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_on_resume;
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
60static void irq_io_loop(struct irq_reg *irq, struct uml_pt_regs *regs)
61{
62/*
63 * irq->active guards against reentry
64 * irq->pending accumulates pending requests
65 * if pending is raised the irq_handler is re-run
66 * until pending is cleared
67 */
68 if (irq->active) {
69 irq->active = false;
70
71 do {
72 irq->pending = false;
73 do_IRQ(irq->irq, regs);
74 } while (irq->pending);
75
76 irq->active = true;
77 } else {
78 irq->pending = true;
79 }
80}
81
82#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
83static void irq_event_handler(struct time_travel_event *ev)
84{
85 struct irq_reg *reg = container_of(ev, struct irq_reg, event);
86
87 /* do nothing if suspended - just to cause a wakeup */
88 if (irqs_suspended)
89 return;
90
91 generic_handle_irq(reg->irq);
92}
93
94static bool irq_do_timetravel_handler(struct irq_entry *entry,
95 enum um_irq_type t)
96{
97 struct irq_reg *reg = &entry->reg[t];
98
99 if (!reg->timetravel_handler)
100 return false;
101
102 /*
103 * Handle all messages - we might get multiple even while
104 * interrupts are already suspended, due to suspend order
105 * etc. Note that time_travel_add_irq_event() will not add
106 * an event twice, if it's pending already "first wins".
107 */
108 reg->timetravel_handler(reg->irq, entry->fd, reg->id, &reg->event);
109
110 if (!reg->event.pending)
111 return false;
112
113 if (irqs_suspended)
114 reg->pending_on_resume = true;
115 return true;
116}
117#else
118static bool irq_do_timetravel_handler(struct irq_entry *entry,
119 enum um_irq_type t)
120{
121 return false;
122}
123#endif
124
125static void sigio_reg_handler(int idx, struct irq_entry *entry, enum um_irq_type t,
126 struct uml_pt_regs *regs,
127 bool timetravel_handlers_only)
128{
129 struct irq_reg *reg = &entry->reg[t];
130
131 if (!reg->events)
132 return;
133
134 if (os_epoll_triggered(idx, reg->events) <= 0)
135 return;
136
137 if (irq_do_timetravel_handler(entry, t: t))
138 return;
139
140 /*
141 * If we're called to only run time-travel handlers then don't
142 * actually proceed but mark sigio as pending (if applicable).
143 * For suspend/resume, timetravel_handlers_only may be true
144 * despite time-travel not being configured and used.
145 */
146 if (timetravel_handlers_only) {
147#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
148 mark_sigio_pending();
149#endif
150 return;
151 }
152
153 irq_io_loop(irq: reg, regs);
154}
155
156static void _sigio_handler(struct uml_pt_regs *regs,
157 bool timetravel_handlers_only)
158{
159 struct irq_entry *irq_entry;
160 int n, i;
161
162 if (timetravel_handlers_only && !um_irq_timetravel_handler_used())
163 return;
164
165 while (1) {
166 /* This is now lockless - epoll keeps back-referencesto the irqs
167 * which have trigger it so there is no need to walk the irq
168 * list and lock it every time. We avoid locking by turning off
169 * IO for a specific fd by executing os_del_epoll_fd(fd) before
170 * we do any changes to the actual data structures
171 */
172 n = os_waiting_for_events_epoll();
173
174 if (n <= 0) {
175 if (n == -EINTR)
176 continue;
177 else
178 break;
179 }
180
181 for (i = 0; i < n ; i++) {
182 enum um_irq_type t;
183
184 irq_entry = os_epoll_get_data_pointer(i);
185
186 for (t = 0; t < NUM_IRQ_TYPES; t++)
187 sigio_reg_handler(idx: i, entry: irq_entry, t: t, regs,
188 timetravel_handlers_only);
189 }
190 }
191
192 if (!timetravel_handlers_only)
193 free_irqs();
194}
195
196void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
197{
198 _sigio_handler(regs, timetravel_handlers_only: irqs_suspended);
199}
200
201static struct irq_entry *get_irq_entry_by_fd(int fd)
202{
203 struct irq_entry *walk;
204
205 lockdep_assert_held(&irq_lock);
206
207 list_for_each_entry(walk, &active_fds, list) {
208 if (walk->fd == fd)
209 return walk;
210 }
211
212 return NULL;
213}
214
215static void free_irq_entry(struct irq_entry *to_free, bool remove)
216{
217 if (!to_free)
218 return;
219
220 if (remove)
221 os_del_epoll_fd(to_free->fd);
222 list_del(entry: &to_free->list);
223 kfree(objp: to_free);
224}
225
226static bool update_irq_entry(struct irq_entry *entry)
227{
228 enum um_irq_type i;
229 int events = 0;
230
231 for (i = 0; i < NUM_IRQ_TYPES; i++)
232 events |= entry->reg[i].events;
233
234 if (events) {
235 /* will modify (instead of add) if needed */
236 os_add_epoll_fd(events, entry->fd, entry);
237 return true;
238 }
239
240 os_del_epoll_fd(entry->fd);
241 return false;
242}
243
244static void update_or_free_irq_entry(struct irq_entry *entry)
245{
246 if (!update_irq_entry(entry))
247 free_irq_entry(to_free: entry, remove: false);
248}
249
250static int activate_fd(int irq, int fd, enum um_irq_type type, void *dev_id,
251 void (*timetravel_handler)(int, int, void *,
252 struct time_travel_event *))
253{
254 struct irq_entry *irq_entry;
255 int err, events = os_event_mask(type);
256 unsigned long flags;
257
258 err = os_set_fd_async(fd);
259 if (err < 0)
260 goto out;
261
262 spin_lock_irqsave(&irq_lock, flags);
263 irq_entry = get_irq_entry_by_fd(fd);
264 if (irq_entry) {
265 /* cannot register the same FD twice with the same type */
266 if (WARN_ON(irq_entry->reg[type].events)) {
267 err = -EALREADY;
268 goto out_unlock;
269 }
270
271 /* temporarily disable to avoid IRQ-side locking */
272 os_del_epoll_fd(fd);
273 } else {
274 irq_entry = kzalloc(size: sizeof(*irq_entry), GFP_ATOMIC);
275 if (!irq_entry) {
276 err = -ENOMEM;
277 goto out_unlock;
278 }
279 irq_entry->fd = fd;
280 list_add_tail(new: &irq_entry->list, head: &active_fds);
281 maybe_sigio_broken(fd);
282 }
283
284 irq_entry->reg[type].id = dev_id;
285 irq_entry->reg[type].irq = irq;
286 irq_entry->reg[type].active = true;
287 irq_entry->reg[type].events = events;
288
289#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
290 if (um_irq_timetravel_handler_used()) {
291 irq_entry->reg[type].timetravel_handler = timetravel_handler;
292 irq_entry->reg[type].event.fn = irq_event_handler;
293 }
294#endif
295
296 WARN_ON(!update_irq_entry(irq_entry));
297 spin_unlock_irqrestore(lock: &irq_lock, flags);
298
299 return 0;
300out_unlock:
301 spin_unlock_irqrestore(lock: &irq_lock, flags);
302out:
303 return err;
304}
305
306/*
307 * Remove the entry or entries for a specific FD, if you
308 * don't want to remove all the possible entries then use
309 * um_free_irq() or deactivate_fd() instead.
310 */
311void free_irq_by_fd(int fd)
312{
313 struct irq_entry *to_free;
314 unsigned long flags;
315
316 spin_lock_irqsave(&irq_lock, flags);
317 to_free = get_irq_entry_by_fd(fd);
318 free_irq_entry(to_free, remove: true);
319 spin_unlock_irqrestore(lock: &irq_lock, flags);
320}
321EXPORT_SYMBOL(free_irq_by_fd);
322
323static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
324{
325 struct irq_entry *entry;
326 unsigned long flags;
327
328 spin_lock_irqsave(&irq_lock, flags);
329 list_for_each_entry(entry, &active_fds, list) {
330 enum um_irq_type i;
331
332 for (i = 0; i < NUM_IRQ_TYPES; i++) {
333 struct irq_reg *reg = &entry->reg[i];
334
335 if (!reg->events)
336 continue;
337 if (reg->irq != irq)
338 continue;
339 if (reg->id != dev)
340 continue;
341
342 os_del_epoll_fd(entry->fd);
343 reg->events = 0;
344 update_or_free_irq_entry(entry);
345 goto out;
346 }
347 }
348out:
349 spin_unlock_irqrestore(lock: &irq_lock, flags);
350}
351
352void deactivate_fd(int fd, int irqnum)
353{
354 struct irq_entry *entry;
355 unsigned long flags;
356 enum um_irq_type i;
357
358 os_del_epoll_fd(fd);
359
360 spin_lock_irqsave(&irq_lock, flags);
361 entry = get_irq_entry_by_fd(fd);
362 if (!entry)
363 goto out;
364
365 for (i = 0; i < NUM_IRQ_TYPES; i++) {
366 if (!entry->reg[i].events)
367 continue;
368 if (entry->reg[i].irq == irqnum)
369 entry->reg[i].events = 0;
370 }
371
372 update_or_free_irq_entry(entry);
373out:
374 spin_unlock_irqrestore(lock: &irq_lock, flags);
375
376 ignore_sigio_fd(fd);
377}
378EXPORT_SYMBOL(deactivate_fd);
379
380/*
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
384 * released the lock.
385 */
386int deactivate_all_fds(void)
387{
388 struct irq_entry *entry;
389
390 /* Stop IO. The IRQ loop has no lock so this is our
391 * only way of making sure we are safe to dispose
392 * of all IRQ handlers
393 */
394 os_set_ioignore();
395
396 /* we can no longer call kfree() here so just deactivate */
397 list_for_each_entry(entry, &active_fds, list)
398 os_del_epoll_fd(entry->fd);
399 os_close_epoll_fd();
400 return 0;
401}
402
403/*
404 * do_IRQ handles all normal device IRQs (the special
405 * SMP cross-CPU interrupts have their own specific
406 * handlers).
407 */
408unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
409{
410 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
411 irq_enter();
412 generic_handle_irq(irq);
413 irq_exit();
414 set_irq_regs(old_regs);
415 return 1;
416}
417
418void um_free_irq(int irq, void *dev)
419{
420 if (WARN(irq < 0 || irq > UM_LAST_SIGNAL_IRQ,
421 "freeing invalid irq %d", irq))
422 return;
423
424 free_irq_by_irq_and_dev(irq, dev);
425 free_irq(irq, dev);
426 clear_bit(nr: irq, addr: irqs_allocated);
427}
428EXPORT_SYMBOL(um_free_irq);
429
430static int
431_um_request_irq(int irq, int fd, enum um_irq_type type,
432 irq_handler_t handler, unsigned long irqflags,
433 const char *devname, void *dev_id,
434 void (*timetravel_handler)(int, int, void *,
435 struct time_travel_event *))
436{
437 int err;
438
439 if (irq == UM_IRQ_ALLOC) {
440 int i;
441
442 for (i = UM_FIRST_DYN_IRQ; i < NR_IRQS; i++) {
443 if (!test_and_set_bit(nr: i, addr: irqs_allocated)) {
444 irq = i;
445 break;
446 }
447 }
448 }
449
450 if (irq < 0)
451 return -ENOSPC;
452
453 if (fd != -1) {
454 err = activate_fd(irq, fd, type: type, dev_id, timetravel_handler);
455 if (err)
456 goto error;
457 }
458
459 err = request_irq(irq, handler, flags: irqflags, name: devname, dev: dev_id);
460 if (err < 0)
461 goto error;
462
463 return irq;
464error:
465 clear_bit(nr: irq, addr: irqs_allocated);
466 return err;
467}
468
469int um_request_irq(int irq, int fd, enum um_irq_type type,
470 irq_handler_t handler, unsigned long irqflags,
471 const char *devname, void *dev_id)
472{
473 return _um_request_irq(irq, fd, type: type, handler, irqflags,
474 devname, dev_id, NULL);
475}
476EXPORT_SYMBOL(um_request_irq);
477
478#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
479int um_request_irq_tt(int irq, int fd, enum um_irq_type type,
480 irq_handler_t handler, unsigned long irqflags,
481 const char *devname, void *dev_id,
482 void (*timetravel_handler)(int, int, void *,
483 struct time_travel_event *))
484{
485 return _um_request_irq(irq, fd, type, handler, irqflags,
486 devname, dev_id, timetravel_handler);
487}
488EXPORT_SYMBOL(um_request_irq_tt);
489
490void sigio_run_timetravel_handlers(void)
491{
492 _sigio_handler(NULL, true);
493}
494#endif
495
496#ifdef CONFIG_PM_SLEEP
497void um_irqs_suspend(void)
498{
499 struct irq_entry *entry;
500 unsigned long flags;
501
502 irqs_suspended = true;
503
504 spin_lock_irqsave(&irq_lock, flags);
505 list_for_each_entry(entry, &active_fds, list) {
506 enum um_irq_type t;
507 bool clear = true;
508
509 for (t = 0; t < NUM_IRQ_TYPES; t++) {
510 if (!entry->reg[t].events)
511 continue;
512
513 /*
514 * For the SIGIO_WRITE_IRQ, which is used to handle the
515 * SIGIO workaround thread, we need special handling:
516 * enable wake for it itself, but below we tell it about
517 * any FDs that should be suspended.
518 */
519 if (entry->reg[t].wakeup ||
520 entry->reg[t].irq == SIGIO_WRITE_IRQ
521#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
522 || entry->reg[t].timetravel_handler
523#endif
524 ) {
525 clear = false;
526 break;
527 }
528 }
529
530 if (clear) {
531 entry->suspended = true;
532 os_clear_fd_async(entry->fd);
533 entry->sigio_workaround =
534 !__ignore_sigio_fd(entry->fd);
535 }
536 }
537 spin_unlock_irqrestore(lock: &irq_lock, flags);
538}
539
540void um_irqs_resume(void)
541{
542 struct irq_entry *entry;
543 unsigned long flags;
544
545
546 local_irq_save(flags);
547#ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
548 /*
549 * We don't need to lock anything here since we're in resume
550 * and nothing else is running, but have disabled IRQs so we
551 * don't try anything else with the interrupt list from there.
552 */
553 list_for_each_entry(entry, &active_fds, list) {
554 enum um_irq_type t;
555
556 for (t = 0; t < NUM_IRQ_TYPES; t++) {
557 struct irq_reg *reg = &entry->reg[t];
558
559 if (reg->pending_on_resume) {
560 irq_enter();
561 generic_handle_irq(reg->irq);
562 irq_exit();
563 reg->pending_on_resume = false;
564 }
565 }
566 }
567#endif
568
569 spin_lock(lock: &irq_lock);
570 list_for_each_entry(entry, &active_fds, list) {
571 if (entry->suspended) {
572 int err = os_set_fd_async(entry->fd);
573
574 WARN(err < 0, "os_set_fd_async returned %d\n", err);
575 entry->suspended = false;
576
577 if (entry->sigio_workaround) {
578 err = __add_sigio_fd(entry->fd);
579 WARN(err < 0, "add_sigio_returned %d\n", err);
580 }
581 }
582 }
583 spin_unlock_irqrestore(lock: &irq_lock, flags);
584
585 irqs_suspended = false;
586 send_sigio_to_self();
587}
588
589static int normal_irq_set_wake(struct irq_data *d, unsigned int on)
590{
591 struct irq_entry *entry;
592 unsigned long flags;
593
594 spin_lock_irqsave(&irq_lock, flags);
595 list_for_each_entry(entry, &active_fds, list) {
596 enum um_irq_type t;
597
598 for (t = 0; t < NUM_IRQ_TYPES; t++) {
599 if (!entry->reg[t].events)
600 continue;
601
602 if (entry->reg[t].irq != d->irq)
603 continue;
604 entry->reg[t].wakeup = on;
605 goto unlock;
606 }
607 }
608unlock:
609 spin_unlock_irqrestore(lock: &irq_lock, flags);
610 return 0;
611}
612#else
613#define normal_irq_set_wake NULL
614#endif
615
616/*
617 * irq_chip must define at least enable/disable and ack when
618 * the edge handler is used.
619 */
620static void dummy(struct irq_data *d)
621{
622}
623
624/* This is used for everything other than the timer. */
625static struct irq_chip normal_irq_type = {
626 .name = "SIGIO",
627 .irq_disable = dummy,
628 .irq_enable = dummy,
629 .irq_ack = dummy,
630 .irq_mask = dummy,
631 .irq_unmask = dummy,
632 .irq_set_wake = normal_irq_set_wake,
633};
634
635static struct irq_chip alarm_irq_type = {
636 .name = "SIGALRM",
637 .irq_disable = dummy,
638 .irq_enable = dummy,
639 .irq_ack = dummy,
640 .irq_mask = dummy,
641 .irq_unmask = dummy,
642};
643
644void __init init_IRQ(void)
645{
646 int i;
647
648 irq_set_chip_and_handler(TIMER_IRQ, &alarm_irq_type, handle_edge_irq);
649
650 for (i = 1; i < UM_LAST_SIGNAL_IRQ; i++)
651 irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
652 /* Initialize EPOLL Loop */
653 os_setup_epoll();
654}
655
656/*
657 * IRQ stack entry and exit:
658 *
659 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack
660 * and switch over to the IRQ stack after some preparation. We use
661 * sigaltstack to receive signals on a separate stack from the start.
662 * These two functions make sure the rest of the kernel won't be too
663 * upset by being on a different stack. The IRQ stack has a
664 * thread_info structure at the bottom so that current et al continue
665 * to work.
666 *
667 * to_irq_stack copies the current task's thread_info to the IRQ stack
668 * thread_info and sets the tasks's stack to point to the IRQ stack.
669 *
670 * from_irq_stack copies the thread_info struct back (flags may have
671 * been modified) and resets the task's stack pointer.
672 *
673 * Tricky bits -
674 *
675 * What happens when two signals race each other? UML doesn't block
676 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal
677 * could arrive while a previous one is still setting up the
678 * thread_info.
679 *
680 * There are three cases -
681 * The first interrupt on the stack - sets up the thread_info and
682 * handles the interrupt
683 * A nested interrupt interrupting the copying of the thread_info -
684 * can't handle the interrupt, as the stack is in an unknown state
685 * A nested interrupt not interrupting the copying of the
686 * thread_info - doesn't do any setup, just handles the interrupt
687 *
688 * The first job is to figure out whether we interrupted stack setup.
689 * This is done by xchging the signal mask with thread_info->pending.
690 * If the value that comes back is zero, then there is no setup in
691 * progress, and the interrupt can be handled. If the value is
692 * non-zero, then there is stack setup in progress. In order to have
693 * the interrupt handled, we leave our signal in the mask, and it will
694 * be handled by the upper handler after it has set up the stack.
695 *
696 * Next is to figure out whether we are the outer handler or a nested
697 * one. As part of setting up the stack, thread_info->real_thread is
698 * set to non-NULL (and is reset to NULL on exit). This is the
699 * nesting indicator. If it is non-NULL, then the stack is already
700 * set up and the handler can run.
701 */
702
703static unsigned long pending_mask;
704
705unsigned long to_irq_stack(unsigned long *mask_out)
706{
707 struct thread_info *ti;
708 unsigned long mask, old;
709 int nested;
710
711 mask = xchg(&pending_mask, *mask_out);
712 if (mask != 0) {
713 /*
714 * If any interrupts come in at this point, we want to
715 * make sure that their bits aren't lost by our
716 * putting our bit in. So, this loop accumulates bits
717 * until xchg returns the same value that we put in.
718 * When that happens, there were no new interrupts,
719 * and pending_mask contains a bit for each interrupt
720 * that came in.
721 */
722 old = *mask_out;
723 do {
724 old |= mask;
725 mask = xchg(&pending_mask, old);
726 } while (mask != old);
727 return 1;
728 }
729
730 ti = current_thread_info();
731 nested = (ti->real_thread != NULL);
732 if (!nested) {
733 struct task_struct *task;
734 struct thread_info *tti;
735
736 task = cpu_tasks[ti->cpu].task;
737 tti = task_thread_info(task);
738
739 *ti = *tti;
740 ti->real_thread = tti;
741 task->stack = ti;
742 }
743
744 mask = xchg(&pending_mask, 0);
745 *mask_out |= mask | nested;
746 return 0;
747}
748
749unsigned long from_irq_stack(int nested)
750{
751 struct thread_info *ti, *to;
752 unsigned long mask;
753
754 ti = current_thread_info();
755
756 pending_mask = 1;
757
758 to = ti->real_thread;
759 current->stack = to;
760 ti->real_thread = NULL;
761 *to = *ti;
762
763 mask = xchg(&pending_mask, 0);
764 return mask & ~1;
765}
766
767

source code of linux/arch/um/kernel/irq.c