xref: /openbmc/linux/kernel/irq/spurious.c (revision 1eb4c977)
1 /*
2  * linux/kernel/irq/spurious.c
3  *
4  * Copyright (C) 1992, 1998-2004 Linus Torvalds, Ingo Molnar
5  *
6  * This file contains spurious interrupt handling.
7  */
8 
9 #include <linux/jiffies.h>
10 #include <linux/irq.h>
11 #include <linux/module.h>
12 #include <linux/kallsyms.h>
13 #include <linux/interrupt.h>
14 #include <linux/moduleparam.h>
15 #include <linux/timer.h>
16 
17 #include "internals.h"
18 
19 static int irqfixup __read_mostly;
20 
21 #define POLL_SPURIOUS_IRQ_INTERVAL (HZ/10)
22 static void poll_spurious_irqs(unsigned long dummy);
23 static DEFINE_TIMER(poll_spurious_irq_timer, poll_spurious_irqs, 0, 0);
24 static int irq_poll_cpu;
25 static atomic_t irq_poll_active;
26 
27 /*
28  * We wait here for a poller to finish.
29  *
30  * If the poll runs on this CPU, then we yell loudly and return
31  * false. That will leave the interrupt line disabled in the worst
32  * case, but it should never happen.
33  *
34  * We wait until the poller is done and then recheck disabled and
35  * action (about to be disabled). Only if it's still active, we return
36  * true and let the handler run.
37  */
38 bool irq_wait_for_poll(struct irq_desc *desc)
39 {
40 	if (WARN_ONCE(irq_poll_cpu == smp_processor_id(),
41 		      "irq poll in progress on cpu %d for irq %d\n",
42 		      smp_processor_id(), desc->irq_data.irq))
43 		return false;
44 
45 #ifdef CONFIG_SMP
46 	do {
47 		raw_spin_unlock(&desc->lock);
48 		while (irqd_irq_inprogress(&desc->irq_data))
49 			cpu_relax();
50 		raw_spin_lock(&desc->lock);
51 	} while (irqd_irq_inprogress(&desc->irq_data));
52 	/* Might have been disabled in meantime */
53 	return !irqd_irq_disabled(&desc->irq_data) && desc->action;
54 #else
55 	return false;
56 #endif
57 }
58 
59 
60 /*
61  * Recovery handler for misrouted interrupts.
62  */
63 static int try_one_irq(int irq, struct irq_desc *desc, bool force)
64 {
65 	irqreturn_t ret = IRQ_NONE;
66 	struct irqaction *action;
67 
68 	raw_spin_lock(&desc->lock);
69 
70 	/* PER_CPU and nested thread interrupts are never polled */
71 	if (irq_settings_is_per_cpu(desc) || irq_settings_is_nested_thread(desc))
72 		goto out;
73 
74 	/*
75 	 * Do not poll disabled interrupts unless the spurious
76 	 * disabled poller asks explicitely.
77 	 */
78 	if (irqd_irq_disabled(&desc->irq_data) && !force)
79 		goto out;
80 
81 	/*
82 	 * All handlers must agree on IRQF_SHARED, so we test just the
83 	 * first. Check for action->next as well.
84 	 */
85 	action = desc->action;
86 	if (!action || !(action->flags & IRQF_SHARED) ||
87 	    (action->flags & __IRQF_TIMER) ||
88 	    (action->handler(irq, action->dev_id) == IRQ_HANDLED) ||
89 	    !action->next)
90 		goto out;
91 
92 	/* Already running on another processor */
93 	if (irqd_irq_inprogress(&desc->irq_data)) {
94 		/*
95 		 * Already running: If it is shared get the other
96 		 * CPU to go looking for our mystery interrupt too
97 		 */
98 		desc->istate |= IRQS_PENDING;
99 		goto out;
100 	}
101 
102 	/* Mark it poll in progress */
103 	desc->istate |= IRQS_POLL_INPROGRESS;
104 	do {
105 		if (handle_irq_event(desc) == IRQ_HANDLED)
106 			ret = IRQ_HANDLED;
107 		action = desc->action;
108 	} while ((desc->istate & IRQS_PENDING) && action);
109 	desc->istate &= ~IRQS_POLL_INPROGRESS;
110 out:
111 	raw_spin_unlock(&desc->lock);
112 	return ret == IRQ_HANDLED;
113 }
114 
115 static int misrouted_irq(int irq)
116 {
117 	struct irq_desc *desc;
118 	int i, ok = 0;
119 
120 	if (atomic_inc_return(&irq_poll_active) != 1)
121 		goto out;
122 
123 	irq_poll_cpu = smp_processor_id();
124 
125 	for_each_irq_desc(i, desc) {
126 		if (!i)
127 			 continue;
128 
129 		if (i == irq)	/* Already tried */
130 			continue;
131 
132 		if (try_one_irq(i, desc, false))
133 			ok = 1;
134 	}
135 out:
136 	atomic_dec(&irq_poll_active);
137 	/* So the caller can adjust the irq error counts */
138 	return ok;
139 }
140 
141 static void poll_spurious_irqs(unsigned long dummy)
142 {
143 	struct irq_desc *desc;
144 	int i;
145 
146 	if (atomic_inc_return(&irq_poll_active) != 1)
147 		goto out;
148 	irq_poll_cpu = smp_processor_id();
149 
150 	for_each_irq_desc(i, desc) {
151 		unsigned int state;
152 
153 		if (!i)
154 			 continue;
155 
156 		/* Racy but it doesn't matter */
157 		state = desc->istate;
158 		barrier();
159 		if (!(state & IRQS_SPURIOUS_DISABLED))
160 			continue;
161 
162 		local_irq_disable();
163 		try_one_irq(i, desc, true);
164 		local_irq_enable();
165 	}
166 out:
167 	atomic_dec(&irq_poll_active);
168 	mod_timer(&poll_spurious_irq_timer,
169 		  jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
170 }
171 
172 static inline int bad_action_ret(irqreturn_t action_ret)
173 {
174 	if (likely(action_ret <= (IRQ_HANDLED | IRQ_WAKE_THREAD)))
175 		return 0;
176 	return 1;
177 }
178 
179 /*
180  * If 99,900 of the previous 100,000 interrupts have not been handled
181  * then assume that the IRQ is stuck in some manner. Drop a diagnostic
182  * and try to turn the IRQ off.
183  *
184  * (The other 100-of-100,000 interrupts may have been a correctly
185  *  functioning device sharing an IRQ with the failing one)
186  */
187 static void
188 __report_bad_irq(unsigned int irq, struct irq_desc *desc,
189 		 irqreturn_t action_ret)
190 {
191 	struct irqaction *action;
192 	unsigned long flags;
193 
194 	if (bad_action_ret(action_ret)) {
195 		printk(KERN_ERR "irq event %d: bogus return value %x\n",
196 				irq, action_ret);
197 	} else {
198 		printk(KERN_ERR "irq %d: nobody cared (try booting with "
199 				"the \"irqpoll\" option)\n", irq);
200 	}
201 	dump_stack();
202 	printk(KERN_ERR "handlers:\n");
203 
204 	/*
205 	 * We need to take desc->lock here. note_interrupt() is called
206 	 * w/o desc->lock held, but IRQ_PROGRESS set. We might race
207 	 * with something else removing an action. It's ok to take
208 	 * desc->lock here. See synchronize_irq().
209 	 */
210 	raw_spin_lock_irqsave(&desc->lock, flags);
211 	action = desc->action;
212 	while (action) {
213 		printk(KERN_ERR "[<%p>] %pf", action->handler, action->handler);
214 		if (action->thread_fn)
215 			printk(KERN_CONT " threaded [<%p>] %pf",
216 					action->thread_fn, action->thread_fn);
217 		printk(KERN_CONT "\n");
218 		action = action->next;
219 	}
220 	raw_spin_unlock_irqrestore(&desc->lock, flags);
221 }
222 
223 static void
224 report_bad_irq(unsigned int irq, struct irq_desc *desc, irqreturn_t action_ret)
225 {
226 	static int count = 100;
227 
228 	if (count > 0) {
229 		count--;
230 		__report_bad_irq(irq, desc, action_ret);
231 	}
232 }
233 
234 static inline int
235 try_misrouted_irq(unsigned int irq, struct irq_desc *desc,
236 		  irqreturn_t action_ret)
237 {
238 	struct irqaction *action;
239 
240 	if (!irqfixup)
241 		return 0;
242 
243 	/* We didn't actually handle the IRQ - see if it was misrouted? */
244 	if (action_ret == IRQ_NONE)
245 		return 1;
246 
247 	/*
248 	 * But for 'irqfixup == 2' we also do it for handled interrupts if
249 	 * they are marked as IRQF_IRQPOLL (or for irq zero, which is the
250 	 * traditional PC timer interrupt.. Legacy)
251 	 */
252 	if (irqfixup < 2)
253 		return 0;
254 
255 	if (!irq)
256 		return 1;
257 
258 	/*
259 	 * Since we don't get the descriptor lock, "action" can
260 	 * change under us.  We don't really care, but we don't
261 	 * want to follow a NULL pointer. So tell the compiler to
262 	 * just load it once by using a barrier.
263 	 */
264 	action = desc->action;
265 	barrier();
266 	return action && (action->flags & IRQF_IRQPOLL);
267 }
268 
269 void note_interrupt(unsigned int irq, struct irq_desc *desc,
270 		    irqreturn_t action_ret)
271 {
272 	if (desc->istate & IRQS_POLL_INPROGRESS)
273 		return;
274 
275 	/* we get here again via the threaded handler */
276 	if (action_ret == IRQ_WAKE_THREAD)
277 		return;
278 
279 	if (bad_action_ret(action_ret)) {
280 		report_bad_irq(irq, desc, action_ret);
281 		return;
282 	}
283 
284 	if (unlikely(action_ret == IRQ_NONE)) {
285 		/*
286 		 * If we are seeing only the odd spurious IRQ caused by
287 		 * bus asynchronicity then don't eventually trigger an error,
288 		 * otherwise the counter becomes a doomsday timer for otherwise
289 		 * working systems
290 		 */
291 		if (time_after(jiffies, desc->last_unhandled + HZ/10))
292 			desc->irqs_unhandled = 1;
293 		else
294 			desc->irqs_unhandled++;
295 		desc->last_unhandled = jiffies;
296 	}
297 
298 	if (unlikely(try_misrouted_irq(irq, desc, action_ret))) {
299 		int ok = misrouted_irq(irq);
300 		if (action_ret == IRQ_NONE)
301 			desc->irqs_unhandled -= ok;
302 	}
303 
304 	desc->irq_count++;
305 	if (likely(desc->irq_count < 100000))
306 		return;
307 
308 	desc->irq_count = 0;
309 	if (unlikely(desc->irqs_unhandled > 99900)) {
310 		/*
311 		 * The interrupt is stuck
312 		 */
313 		__report_bad_irq(irq, desc, action_ret);
314 		/*
315 		 * Now kill the IRQ
316 		 */
317 		printk(KERN_EMERG "Disabling IRQ #%d\n", irq);
318 		desc->istate |= IRQS_SPURIOUS_DISABLED;
319 		desc->depth++;
320 		irq_disable(desc);
321 
322 		mod_timer(&poll_spurious_irq_timer,
323 			  jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
324 	}
325 	desc->irqs_unhandled = 0;
326 }
327 
328 bool noirqdebug __read_mostly;
329 
330 int noirqdebug_setup(char *str)
331 {
332 	noirqdebug = 1;
333 	printk(KERN_INFO "IRQ lockup detection disabled\n");
334 
335 	return 1;
336 }
337 
338 __setup("noirqdebug", noirqdebug_setup);
339 module_param(noirqdebug, bool, 0644);
340 MODULE_PARM_DESC(noirqdebug, "Disable irq lockup detection when true");
341 
342 static int __init irqfixup_setup(char *str)
343 {
344 	irqfixup = 1;
345 	printk(KERN_WARNING "Misrouted IRQ fixup support enabled.\n");
346 	printk(KERN_WARNING "This may impact system performance.\n");
347 
348 	return 1;
349 }
350 
351 __setup("irqfixup", irqfixup_setup);
352 module_param(irqfixup, int, 0644);
353 
354 static int __init irqpoll_setup(char *str)
355 {
356 	irqfixup = 2;
357 	printk(KERN_WARNING "Misrouted IRQ fixup and polling support "
358 				"enabled\n");
359 	printk(KERN_WARNING "This may significantly impact system "
360 				"performance\n");
361 	return 1;
362 }
363 
364 __setup("irqpoll", irqpoll_setup);
365