned long irqflags,
478 const char *devname, void *dev_id,
479 void (*timetravel_handler)(int, int, void *,
480 struct time_travel_event *))
481{
482 int err;
483
484 if (irq == UM_IRQ_ALLOC) {
485 int i;
486
487 for (i = UM_FIRST_DYN_IRQ; i < NR_IRQS; i++) {
488 if (!test_and_set_bit(i, irqs_allocated)) {
489 irq = i;
490 break;
491 }
492 }
493 }
494
495 if (irq < 0)
496 return -ENOSPC;
497
498 if (fd != -1) {
499 err = activate_fd(irq, fd, type, dev_id, timetravel_handler);
500 if (err)
501 goto error;
502 }
503
504 err = request_irq(irq, handler, irqflags, devname, dev_id);
505 if (err < 0)
506 goto error;
507
508 return irq;
509error:
510 clear_bit(irq, irqs_allocated);
511 return err;
512}
513
514int um_request_irq(int irq, int fd, enum um_irq_type type,
515 irq_handler_t handler, unsigned long irqflags,
516 const char *devname, void *dev_id)
517{
518 return _um_request_irq(irq, fd, type, handler, irqflags,
519 devname, dev_id, NULL);
520}
521EXPORT_SYMBOL(um_request_irq);
522
523#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
524int um_request_irq_tt(int irq, int fd, enum um_irq_type type,
525 irq_handler_t handler, unsigned long irqflags,
526 const char *devname, void *dev_id,
527 void (*timetravel_handler)(int, int, void *,
528 struct time_travel_event *))
529{
530 return _um_request_irq(irq, fd, type, handler, irqflags,
531 devname, dev_id, timetravel_handler);
532}
533EXPORT_SYMBOL(um_request_irq_tt);
534
535void sigio_run_timetravel_handlers(void)
536{
537 _sigio_handler(NULL, true);
538}
539#endif
540
541#ifdef CONFIG_PM_SLEEP
542void um_irqs_suspend(void)
543{
544 struct irq_entry *entry;
545 unsigned long flags;
546
547 irqs_suspended = true;
548
549 spin_lock_irqsave(&irq_lock, flags);
550 list_for_each_entry(entry, &active_fds, list) {
551 enum um_irq_type t;
552 bool clear = true;
553
554 for (t = 0; t < NUM_IRQ_TYPES; t++) {
555 if (!entry->reg[t].events)
556 continue;
557
558 /*
559 * For the SIGIO_WRITE_IRQ, which is used to handle the
560 * SIGIO workaround thread, we need special handling:
561 * enable wake for it itself, but below we tell it about
562 * any FDs that should be suspended.
563 */
564 if (entry->reg[t].wakeup ||
565 entry->reg[t].irq == SIGIO_WRITE_IRQ
566#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
567 || entry->reg[t].timetravel_handler
568#endif
569 ) {
570 clear = false;
571 break;
572 }
573 }
574
575 if (clear) {
576 entry->suspended = true;
577 os_clear_fd_async(entry->fd);
578 entry->sigio_workaround =
579 !__ignore_sigio_fd(entry->fd);
580 }
581 }
582 spin_unlock_irqrestore(&irq_lock, flags);
583}
584
585void um_irqs_resume(void)
586{
587 struct irq_entry *entry;
588 unsigned long flags;
589
590
591 spin_lock_irqsave(&irq_lock, flags);
592 list_for_each_entry(entry, &active_fds, list) {
593 if (entry->suspended) {
594 int err = os_set_fd_async(entry->fd);
595
596 WARN(err < 0, "os_set_fd_async returned %d\n", err);
597 entry->suspended = false;
598
599 if (entry->sigio_workaround) {
600 err = __add_sigio_fd(entry->fd);
601 WARN(err < 0, "add_sigio_returned %d\n", err);
602 }
603 }
604 }
605 spin_unlock_irqrestore(&irq_lock, flags);
606
607 irqs_suspended = false;
608 send_sigio_to_self();
609}
610
611static int normal_irq_set_wake(struct irq_data *d, unsigned int on)
612{
613 struct irq_entry *entry;
614 unsigned long flags;
615
616 spin_lock_irqsave(&irq_lock, flags);
617 list_for_each_entry(entry, &active_fds, list) {
618 enum um_irq_type t;
619
620 for (t = 0; t < NUM_IRQ_TYPES; t++) {
621 if (!entry->reg[t].events)
622 continue;
623
624 if (entry->reg[t].irq != d->irq)
625 continue;
626 entry->reg[t].wakeup = on;
627 goto unlock;
628 }
629 }
630unlock:
631 spin_unlock_irqrestore(&irq_lock, flags);
632 return 0;
633}
634#else
635#define normal_irq_set_wake NULL
636#endif
637
638/*
639 * irq_chip must define at least enable/disable and ack when
640 * the edge handler is used.
641 */
642static void dummy(struct irq_data *d)
643{
644}
645
646/* This is used for everything other than the timer. */
647static struct irq_chip normal_irq_type = {
648 .name = "SIGIO",
649 .irq_disable = dummy,
650 .irq_enable = dummy,
651 .irq_ack = dummy,
652 .irq_mask = dummy,
653 .irq_unmask = dummy,
654 .irq_set_wake = normal_irq_set_wake,
655};
656
657static struct irq_chip alarm_irq_type = {
658 .name = "SIGALRM",
659 .irq_disable = dummy,
660 .irq_enable = dummy,
661 .irq_ack = dummy,
662 .irq_mask = dummy,
663 .irq_unmask = dummy,
664};
665
666void __init init_IRQ(void)
667{
668 int i;
669
670 irq_set_chip_and_handler(TIMER_IRQ, &alarm_irq_type, handle_edge_irq);
671
672 for (i = 1; i < UM_LAST_SIGNAL_IRQ; i++)
673 irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
674 /* Initialize EPOLL Loop */
675 os_setup_epoll();
676}
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 * 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.
27 */
28static struct irq_fd *active_fds = NULL;
29static struct irq_fd **last_irq_ptr = &active_fds;
30
31extern void free_irqs(void);
32
33void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
34{
35 struct irq_fd *irq_fd;
36 int n;
37
38 while (1) {
39 n = os_waiting_for_events(active_fds);
40 if (n <= 0) {
41 if (n == -EINTR)
42 continue;
43 else break;
44 }
45
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);
51 }
52 }
53 }
54
55 free_irqs();
56}
57
58static DEFINE_SPINLOCK(irq_lock);
59
60static int activate_fd(int irq, int fd, int type, void *dev_id)
61{
62 struct pollfd *tmp_pfd;
63 struct irq_fd *new_fd, *irq_fd;
64 unsigned long flags;
65 int events, err, n;
66
67 err = os_set_fd_async(fd);
68 if (err < 0)
69 goto out;
70
71 err = -ENOMEM;
72 new_fd = kmalloc(sizeof(struct irq_fd), GFP_KERNEL);
73 if (new_fd == NULL)
74 goto out;
75
76 if (type == IRQ_READ)
77 events = UM_POLLIN | UM_POLLPRI;
78 else events = UM_POLLOUT;
79 *new_fd = ((struct irq_fd) { .next = NULL,
80 .id = dev_id,
81 .fd = fd,
82 .type = type,
83 .irq = irq,
84 .events = events,
85 .current_events = 0 } );
86
87 err = -EBUSY;
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,
94 dev_id);
95 goto out_unlock;
96 }
97 }
98
99 if (type == IRQ_WRITE)
100 fd = -1;
101
102 tmp_pfd = NULL;
103 n = 0;
104
105 while (1) {
106 n = os_create_pollfd(fd, events, tmp_pfd, n);
107 if (n == 0)
108 break;
109
110 /*
111 * n > 0
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.
115 *
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.
121 */
122 spin_unlock_irqrestore(&irq_lock, flags);
123 kfree(tmp_pfd);
124
125 tmp_pfd = kmalloc(n, GFP_KERNEL);
126 if (tmp_pfd == NULL)
127 goto out_kfree;
128
129 spin_lock_irqsave(&irq_lock, flags);
130 }
131
132 *last_irq_ptr = new_fd;
133 last_irq_ptr = &new_fd->next;
134
135 spin_unlock_irqrestore(&irq_lock, flags);
136
137 /*
138 * This calls activate_fd, so it has to be outside the critical
139 * section.
140 */
141 maybe_sigio_broken(fd, (type == IRQ_READ));
142
143 return 0;
144
145 out_unlock:
146 spin_unlock_irqrestore(&irq_lock, flags);
147 out_kfree:
148 kfree(new_fd);
149 out:
150 return err;
151}
152
153static void free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg)
154{
155 unsigned long flags;
156
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);
160}
161
162struct irq_and_dev {
163 int irq;
164 void *dev;
165};
166
167static int same_irq_and_dev(struct irq_fd *irq, void *d)
168{
169 struct irq_and_dev *data = d;
170
171 return ((irq->irq == data->irq) && (irq->id == data->dev));
172}
173
174static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
175{
176 struct irq_and_dev data = ((struct irq_and_dev) { .irq = irq,
177 .dev = dev });
178
179 free_irq_by_cb(same_irq_and_dev, &data);
180}
181
182static int same_fd(struct irq_fd *irq, void *fd)
183{
184 return (irq->fd == *((int *)fd));
185}
186
187void free_irq_by_fd(int fd)
188{
189 free_irq_by_cb(same_fd, &fd);
190}
191
192/* Must be called with irq_lock held */
193static struct irq_fd *find_irq_by_fd(int fd, int irqnum, int *index_out)
194{
195 struct irq_fd *irq;
196 int i = 0;
197 int fdi;
198
199 for (irq = active_fds; irq != NULL; irq = irq->next) {
200 if ((irq->fd == fd) && (irq->irq == irqnum))
201 break;
202 i++;
203 }
204 if (irq == NULL) {
205 printk(KERN_ERR "find_irq_by_fd doesn't have descriptor %d\n",
206 fd);
207 goto out;
208 }
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,
213 fdi, fd);
214 irq = NULL;
215 goto out;
216 }
217 *index_out = i;
218 out:
219 return irq;
220}
221
222void reactivate_fd(int fd, int irqnum)
223{
224 struct irq_fd *irq;
225 unsigned long flags;
226 int i;
227
228 spin_lock_irqsave(&irq_lock, flags);
229 irq = find_irq_by_fd(fd, irqnum, &i);
230 if (irq == NULL) {
231 spin_unlock_irqrestore(&irq_lock, flags);
232 return;
233 }
234 os_set_pollfd(i, irq->fd);
235 spin_unlock_irqrestore(&irq_lock, flags);
236
237 add_sigio_fd(fd);
238}
239
240void deactivate_fd(int fd, int irqnum)
241{
242 struct irq_fd *irq;
243 unsigned long flags;
244 int i;
245
246 spin_lock_irqsave(&irq_lock, flags);
247 irq = find_irq_by_fd(fd, irqnum, &i);
248 if (irq == NULL) {
249 spin_unlock_irqrestore(&irq_lock, flags);
250 return;
251 }
252
253 os_set_pollfd(i, -1);
254 spin_unlock_irqrestore(&irq_lock, flags);
255
256 ignore_sigio_fd(fd);
257}
258EXPORT_SYMBOL(deactivate_fd);
259
260/*
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
264 * released the lock.
265 */
266int deactivate_all_fds(void)
267{
268 struct irq_fd *irq;
269 int err;
270
271 for (irq = active_fds; irq != NULL; irq = irq->next) {
272 err = os_clear_fd_async(irq->fd);
273 if (err)
274 return err;
275 }
276 /* If there is a signal already queued, after unblocking ignore it */
277 os_set_ioignore();
278
279 return 0;
280}
281
282/*
283 * do_IRQ handles all normal device IRQs (the special
284 * SMP cross-CPU interrupts have their own specific
285 * handlers).
286 */
287unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
288{
289 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
290 irq_enter();
291 generic_handle_irq(irq);
292 irq_exit();
293 set_irq_regs(old_regs);
294 return 1;
295}
296
297void um_free_irq(unsigned int irq, void *dev)
298{
299 free_irq_by_irq_and_dev(irq, dev);
300 free_irq(irq, dev);
301}
302EXPORT_SYMBOL(um_free_irq);
303
304int um_request_irq(unsigned int irq, int fd, int type,
305 irq_handler_t handler,
306 unsigned long irqflags, const char * devname,
307 void *dev_id)
308{
309 int err;
310
311 if (fd != -1) {
312 err = activate_fd(irq, fd, type, dev_id);
313 if (err)
314 return err;
315 }
316
317 return request_irq(irq, handler, irqflags, devname, dev_id);
318}
319
320EXPORT_SYMBOL(um_request_irq);
321EXPORT_SYMBOL(reactivate_fd);
322
323/*
324 * irq_chip must define at least enable/disable and ack when
325 * the edge handler is used.
326 */
327static void dummy(struct irq_data *d)
328{
329}
330
331/* This is used for everything else than the timer. */
332static struct irq_chip normal_irq_type = {
333 .name = "SIGIO",
334 .irq_disable = dummy,
335 .irq_enable = dummy,
336 .irq_ack = dummy,
337 .irq_mask = dummy,
338 .irq_unmask = dummy,
339};
340
341static struct irq_chip SIGVTALRM_irq_type = {
342 .name = "SIGVTALRM",
343 .irq_disable = dummy,
344 .irq_enable = dummy,
345 .irq_ack = dummy,
346 .irq_mask = dummy,
347 .irq_unmask = dummy,
348};
349
350void __init init_IRQ(void)
351{
352 int i;
353
354 irq_set_chip_and_handler(TIMER_IRQ, &SIGVTALRM_irq_type, handle_edge_irq);
355
356 for (i = 1; i < NR_IRQS; i++)
357 irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
358}
359
360/*
361 * IRQ stack entry and exit:
362 *
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
369 * to work.
370 *
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.
373 *
374 * from_irq_stack copies the thread_info struct back (flags may have
375 * been modified) and resets the task's stack pointer.
376 *
377 * Tricky bits -
378 *
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
382 * thread_info.
383 *
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
391 *
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.
399 *
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.
405 */
406
407static unsigned long pending_mask;
408
409unsigned long to_irq_stack(unsigned long *mask_out)
410{
411 struct thread_info *ti;
412 unsigned long mask, old;
413 int nested;
414
415 mask = xchg(&pending_mask, *mask_out);
416 if (mask != 0) {
417 /*
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
424 * that came in.
425 */
426 old = *mask_out;
427 do {
428 old |= mask;
429 mask = xchg(&pending_mask, old);
430 } while (mask != old);
431 return 1;
432 }
433
434 ti = current_thread_info();
435 nested = (ti->real_thread != NULL);
436 if (!nested) {
437 struct task_struct *task;
438 struct thread_info *tti;
439
440 task = cpu_tasks[ti->cpu].task;
441 tti = task_thread_info(task);
442
443 *ti = *tti;
444 ti->real_thread = tti;
445 task->stack = ti;
446 }
447
448 mask = xchg(&pending_mask, 0);
449 *mask_out |= mask | nested;
450 return 0;
451}
452
453unsigned long from_irq_stack(int nested)
454{
455 struct thread_info *ti, *to;
456 unsigned long mask;
457
458 ti = current_thread_info();
459
460 pending_mask = 1;
461
462 to = ti->real_thread;
463 current->stack = to;
464 ti->real_thread = NULL;
465 *to = *ti;
466
467 mask = xchg(&pending_mask, 0);
468 return mask & ~1;
469}
470