Loading...
1/*
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
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
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
28 */
29
30struct irq_entry {
31 struct irq_entry *next;
32 int fd;
33 struct irq_fd *irq_array[MAX_IRQ_TYPE + 1];
34};
35
36static struct irq_entry *active_fds;
37
38static DEFINE_SPINLOCK(irq_lock);
39
40static void irq_io_loop(struct irq_fd *irq, struct uml_pt_regs *regs)
41{
42/*
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
47 */
48 if (irq->active) {
49 irq->active = false;
50 do {
51 irq->pending = false;
52 do_IRQ(irq->irq, regs);
53 } while (irq->pending && (!irq->purge));
54 if (!irq->purge)
55 irq->active = true;
56 } else {
57 irq->pending = true;
58 }
59}
60
61void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
62{
63 struct irq_entry *irq_entry;
64 struct irq_fd *irq;
65
66 int n, i, j;
67
68 while (1) {
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
74 */
75 n = os_waiting_for_events_epoll();
76
77 if (n <= 0) {
78 if (n == -EINTR)
79 continue;
80 else
81 break;
82 }
83
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.
87 */
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];
92 if (irq == NULL)
93 continue;
94 if (os_epoll_triggered(i, irq->events) > 0)
95 irq_io_loop(irq, regs);
96 if (irq->purge) {
97 irq_entry->irq_array[j] = NULL;
98 kfree(irq);
99 }
100 }
101 }
102 }
103}
104
105static int assign_epoll_events_to_irq(struct irq_entry *irq_entry)
106{
107 int i;
108 int events = 0;
109 struct irq_fd *irq;
110
111 for (i = 0; i < MAX_IRQ_TYPE ; i++) {
112 irq = irq_entry->irq_array[i];
113 if (irq != NULL)
114 events = irq->events | events;
115 }
116 if (events > 0) {
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);
119 }
120 /* No events - delete */
121 return os_del_epoll_fd(irq_entry->fd);
122}
123
124
125
126static int activate_fd(int irq, int fd, int type, void *dev_id)
127{
128 struct irq_fd *new_fd;
129 struct irq_entry *irq_entry;
130 int i, err, events;
131 unsigned long flags;
132
133 err = os_set_fd_async(fd);
134 if (err < 0)
135 goto out;
136
137 spin_lock_irqsave(&irq_lock, flags);
138
139 /* Check if we have an entry for this fd */
140
141 err = -EBUSY;
142 for (irq_entry = active_fds;
143 irq_entry != NULL; irq_entry = irq_entry->next) {
144 if (irq_entry->fd == fd)
145 break;
146 }
147
148 if (irq_entry == NULL) {
149 /* This needs to be atomic as it may be called from an
150 * IRQ context.
151 */
152 irq_entry = kmalloc(sizeof(struct irq_entry), GFP_ATOMIC);
153 if (irq_entry == NULL) {
154 printk(KERN_ERR
155 "Failed to allocate new IRQ entry\n");
156 goto out_unlock;
157 }
158 irq_entry->fd = fd;
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;
163 }
164
165 /* Check if we are trying to re-register an interrupt for a
166 * particular fd
167 */
168
169 if (irq_entry->irq_array[type] != NULL) {
170 printk(KERN_ERR
171 "Trying to reregister IRQ %d FD %d TYPE %d ID %p\n",
172 irq, fd, type, dev_id
173 );
174 goto out_unlock;
175 } else {
176 /* New entry for this fd */
177
178 err = -ENOMEM;
179 new_fd = kmalloc(sizeof(struct irq_fd), GFP_ATOMIC);
180 if (new_fd == NULL)
181 goto out_unlock;
182
183 events = os_event_mask(type);
184
185 *new_fd = ((struct irq_fd) {
186 .id = dev_id,
187 .irq = irq,
188 .type = type,
189 .events = events,
190 .active = true,
191 .pending = false,
192 .purge = false
193 });
194 /* Turn off any IO on this fd - allows us to
195 * avoid locking the IRQ loop
196 */
197 os_del_epoll_fd(irq_entry->fd);
198 irq_entry->irq_array[type] = new_fd;
199 }
200
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));
205
206 return 0;
207out_unlock:
208 spin_unlock_irqrestore(&irq_lock, flags);
209out:
210 return err;
211}
212
213/*
214 * Walk the IRQ list and dispose of any unused entries.
215 * Should be done under irq_lock.
216 */
217
218static void garbage_collect_irq_entries(void)
219{
220 int i;
221 bool reap;
222 struct irq_entry *walk;
223 struct irq_entry *previous = NULL;
224 struct irq_entry *to_free;
225
226 if (active_fds == NULL)
227 return;
228 walk = active_fds;
229 while (walk != NULL) {
230 reap = true;
231 for (i = 0; i < MAX_IRQ_TYPE ; i++) {
232 if (walk->irq_array[i] != NULL) {
233 reap = false;
234 break;
235 }
236 }
237 if (reap) {
238 if (previous == NULL)
239 active_fds = walk->next;
240 else
241 previous->next = walk->next;
242 to_free = walk;
243 } else {
244 to_free = NULL;
245 }
246 walk = walk->next;
247 if (to_free != NULL)
248 kfree(to_free);
249 }
250}
251
252/*
253 * Walk the IRQ list and get the descriptor for our FD
254 */
255
256static struct irq_entry *get_irq_entry_by_fd(int fd)
257{
258 struct irq_entry *walk = active_fds;
259
260 while (walk != NULL) {
261 if (walk->fd == fd)
262 return walk;
263 walk = walk->next;
264 }
265 return NULL;
266}
267
268
269/*
270 * Walk the IRQ list and dispose of an entry for a specific
271 * device, fd and number. Note - if sharing an IRQ for read
272 * and writefor the same FD it will be disposed in either case.
273 * If this behaviour is undesirable use different IRQ ids.
274 */
275
276#define IGNORE_IRQ 1
277#define IGNORE_DEV (1<<1)
278
279static void do_free_by_irq_and_dev(
280 struct irq_entry *irq_entry,
281 unsigned int irq,
282 void *dev,
283 int flags
284)
285{
286 int i;
287 struct irq_fd *to_free;
288
289 for (i = 0; i < MAX_IRQ_TYPE ; i++) {
290 if (irq_entry->irq_array[i] != NULL) {
291 if (
292 ((flags & IGNORE_IRQ) ||
293 (irq_entry->irq_array[i]->irq == irq)) &&
294 ((flags & IGNORE_DEV) ||
295 (irq_entry->irq_array[i]->id == dev))
296 ) {
297 /* Turn off any IO on this fd - allows us to
298 * avoid locking the IRQ loop
299 */
300 os_del_epoll_fd(irq_entry->fd);
301 to_free = irq_entry->irq_array[i];
302 irq_entry->irq_array[i] = NULL;
303 assign_epoll_events_to_irq(irq_entry);
304 if (to_free->active)
305 to_free->purge = true;
306 else
307 kfree(to_free);
308 }
309 }
310 }
311}
312
313void free_irq_by_fd(int fd)
314{
315 struct irq_entry *to_free;
316 unsigned long flags;
317
318 spin_lock_irqsave(&irq_lock, flags);
319 to_free = get_irq_entry_by_fd(fd);
320 if (to_free != NULL) {
321 do_free_by_irq_and_dev(
322 to_free,
323 -1,
324 NULL,
325 IGNORE_IRQ | IGNORE_DEV
326 );
327 }
328 garbage_collect_irq_entries();
329 spin_unlock_irqrestore(&irq_lock, flags);
330}
331EXPORT_SYMBOL(free_irq_by_fd);
332
333static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
334{
335 struct irq_entry *to_free;
336 unsigned long flags;
337
338 spin_lock_irqsave(&irq_lock, flags);
339 to_free = active_fds;
340 while (to_free != NULL) {
341 do_free_by_irq_and_dev(
342 to_free,
343 irq,
344 dev,
345 0
346 );
347 to_free = to_free->next;
348 }
349 garbage_collect_irq_entries();
350 spin_unlock_irqrestore(&irq_lock, flags);
351}
352
353
354void reactivate_fd(int fd, int irqnum)
355{
356 /** NOP - we do auto-EOI now **/
357}
358
359void deactivate_fd(int fd, int irqnum)
360{
361 struct irq_entry *to_free;
362 unsigned long flags;
363
364 os_del_epoll_fd(fd);
365 spin_lock_irqsave(&irq_lock, flags);
366 to_free = get_irq_entry_by_fd(fd);
367 if (to_free != NULL) {
368 do_free_by_irq_and_dev(
369 to_free,
370 irqnum,
371 NULL,
372 IGNORE_DEV
373 );
374 }
375 garbage_collect_irq_entries();
376 spin_unlock_irqrestore(&irq_lock, flags);
377 ignore_sigio_fd(fd);
378}
379EXPORT_SYMBOL(deactivate_fd);
380
381/*
382 * Called just before shutdown in order to provide a clean exec
383 * environment in case the system is rebooting. No locking because
384 * that would cause a pointless shutdown hang if something hadn't
385 * released the lock.
386 */
387int deactivate_all_fds(void)
388{
389 unsigned long flags;
390 struct irq_entry *to_free;
391
392 spin_lock_irqsave(&irq_lock, flags);
393 /* Stop IO. The IRQ loop has no lock so this is our
394 * only way of making sure we are safe to dispose
395 * of all IRQ handlers
396 */
397 os_set_ioignore();
398 to_free = active_fds;
399 while (to_free != NULL) {
400 do_free_by_irq_and_dev(
401 to_free,
402 -1,
403 NULL,
404 IGNORE_IRQ | IGNORE_DEV
405 );
406 to_free = to_free->next;
407 }
408 garbage_collect_irq_entries();
409 spin_unlock_irqrestore(&irq_lock, flags);
410 os_close_epoll_fd();
411 return 0;
412}
413
414/*
415 * do_IRQ handles all normal device IRQs (the special
416 * SMP cross-CPU interrupts have their own specific
417 * handlers).
418 */
419unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
420{
421 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
422 irq_enter();
423 generic_handle_irq(irq);
424 irq_exit();
425 set_irq_regs(old_regs);
426 return 1;
427}
428
429void um_free_irq(unsigned int irq, void *dev)
430{
431 free_irq_by_irq_and_dev(irq, dev);
432 free_irq(irq, dev);
433}
434EXPORT_SYMBOL(um_free_irq);
435
436int um_request_irq(unsigned int irq, int fd, int type,
437 irq_handler_t handler,
438 unsigned long irqflags, const char * devname,
439 void *dev_id)
440{
441 int err;
442
443 if (fd != -1) {
444 err = activate_fd(irq, fd, type, dev_id);
445 if (err)
446 return err;
447 }
448
449 return request_irq(irq, handler, irqflags, devname, dev_id);
450}
451
452EXPORT_SYMBOL(um_request_irq);
453EXPORT_SYMBOL(reactivate_fd);
454
455/*
456 * irq_chip must define at least enable/disable and ack when
457 * the edge handler is used.
458 */
459static void dummy(struct irq_data *d)
460{
461}
462
463/* This is used for everything else than the timer. */
464static struct irq_chip normal_irq_type = {
465 .name = "SIGIO",
466 .irq_disable = dummy,
467 .irq_enable = dummy,
468 .irq_ack = dummy,
469 .irq_mask = dummy,
470 .irq_unmask = dummy,
471};
472
473static struct irq_chip SIGVTALRM_irq_type = {
474 .name = "SIGVTALRM",
475 .irq_disable = dummy,
476 .irq_enable = dummy,
477 .irq_ack = dummy,
478 .irq_mask = dummy,
479 .irq_unmask = dummy,
480};
481
482void __init init_IRQ(void)
483{
484 int i;
485
486 irq_set_chip_and_handler(TIMER_IRQ, &SIGVTALRM_irq_type, handle_edge_irq);
487
488
489 for (i = 1; i < NR_IRQS; i++)
490 irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
491 /* Initialize EPOLL Loop */
492 os_setup_epoll();
493}
494
495/*
496 * IRQ stack entry and exit:
497 *
498 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack
499 * and switch over to the IRQ stack after some preparation. We use
500 * sigaltstack to receive signals on a separate stack from the start.
501 * These two functions make sure the rest of the kernel won't be too
502 * upset by being on a different stack. The IRQ stack has a
503 * thread_info structure at the bottom so that current et al continue
504 * to work.
505 *
506 * to_irq_stack copies the current task's thread_info to the IRQ stack
507 * thread_info and sets the tasks's stack to point to the IRQ stack.
508 *
509 * from_irq_stack copies the thread_info struct back (flags may have
510 * been modified) and resets the task's stack pointer.
511 *
512 * Tricky bits -
513 *
514 * What happens when two signals race each other? UML doesn't block
515 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal
516 * could arrive while a previous one is still setting up the
517 * thread_info.
518 *
519 * There are three cases -
520 * The first interrupt on the stack - sets up the thread_info and
521 * handles the interrupt
522 * A nested interrupt interrupting the copying of the thread_info -
523 * can't handle the interrupt, as the stack is in an unknown state
524 * A nested interrupt not interrupting the copying of the
525 * thread_info - doesn't do any setup, just handles the interrupt
526 *
527 * The first job is to figure out whether we interrupted stack setup.
528 * This is done by xchging the signal mask with thread_info->pending.
529 * If the value that comes back is zero, then there is no setup in
530 * progress, and the interrupt can be handled. If the value is
531 * non-zero, then there is stack setup in progress. In order to have
532 * the interrupt handled, we leave our signal in the mask, and it will
533 * be handled by the upper handler after it has set up the stack.
534 *
535 * Next is to figure out whether we are the outer handler or a nested
536 * one. As part of setting up the stack, thread_info->real_thread is
537 * set to non-NULL (and is reset to NULL on exit). This is the
538 * nesting indicator. If it is non-NULL, then the stack is already
539 * set up and the handler can run.
540 */
541
542static unsigned long pending_mask;
543
544unsigned long to_irq_stack(unsigned long *mask_out)
545{
546 struct thread_info *ti;
547 unsigned long mask, old;
548 int nested;
549
550 mask = xchg(&pending_mask, *mask_out);
551 if (mask != 0) {
552 /*
553 * If any interrupts come in at this point, we want to
554 * make sure that their bits aren't lost by our
555 * putting our bit in. So, this loop accumulates bits
556 * until xchg returns the same value that we put in.
557 * When that happens, there were no new interrupts,
558 * and pending_mask contains a bit for each interrupt
559 * that came in.
560 */
561 old = *mask_out;
562 do {
563 old |= mask;
564 mask = xchg(&pending_mask, old);
565 } while (mask != old);
566 return 1;
567 }
568
569 ti = current_thread_info();
570 nested = (ti->real_thread != NULL);
571 if (!nested) {
572 struct task_struct *task;
573 struct thread_info *tti;
574
575 task = cpu_tasks[ti->cpu].task;
576 tti = task_thread_info(task);
577
578 *ti = *tti;
579 ti->real_thread = tti;
580 task->stack = ti;
581 }
582
583 mask = xchg(&pending_mask, 0);
584 *mask_out |= mask | nested;
585 return 0;
586}
587
588unsigned long from_irq_stack(int nested)
589{
590 struct thread_info *ti, *to;
591 unsigned long mask;
592
593 ti = current_thread_info();
594
595 pending_mask = 1;
596
597 to = ti->real_thread;
598 current->stack = to;
599 ti->real_thread = NULL;
600 *to = *ti;
601
602 mask = xchg(&pending_mask, 0);
603 return mask & ~1;
604}
605
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 uml_pt_regs *regs)
34{
35 struct irq_fd *irq_fd;
36 int n;
37
38 if (smp_sigio_handler())
39 return;
40
41 while (1) {
42 n = os_waiting_for_events(active_fds);
43 if (n <= 0) {
44 if (n == -EINTR)
45 continue;
46 else break;
47 }
48
49 for (irq_fd = active_fds; irq_fd != NULL;
50 irq_fd = irq_fd->next) {
51 if (irq_fd->current_events != 0) {
52 irq_fd->current_events = 0;
53 do_IRQ(irq_fd->irq, regs);
54 }
55 }
56 }
57
58 free_irqs();
59}
60
61static DEFINE_SPINLOCK(irq_lock);
62
63static int activate_fd(int irq, int fd, int type, void *dev_id)
64{
65 struct pollfd *tmp_pfd;
66 struct irq_fd *new_fd, *irq_fd;
67 unsigned long flags;
68 int events, err, n;
69
70 err = os_set_fd_async(fd);
71 if (err < 0)
72 goto out;
73
74 err = -ENOMEM;
75 new_fd = kmalloc(sizeof(struct irq_fd), GFP_KERNEL);
76 if (new_fd == NULL)
77 goto out;
78
79 if (type == IRQ_READ)
80 events = UM_POLLIN | UM_POLLPRI;
81 else events = UM_POLLOUT;
82 *new_fd = ((struct irq_fd) { .next = NULL,
83 .id = dev_id,
84 .fd = fd,
85 .type = type,
86 .irq = irq,
87 .events = events,
88 .current_events = 0 } );
89
90 err = -EBUSY;
91 spin_lock_irqsave(&irq_lock, flags);
92 for (irq_fd = active_fds; irq_fd != NULL; irq_fd = irq_fd->next) {
93 if ((irq_fd->fd == fd) && (irq_fd->type == type)) {
94 printk(KERN_ERR "Registering fd %d twice\n", fd);
95 printk(KERN_ERR "Irqs : %d, %d\n", irq_fd->irq, irq);
96 printk(KERN_ERR "Ids : 0x%p, 0x%p\n", irq_fd->id,
97 dev_id);
98 goto out_unlock;
99 }
100 }
101
102 if (type == IRQ_WRITE)
103 fd = -1;
104
105 tmp_pfd = NULL;
106 n = 0;
107
108 while (1) {
109 n = os_create_pollfd(fd, events, tmp_pfd, n);
110 if (n == 0)
111 break;
112
113 /*
114 * n > 0
115 * It means we couldn't put new pollfd to current pollfds
116 * and tmp_fds is NULL or too small for new pollfds array.
117 * Needed size is equal to n as minimum.
118 *
119 * Here we have to drop the lock in order to call
120 * kmalloc, which might sleep.
121 * If something else came in and changed the pollfds array
122 * so we will not be able to put new pollfd struct to pollfds
123 * then we free the buffer tmp_fds and try again.
124 */
125 spin_unlock_irqrestore(&irq_lock, flags);
126 kfree(tmp_pfd);
127
128 tmp_pfd = kmalloc(n, GFP_KERNEL);
129 if (tmp_pfd == NULL)
130 goto out_kfree;
131
132 spin_lock_irqsave(&irq_lock, flags);
133 }
134
135 *last_irq_ptr = new_fd;
136 last_irq_ptr = &new_fd->next;
137
138 spin_unlock_irqrestore(&irq_lock, flags);
139
140 /*
141 * This calls activate_fd, so it has to be outside the critical
142 * section.
143 */
144 maybe_sigio_broken(fd, (type == IRQ_READ));
145
146 return 0;
147
148 out_unlock:
149 spin_unlock_irqrestore(&irq_lock, flags);
150 out_kfree:
151 kfree(new_fd);
152 out:
153 return err;
154}
155
156static void free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg)
157{
158 unsigned long flags;
159
160 spin_lock_irqsave(&irq_lock, flags);
161 os_free_irq_by_cb(test, arg, active_fds, &last_irq_ptr);
162 spin_unlock_irqrestore(&irq_lock, flags);
163}
164
165struct irq_and_dev {
166 int irq;
167 void *dev;
168};
169
170static int same_irq_and_dev(struct irq_fd *irq, void *d)
171{
172 struct irq_and_dev *data = d;
173
174 return ((irq->irq == data->irq) && (irq->id == data->dev));
175}
176
177static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
178{
179 struct irq_and_dev data = ((struct irq_and_dev) { .irq = irq,
180 .dev = dev });
181
182 free_irq_by_cb(same_irq_and_dev, &data);
183}
184
185static int same_fd(struct irq_fd *irq, void *fd)
186{
187 return (irq->fd == *((int *)fd));
188}
189
190void free_irq_by_fd(int fd)
191{
192 free_irq_by_cb(same_fd, &fd);
193}
194
195/* Must be called with irq_lock held */
196static struct irq_fd *find_irq_by_fd(int fd, int irqnum, int *index_out)
197{
198 struct irq_fd *irq;
199 int i = 0;
200 int fdi;
201
202 for (irq = active_fds; irq != NULL; irq = irq->next) {
203 if ((irq->fd == fd) && (irq->irq == irqnum))
204 break;
205 i++;
206 }
207 if (irq == NULL) {
208 printk(KERN_ERR "find_irq_by_fd doesn't have descriptor %d\n",
209 fd);
210 goto out;
211 }
212 fdi = os_get_pollfd(i);
213 if ((fdi != -1) && (fdi != fd)) {
214 printk(KERN_ERR "find_irq_by_fd - mismatch between active_fds "
215 "and pollfds, fd %d vs %d, need %d\n", irq->fd,
216 fdi, fd);
217 irq = NULL;
218 goto out;
219 }
220 *index_out = i;
221 out:
222 return irq;
223}
224
225void reactivate_fd(int fd, int irqnum)
226{
227 struct irq_fd *irq;
228 unsigned long flags;
229 int i;
230
231 spin_lock_irqsave(&irq_lock, flags);
232 irq = find_irq_by_fd(fd, irqnum, &i);
233 if (irq == NULL) {
234 spin_unlock_irqrestore(&irq_lock, flags);
235 return;
236 }
237 os_set_pollfd(i, irq->fd);
238 spin_unlock_irqrestore(&irq_lock, flags);
239
240 add_sigio_fd(fd);
241}
242
243void deactivate_fd(int fd, int irqnum)
244{
245 struct irq_fd *irq;
246 unsigned long flags;
247 int i;
248
249 spin_lock_irqsave(&irq_lock, flags);
250 irq = find_irq_by_fd(fd, irqnum, &i);
251 if (irq == NULL) {
252 spin_unlock_irqrestore(&irq_lock, flags);
253 return;
254 }
255
256 os_set_pollfd(i, -1);
257 spin_unlock_irqrestore(&irq_lock, flags);
258
259 ignore_sigio_fd(fd);
260}
261
262/*
263 * Called just before shutdown in order to provide a clean exec
264 * environment in case the system is rebooting. No locking because
265 * that would cause a pointless shutdown hang if something hadn't
266 * released the lock.
267 */
268int deactivate_all_fds(void)
269{
270 struct irq_fd *irq;
271 int err;
272
273 for (irq = active_fds; irq != NULL; irq = irq->next) {
274 err = os_clear_fd_async(irq->fd);
275 if (err)
276 return err;
277 }
278 /* If there is a signal already queued, after unblocking ignore it */
279 os_set_ioignore();
280
281 return 0;
282}
283
284/*
285 * do_IRQ handles all normal device IRQs (the special
286 * SMP cross-CPU interrupts have their own specific
287 * handlers).
288 */
289unsigned int do_IRQ(int irq, struct uml_pt_regs *regs)
290{
291 struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)regs);
292 irq_enter();
293 generic_handle_irq(irq);
294 irq_exit();
295 set_irq_regs(old_regs);
296 return 1;
297}
298
299int um_request_irq(unsigned int irq, int fd, int type,
300 irq_handler_t handler,
301 unsigned long irqflags, const char * devname,
302 void *dev_id)
303{
304 int err;
305
306 if (fd != -1) {
307 err = activate_fd(irq, fd, type, dev_id);
308 if (err)
309 return err;
310 }
311
312 return request_irq(irq, handler, irqflags, devname, dev_id);
313}
314
315EXPORT_SYMBOL(um_request_irq);
316EXPORT_SYMBOL(reactivate_fd);
317
318/*
319 * irq_chip must define at least enable/disable and ack when
320 * the edge handler is used.
321 */
322static void dummy(struct irq_data *d)
323{
324}
325
326/* This is used for everything else than the timer. */
327static struct irq_chip normal_irq_type = {
328 .name = "SIGIO",
329 .release = free_irq_by_irq_and_dev,
330 .irq_disable = dummy,
331 .irq_enable = dummy,
332 .irq_ack = dummy,
333};
334
335static struct irq_chip SIGVTALRM_irq_type = {
336 .name = "SIGVTALRM",
337 .release = free_irq_by_irq_and_dev,
338 .irq_disable = dummy,
339 .irq_enable = dummy,
340 .irq_ack = dummy,
341};
342
343void __init init_IRQ(void)
344{
345 int i;
346
347 irq_set_chip_and_handler(TIMER_IRQ, &SIGVTALRM_irq_type, handle_edge_irq);
348
349 for (i = 1; i < NR_IRQS; i++)
350 irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
351}
352
353/*
354 * IRQ stack entry and exit:
355 *
356 * Unlike i386, UML doesn't receive IRQs on the normal kernel stack
357 * and switch over to the IRQ stack after some preparation. We use
358 * sigaltstack to receive signals on a separate stack from the start.
359 * These two functions make sure the rest of the kernel won't be too
360 * upset by being on a different stack. The IRQ stack has a
361 * thread_info structure at the bottom so that current et al continue
362 * to work.
363 *
364 * to_irq_stack copies the current task's thread_info to the IRQ stack
365 * thread_info and sets the tasks's stack to point to the IRQ stack.
366 *
367 * from_irq_stack copies the thread_info struct back (flags may have
368 * been modified) and resets the task's stack pointer.
369 *
370 * Tricky bits -
371 *
372 * What happens when two signals race each other? UML doesn't block
373 * signals with sigprocmask, SA_DEFER, or sa_mask, so a second signal
374 * could arrive while a previous one is still setting up the
375 * thread_info.
376 *
377 * There are three cases -
378 * The first interrupt on the stack - sets up the thread_info and
379 * handles the interrupt
380 * A nested interrupt interrupting the copying of the thread_info -
381 * can't handle the interrupt, as the stack is in an unknown state
382 * A nested interrupt not interrupting the copying of the
383 * thread_info - doesn't do any setup, just handles the interrupt
384 *
385 * The first job is to figure out whether we interrupted stack setup.
386 * This is done by xchging the signal mask with thread_info->pending.
387 * If the value that comes back is zero, then there is no setup in
388 * progress, and the interrupt can be handled. If the value is
389 * non-zero, then there is stack setup in progress. In order to have
390 * the interrupt handled, we leave our signal in the mask, and it will
391 * be handled by the upper handler after it has set up the stack.
392 *
393 * Next is to figure out whether we are the outer handler or a nested
394 * one. As part of setting up the stack, thread_info->real_thread is
395 * set to non-NULL (and is reset to NULL on exit). This is the
396 * nesting indicator. If it is non-NULL, then the stack is already
397 * set up and the handler can run.
398 */
399
400static unsigned long pending_mask;
401
402unsigned long to_irq_stack(unsigned long *mask_out)
403{
404 struct thread_info *ti;
405 unsigned long mask, old;
406 int nested;
407
408 mask = xchg(&pending_mask, *mask_out);
409 if (mask != 0) {
410 /*
411 * If any interrupts come in at this point, we want to
412 * make sure that their bits aren't lost by our
413 * putting our bit in. So, this loop accumulates bits
414 * until xchg returns the same value that we put in.
415 * When that happens, there were no new interrupts,
416 * and pending_mask contains a bit for each interrupt
417 * that came in.
418 */
419 old = *mask_out;
420 do {
421 old |= mask;
422 mask = xchg(&pending_mask, old);
423 } while (mask != old);
424 return 1;
425 }
426
427 ti = current_thread_info();
428 nested = (ti->real_thread != NULL);
429 if (!nested) {
430 struct task_struct *task;
431 struct thread_info *tti;
432
433 task = cpu_tasks[ti->cpu].task;
434 tti = task_thread_info(task);
435
436 *ti = *tti;
437 ti->real_thread = tti;
438 task->stack = ti;
439 }
440
441 mask = xchg(&pending_mask, 0);
442 *mask_out |= mask | nested;
443 return 0;
444}
445
446unsigned long from_irq_stack(int nested)
447{
448 struct thread_info *ti, *to;
449 unsigned long mask;
450
451 ti = current_thread_info();
452
453 pending_mask = 1;
454
455 to = ti->real_thread;
456 current->stack = to;
457 ti->real_thread = NULL;
458 *to = *ti;
459
460 mask = xchg(&pending_mask, 0);
461 return mask & ~1;
462}
463