/arch/um/kernel/irq.c

  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
 23
 24extern void free_irqs(void);
 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_fd array for each fd -
 29 * one entry per IRQ type
 30 */
 31
 32struct irq_entry {
 33	struct irq_entry *next;
 34	int fd;
 35	struct irq_fd *irq_array[MAX_IRQ_TYPE + 1];
 36};
 37
 38static struct irq_entry *active_fds;
 39
 40static DEFINE_SPINLOCK(irq_lock);
 41
 42static void irq_io_loop(struct irq_fd *irq, struct uml_pt_regs *regs)
 43{
 44/*
 45 * irq->active guards against reentry
 46 * irq->pending accumulates pending requests
 47 * if pending is raised the irq_handler is re-run
 48 * until pending is cleared
 49 */
 50	if (irq->active) {
 51		irq->active = false;
 52		do {
 53			irq->pending = false;
 54			do_IRQ(irq->irq, regs);
 55		} while (irq->pending && (!irq->purge));
 56		if (!irq->purge)
 57			irq->active = true;
 58	} else {
 59		irq->pending = true;
 60	}
 61}
 62
 63void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
 64{
 65	struct irq_entry *irq_entry;
 66	struct irq_fd *irq;
 67
 68	int n, i, j;
 69
 70	while (1) {
 71		/* This is now lockless - epoll keeps back-referencesto the irqs
 72		 * which have trigger it so there is no need to walk the irq
 73		 * list and lock it every time. We avoid locking by turning off
 74		 * IO for a specific fd by executing os_del_epoll_fd(fd) before
 75		 * we do any changes to the actual data structures
 76		 */
 77		n = os_waiting_for_events_epoll();
 78
 79		if (n <= 0) {
 80			if (n == -EINTR)
 81				continue;
 82			else
 83				break;
 84		}
 85
 86		for (i = 0; i < n ; i++) {
 87			/* Epoll back reference is the entry with 3 irq_fd
 88			 * leaves - one for each irq type.
 89			 */
 90			irq_entry = (struct irq_entry *)
 91				os_epoll_get_data_pointer(i);
 92			for (j = 0; j < MAX_IRQ_TYPE ; j++) {
 93				irq = irq_entry->irq_array[j];
 94				if (irq == NULL)
 95					continue;
 96				if (os_epoll_triggered(i, irq->events) > 0)
 97					irq_io_loop(irq, regs);
 98				if (irq->purge) {
 99					irq_entry->irq_array[j] = NULL;
100					kfree(irq);
101				}
102			}
103		}
104	}
105
106	free_irqs();
107}
108
109static int assign_epoll_events_to_irq(struct irq_entry *irq_entry)
110{
111	int i;
112	int events = 0;
113	struct irq_fd *irq;
114
115	for (i = 0; i < MAX_IRQ_TYPE ; i++) {
116		irq = irq_entry->irq_array[i];
117		if (irq != NULL)
118			events = irq->events | events;
119	}
120	if (events > 0) {
121	/* os_add_epoll will call os_mod_epoll if this already exists */
122		return os_add_epoll_fd(events, irq_entry->fd, irq_entry);
123	}
124	/* No events - delete */
125	return os_del_epoll_fd(irq_entry->fd);
126}
127
128
129
130static int activate_fd(int irq, int fd, int type, void *dev_id)
131{
132	struct irq_fd *new_fd;
133	struct irq_entry *irq_entry;
134	int i, err, events;
135	unsigned long flags;
136
137	err = os_set_fd_async(fd);
138	if (err < 0)
139		goto out;
140
141	spin_lock_irqsave(&irq_lock, flags);
142
143	/* Check if we have an entry for this fd */
144
145	err = -EBUSY;
146	for (irq_entry = active_fds;
147		irq_entry != NULL; irq_entry = irq_entry->next) {
148		if (irq_entry->fd == fd)
149			break;
150	}
151
152	if (irq_entry == NULL) {
153		/* This needs to be atomic as it may be called from an
154		 * IRQ context.
155		 */
156		irq_entry = kmalloc(sizeof(struct irq_entry), GFP_ATOMIC);
157		if (irq_entry == NULL) {
158			printk(KERN_ERR
159				"Failed to allocate new IRQ entry\n");
160			goto out_unlock;
161		}
162		irq_entry->fd = fd;
163		for (i = 0; i < MAX_IRQ_TYPE; i++)
164			irq_entry->irq_array[i] = NULL;
165		irq_entry->next = active_fds;
166		active_fds = irq_entry;
167	}
168
169	/* Check if we are trying to re-register an interrupt for a
170	 * particular fd
171	 */
172
173	if (irq_entry->irq_array[type] != NULL) {
174		printk(KERN_ERR
175			"Trying to reregister IRQ %d FD %d TYPE %d ID %p\n",
176			irq, fd, type, dev_id
177		);
178		goto out_unlock;
179	} else {
180		/* New entry for this fd */
181
182		err = -ENOMEM;
183		new_fd = kmalloc(sizeof(struct irq_fd), GFP_ATOMIC);
184		if (new_fd == NULL)
185			goto out_unlock;
186
187		events = os_event_mask(type);
188
189		*new_fd = ((struct irq_fd) {
190			.id		= dev_id,
191			.irq		= irq,
192			.type		= type,
193			.events		= events,
194			.active		= true,
195			.pending	= false,
196			.purge		= false
197		});
198		/* Turn off any IO on this fd - allows us to
199		 * avoid locking the IRQ loop
200		 */
201		os_del_epoll_fd(irq_entry->fd);
202		irq_entry->irq_array[type] = new_fd;
203	}
204
205	/* Turn back IO on with the correct (new) IO event mask */
206	assign_epoll_events_to_irq(irq_entry);
207	spin_unlock_irqrestore(&irq_lock, flags);
208	maybe_sigio_broken(fd, (type != IRQ_NONE));
209
210	return 0;
211out_unlock:
212	spin_unlock_irqrestore(&irq_lock, flags);
213out:
214	return err;
215}
216
217/*
218 * Walk the IRQ list and dispose of any unused entries.
219 * Should be done under irq_lock.
220 */
221
222static void garbage_collect_irq_entries(void)
223{
224	int i;
225	bool reap;
226	struct irq_entry *walk;
227	struct irq_entry *previous = NULL;
228	struct irq_entry *to_free;
229
230	if (active_fds == NULL)
231		return;
232	walk = active_fds;
233	while (walk != NULL) {
234		reap = true;
235		for (i = 0; i < MAX_IRQ_TYPE ; i++) {
236			if (walk->irq_array[i] != NULL) {
237				reap = false;
238				break;
239			}
240		}
241		if (reap) {
242			if (previous == NULL)
243				active_fds = walk->next;
244			else
245				previous->next = walk->next;
246			to_free = walk;
247		} else {
248			to_free = NULL;
249		}
250		walk = walk->next;
251		kfree(to_free);
252	}
253}
254
255/*
256 * Walk the IRQ list and get the descriptor for our FD
257 */
258
259static struct irq_entry *get_irq_entry_by_fd(int fd)
260{
261	struct irq_entry *walk = active_fds;
262
263	while (walk != NULL) {
264		if (walk->fd == fd)
265			return walk;
266		walk = walk->next;
267	}
268	return NULL;
269}
270
271
272/*
273 * Walk the IRQ list and dispose of an entry for a specific
274 * device, fd and number. Note - if sharing an IRQ for read
275 * and writefor the same FD it will be disposed in either case.
276 * If this behaviour is undesirable use different IRQ ids.
277 */
278
279#define IGNORE_IRQ 1
280#define IGNORE_DEV (1<<1)
281
282static void do_free_by_irq_and_dev(
283	struct irq_entry *irq_entry,
284	unsigned int irq,
285	void *dev,
286	int flags
287)
288{
289	int i;
290	struct irq_fd *to_free;
291
292	for (i = 0; i < MAX_IRQ_TYPE ; i++) {
293		if (irq_entry->irq_array[i] != NULL) {
294			if (
295			((flags & IGNORE_IRQ) ||
296				(irq_entry->irq_array[i]->irq == irq)) &&
297			((flags & IGNORE_DEV) ||
298				(irq_entry->irq_array[i]->id == dev))
299			) {
300				/* Turn off any IO on this fd - allows us to
301				 * avoid locking the IRQ loop
302				 */
303				os_del_epoll_fd(irq_entry->fd);
304				to_free = irq_entry->irq_array[i];
305				irq_entry->irq_array[i] = NULL;
306				assign_epoll_events_to_irq(irq_entry);
307				if (to_free->active)
308					to_free->purge = true;
309				else
310					kfree(to_free);
311			}
312		}
313	}
314}
315
316void free_irq_by_fd(int fd)
317{
318	struct irq_entry *to_free;
319	unsigned long flags;
320
321	spin_lock_irqsave(&irq_lock, flags);
322	to_free = get_irq_entry_by_fd(fd);
323	if (to_free != NULL) {
324		do_free_by_irq_and_dev(
325			to_free,
326			-1,
327			NULL,
328			IGNORE_IRQ | IGNORE_DEV
329		);
330	}
331	garbage_collect_irq_entries();
332	spin_unlock_irqrestore(&irq_lock, flags);
333}
334EXPORT_SYMBOL(free_irq_by_fd);
335
336static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
337{
338	struct irq_entry *to_free;
339	unsigned long flags;
340
341	spin_lock_irqsave(&irq_lock, flags);
342	to_free = active_fds;
343	while (to_free != NULL) {
344		do_free_by_irq_and_dev(
345			to_free,
346			irq,
347			dev,
348			0
349		);
350		to_free = to_free->next;
351	}
352	garbage_collect_irq_entries();
353	spin_unlock_irqrestore(&irq_lock, flags);
354}
355
356
357void deactivate_fd(int fd, int irqnum)
358{
359	struct irq_entry *to_free;
360	unsigned long flags;
361
362	os_del_epoll_fd(fd);
363	spin_lock_irqsave(&irq_lock, flags);
364	to_free = get_irq_entry_by_fd(fd);
365	if (to_free != NULL) {
366		do_free_by_irq_and_dev(
367			to_free,
368			irqnum,
369			NULL,
370			IGNORE_DEV
371		);
372	}
373	garbage_collect_irq_entries();
374	spin_unlock_irqrestore(&irq_lock, flags);
375	ignore_sigio_fd(fd);
376}
377EXPORT_SYMBOL(deactivate_fd);
378
379/*
380 * Called just before shutdown in order to provide a clean exec
381 * environment in case the system is rebooting.  No locking because
382 * that would cause a pointless shutdown hang if something hadn't
383 * released the lock.
384 */
385int deactivate_all_fds(void)
386{
387	struct irq_entry *to_free;
388
389	/* Stop IO. The IRQ loop has no lock so this is our
390	 * only way of making sure we are safe to dispose
391	 * of all IRQ handlers
392	 */
393	os_set_ioignore();
394	to_free = active_fds;
395	while (to_free != NULL) {
396		do_free_by_irq_and_dev(
397			to_free,
398			-1,
399			NULL,
400			IGNORE_IRQ | IGNORE_DEV
401		);
402		to_free = to_free->next;
403	}
404	/* don't garbage collect - we can no longer call kfree() here */
405	os_close_epoll_fd();
406	return 0;
407}
408
409/*
410 * do_IRQ handles all normal device IRQs (the special
411 * SMP cross-CPU interrupts have their own specific
412 * handlers).
413 */
414unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
415{
416	struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
417	irq_enter();
418	generic_handle_irq(irq);
419	irq_exit();
420	set_irq_regs(old_regs);
421	return 1;
422}
423
424void um_free_irq(unsigned int irq, void *dev)
425{
426	free_irq_by_irq_and_dev(irq, dev);
427	free_irq(irq, dev);
428}
429EXPORT_SYMBOL(um_free_irq);
430
431int um_request_irq(unsigned int irq, int fd, int type,
432		   irq_handler_t handler,
433		   unsigned long irqflags, const char * devname,
434		   void *dev_id)
435{
436	int err;
437
438	if (fd != -1) {
439		err = activate_fd(irq, fd, type, dev_id);
440		if (err)
441			return err;
442	}
443
444	return request_irq(irq, handler, irqflags, devname, dev_id);
445}
446
447EXPORT_SYMBOL(um_request_irq);
448
449/*
450 * irq_chip must define at least enable/disable and ack when
451 * the edge handler is used.
452 */
453static void dummy(struct irq_data *d)
454{
455}
456
457/* This is used for everything else than the timer. */
458static struct irq_chip normal_irq_type = {
459	.name = "SIGIO",
460	.irq_disable = dummy,
461	.irq_enable = dummy,
462	.irq_ack = dummy,
463	.irq_mask = dummy,
464	.irq_unmask = dummy,
465};
466
467static struct irq_chip SIGVTALRM_irq_type = {
468	.name = "SIGVTALRM",
469	.irq_disable = dummy,
470	.irq_enable = dummy,
471	.irq_ack = dummy,
472	.irq_mask = dummy,
473	.irq_unmask = dummy,
474};
475
476void __init init_IRQ(void)
477{
478	int i;
479
480	irq_set_chip_and_handler(TIMER_IRQ, &SIGVTALRM_irq_type, handle_edge_irq);
481
482
483	for (i = 1; i <= LAST_IRQ; i++)
484		irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
485	/* Initialize EPOLL Loop */
486	os_setup_epoll();
487}
488
489/*
490 * IRQ stack entry and exit:
491 *
492 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack
493 * and switch over to the IRQ stack after some preparation.  We use
494 * sigaltstack to receive signals on a separate stack from the start.
495 * These two functions make sure the rest of the kernel won't be too
496 * upset by being on a different stack.  The IRQ stack has a
497 * thread_info structure at the bottom so that current et al continue
498 * to work.
499 *
500 * to_irq_stack copies the current task's thread_info to the IRQ stack
501 * thread_info and sets the tasks's stack to point to the IRQ stack.
502 *
503 * from_irq_stack copies the thread_info struct back (flags may have
504 * been modified) and resets the task's stack pointer.
505 *
506 * Tricky bits -
507 *
508 * What happens when two signals race each other?  UML doesn't block
509 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal
510 * could arrive while a previous one is still setting up the
511 * thread_info.
512 *
513 * There are three cases -
514 *     The first interrupt on the stack - sets up the thread_info and
515 * handles the interrupt
516 *     A nested interrupt interrupting the copying of the thread_info -
517 * can't handle the interrupt, as the stack is in an unknown state
518 *     A nested interrupt not interrupting the copying of the
519 * thread_info - doesn't do any setup, just handles the interrupt
520 *
521 * The first job is to figure out whether we interrupted stack setup.
522 * This is done by xchging the signal mask with thread_info->pending.
523 * If the value that comes back is zero, then there is no setup in
524 * progress, and the interrupt can be handled.  If the value is
525 * non-zero, then there is stack setup in progress.  In order to have
526 * the interrupt handled, we leave our signal in the mask, and it will
527 * be handled by the upper handler after it has set up the stack.
528 *
529 * Next is to figure out whether we are the outer handler or a nested
530 * one.  As part of setting up the stack, thread_info->real_thread is
531 * set to non-NULL (and is reset to NULL on exit).  This is the
532 * nesting indicator.  If it is non-NULL, then the stack is already
533 * set up and the handler can run.
534 */
535
536static unsigned long pending_mask;
537
538unsigned long to_irq_stack(unsigned long *mask_out)
539{
540	struct thread_info *ti;
541	unsigned long mask, old;
542	int nested;
543
544	mask = xchg(&pending_mask, *mask_out);
545	if (mask != 0) {
546		/*
547		 * If any interrupts come in at this point, we want to
548		 * make sure that their bits aren't lost by our
549		 * putting our bit in.  So, this loop accumulates bits
550		 * until xchg returns the same value that we put in.
551		 * When that happens, there were no new interrupts,
552		 * and pending_mask contains a bit for each interrupt
553		 * that came in.
554		 */
555		old = *mask_out;
556		do {
557			old |= mask;
558			mask = xchg(&pending_mask, old);
559		} while (mask != old);
560		return 1;
561	}
562
563	ti = current_thread_info();
564	nested = (ti->real_thread != NULL);
565	if (!nested) {
566		struct task_struct *task;
567		struct thread_info *tti;
568
569		task = cpu_tasks[ti->cpu].task;
570		tti = task_thread_info(task);
571
572		*ti = *tti;
573		ti->real_thread = tti;
574		task->stack = ti;
575	}
576
577	mask = xchg(&pending_mask, 0);
578	*mask_out |= mask | nested;
579	return 0;
580}
581
582unsigned long from_irq_stack(int nested)
583{
584	struct thread_info *ti, *to;
585	unsigned long mask;
586
587	ti = current_thread_info();
588
589	pending_mask = 1;
590
591	to = ti->real_thread;
592	current->stack = to;
593	ti->real_thread = NULL;
594	*to = *ti;
595
596	mask = xchg(&pending_mask, 0);
597	return mask & ~1;
598}
599