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 */
31 struct 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
47 struct 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
55 static DEFINE_SPINLOCK(irq_lock);
56 static LIST_HEAD(active_fds);
57 static DECLARE_BITMAP(irqs_allocated, UM_LAST_SIGNAL_IRQ);
58 static bool irqs_suspended;
59 #ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
60 static bool irqs_pending;
61 #endif
62
63 static 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
86 static 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
100 static 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
122 static 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
151 static bool irq_do_timetravel_handler(struct irq_entry *entry,
152 enum um_irq_type t)
153 {
154 return false;
155 }
156
157 static void irq_do_pending_events(bool timetravel_handlers_only)
158 {
159 }
160 #endif
161
162 static 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
195 static 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
239 void 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
246 static 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
260 static 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
271 static 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
289 static 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
295 static 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;
345 out_unlock:
346 spin_unlock_irqrestore(&irq_lock, flags);
347 out:
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 */
356 void 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 }
366 EXPORT_SYMBOL(free_irq_by_fd);
367
368 static 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 }
393 out:
394 spin_unlock_irqrestore(&irq_lock, flags);
395 }
396
397 void 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);
418 out:
419 spin_unlock_irqrestore(&irq_lock, flags);
420
421 ignore_sigio_fd(fd);
422 }
423 EXPORT_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 */
431 int 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 */
453 unsigned 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
463 void 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 }
473 EXPORT_SYMBOL(um_free_irq);
474
475 static 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;
509 error:
510 clear_bit(irq, irqs_allocated);
511 return err;
512 }
513
514 int 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 }
521 EXPORT_SYMBOL(um_request_irq);
522
523 #ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
524 int 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 }
533 EXPORT_SYMBOL(um_request_irq_tt);
534
535 void sigio_run_timetravel_handlers(void)
536 {
537 _sigio_handler(NULL, true);
538 }
539 #endif
540
541 #ifdef CONFIG_PM_SLEEP
542 void 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
585 void 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
611 static 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 }
630 unlock:
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 */
642 static void dummy(struct irq_data *d)
643 {
644 }
645
646 /* This is used for everything other than the timer. */
647 static 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
657 static 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
666 void __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 }
677
678 /*
679 * IRQ stack entry and exit:
680 *
681 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack
682 * and switch over to the IRQ stack after some preparation. We use
683 * sigaltstack to receive signals on a separate stack from the start.
684 * These two functions make sure the rest of the kernel won't be too
685 * upset by being on a different stack. The IRQ stack has a
686 * thread_info structure at the bottom so that current et al continue
687 * to work.
688 *
689 * to_irq_stack copies the current task's thread_info to the IRQ stack
690 * thread_info and sets the tasks's stack to point to the IRQ stack.
691 *
692 * from_irq_stack copies the thread_info struct back (flags may have
693 * been modified) and resets the task's stack pointer.
694 *
695 * Tricky bits -
696 *
697 * What happens when two signals race each other? UML doesn't block
698 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal
699 * could arrive while a previous one is still setting up the
700 * thread_info.
701 *
702 * There are three cases -
703 * The first interrupt on the stack - sets up the thread_info and
704 * handles the interrupt
705 * A nested interrupt interrupting the copying of the thread_info -
706 * can't handle the interrupt, as the stack is in an unknown state
707 * A nested interrupt not interrupting the copying of the
708 * thread_info - doesn't do any setup, just handles the interrupt
709 *
710 * The first job is to figure out whether we interrupted stack setup.
711 * This is done by xchging the signal mask with thread_info->pending.
712 * If the value that comes back is zero, then there is no setup in
713 * progress, and the interrupt can be handled. If the value is
714 * non-zero, then there is stack setup in progress. In order to have
715 * the interrupt handled, we leave our signal in the mask, and it will
716 * be handled by the upper handler after it has set up the stack.
717 *
718 * Next is to figure out whether we are the outer handler or a nested
719 * one. As part of setting up the stack, thread_info->real_thread is
720 * set to non-NULL (and is reset to NULL on exit). This is the
721 * nesting indicator. If it is non-NULL, then the stack is already
722 * set up and the handler can run.
723 */
724
725 static unsigned long pending_mask;
726
727 unsigned long to_irq_stack(unsigned long *mask_out)
728 {
729 struct thread_info *ti;
730 unsigned long mask, old;
731 int nested;
732
733 mask = xchg(&pending_mask, *mask_out);
734 if (mask != 0) {
735 /*
736 * If any interrupts come in at this point, we want to
737 * make sure that their bits aren't lost by our
738 * putting our bit in. So, this loop accumulates bits
739 * until xchg returns the same value that we put in.
740 * When that happens, there were no new interrupts,
741 * and pending_mask contains a bit for each interrupt
742 * that came in.
743 */
744 old = *mask_out;
745 do {
746 old |= mask;
747 mask = xchg(&pending_mask, old);
748 } while (mask != old);
749 return 1;
750 }
751
752 ti = current_thread_info();
753 nested = (ti->real_thread != NULL);
754 if (!nested) {
755 struct task_struct *task;
756 struct thread_info *tti;
757
758 task = cpu_tasks[ti->cpu].task;
759 tti = task_thread_info(task);
760
761 *ti = *tti;
762 ti->real_thread = tti;
763 task->stack = ti;
764 }
765
766 mask = xchg(&pending_mask, 0);
767 *mask_out |= mask | nested;
768 return 0;
769 }
770
771 unsigned long from_irq_stack(int nested)
772 {
773 struct thread_info *ti, *to;
774 unsigned long mask;
775
776 ti = current_thread_info();
777
778 pending_mask = 1;
779
780 to = ti->real_thread;
781 current->stack = to;
782 ti->real_thread = NULL;
783 *to = *ti;
784
785 mask = xchg(&pending_mask, 0);
786 return mask & ~1;
787 }
788
789
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.