xref: /openbmc/linux/kernel/irq/spurious.c (revision 565d76cb)
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 (desc->istate & IRQS_INPROGRESS)
49 			cpu_relax();
50 		raw_spin_lock(&desc->lock);
51 	} while (desc->istate & IRQS_INPROGRESS);
52 	/* Might have been disabled in meantime */
53 	return !(desc->istate & IRQS_DISABLED) && 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 ((desc->istate & IRQS_DISABLED) && !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) || !action->next)
88 		goto out;
89 
90 	/* Already running on another processor */
91 	if (desc->istate & IRQS_INPROGRESS) {
92 		/*
93 		 * Already running: If it is shared get the other
94 		 * CPU to go looking for our mystery interrupt too
95 		 */
96 		irq_compat_set_pending(desc);
97 		desc->istate |= IRQS_PENDING;
98 		goto out;
99 	}
100 
101 	/* Mark it poll in progress */
102 	desc->istate |= IRQS_POLL_INPROGRESS;
103 	do {
104 		if (handle_irq_event(desc) == IRQ_HANDLED)
105 			ret = IRQ_HANDLED;
106 		action = desc->action;
107 	} while ((desc->istate & IRQS_PENDING) && action);
108 	desc->istate &= ~IRQS_POLL_INPROGRESS;
109 out:
110 	raw_spin_unlock(&desc->lock);
111 	return ret == IRQ_HANDLED;
112 }
113 
114 static int misrouted_irq(int irq)
115 {
116 	struct irq_desc *desc;
117 	int i, ok = 0;
118 
119 	if (atomic_inc_return(&irq_poll_active) == 1)
120 		goto out;
121 
122 	irq_poll_cpu = smp_processor_id();
123 
124 	for_each_irq_desc(i, desc) {
125 		if (!i)
126 			 continue;
127 
128 		if (i == irq)	/* Already tried */
129 			continue;
130 
131 		if (try_one_irq(i, desc, false))
132 			ok = 1;
133 	}
134 out:
135 	atomic_dec(&irq_poll_active);
136 	/* So the caller can adjust the irq error counts */
137 	return ok;
138 }
139 
140 static void poll_spurious_irqs(unsigned long dummy)
141 {
142 	struct irq_desc *desc;
143 	int i;
144 
145 	if (atomic_inc_return(&irq_poll_active) != 1)
146 		goto out;
147 	irq_poll_cpu = smp_processor_id();
148 
149 	for_each_irq_desc(i, desc) {
150 		unsigned int state;
151 
152 		if (!i)
153 			 continue;
154 
155 		/* Racy but it doesn't matter */
156 		state = desc->istate;
157 		barrier();
158 		if (!(state & IRQS_SPURIOUS_DISABLED))
159 			continue;
160 
161 		local_irq_disable();
162 		try_one_irq(i, desc, true);
163 		local_irq_enable();
164 	}
165 out:
166 	atomic_dec(&irq_poll_active);
167 	mod_timer(&poll_spurious_irq_timer,
168 		  jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
169 }
170 
171 /*
172  * If 99,900 of the previous 100,000 interrupts have not been handled
173  * then assume that the IRQ is stuck in some manner. Drop a diagnostic
174  * and try to turn the IRQ off.
175  *
176  * (The other 100-of-100,000 interrupts may have been a correctly
177  *  functioning device sharing an IRQ with the failing one)
178  */
179 static void
180 __report_bad_irq(unsigned int irq, struct irq_desc *desc,
181 		 irqreturn_t action_ret)
182 {
183 	struct irqaction *action;
184 	unsigned long flags;
185 
186 	if (action_ret != IRQ_HANDLED && action_ret != IRQ_NONE) {
187 		printk(KERN_ERR "irq event %d: bogus return value %x\n",
188 				irq, action_ret);
189 	} else {
190 		printk(KERN_ERR "irq %d: nobody cared (try booting with "
191 				"the \"irqpoll\" option)\n", irq);
192 	}
193 	dump_stack();
194 	printk(KERN_ERR "handlers:\n");
195 
196 	/*
197 	 * We need to take desc->lock here. note_interrupt() is called
198 	 * w/o desc->lock held, but IRQ_PROGRESS set. We might race
199 	 * with something else removing an action. It's ok to take
200 	 * desc->lock here. See synchronize_irq().
201 	 */
202 	raw_spin_lock_irqsave(&desc->lock, flags);
203 	action = desc->action;
204 	while (action) {
205 		printk(KERN_ERR "[<%p>]", action->handler);
206 		print_symbol(" (%s)",
207 			(unsigned long)action->handler);
208 		printk("\n");
209 		action = action->next;
210 	}
211 	raw_spin_unlock_irqrestore(&desc->lock, flags);
212 }
213 
214 static void
215 report_bad_irq(unsigned int irq, struct irq_desc *desc, irqreturn_t action_ret)
216 {
217 	static int count = 100;
218 
219 	if (count > 0) {
220 		count--;
221 		__report_bad_irq(irq, desc, action_ret);
222 	}
223 }
224 
225 static inline int
226 try_misrouted_irq(unsigned int irq, struct irq_desc *desc,
227 		  irqreturn_t action_ret)
228 {
229 	struct irqaction *action;
230 
231 	if (!irqfixup)
232 		return 0;
233 
234 	/* We didn't actually handle the IRQ - see if it was misrouted? */
235 	if (action_ret == IRQ_NONE)
236 		return 1;
237 
238 	/*
239 	 * But for 'irqfixup == 2' we also do it for handled interrupts if
240 	 * they are marked as IRQF_IRQPOLL (or for irq zero, which is the
241 	 * traditional PC timer interrupt.. Legacy)
242 	 */
243 	if (irqfixup < 2)
244 		return 0;
245 
246 	if (!irq)
247 		return 1;
248 
249 	/*
250 	 * Since we don't get the descriptor lock, "action" can
251 	 * change under us.  We don't really care, but we don't
252 	 * want to follow a NULL pointer. So tell the compiler to
253 	 * just load it once by using a barrier.
254 	 */
255 	action = desc->action;
256 	barrier();
257 	return action && (action->flags & IRQF_IRQPOLL);
258 }
259 
260 void note_interrupt(unsigned int irq, struct irq_desc *desc,
261 		    irqreturn_t action_ret)
262 {
263 	if (desc->istate & IRQS_POLL_INPROGRESS)
264 		return;
265 
266 	if (unlikely(action_ret != IRQ_HANDLED)) {
267 		/*
268 		 * If we are seeing only the odd spurious IRQ caused by
269 		 * bus asynchronicity then don't eventually trigger an error,
270 		 * otherwise the counter becomes a doomsday timer for otherwise
271 		 * working systems
272 		 */
273 		if (time_after(jiffies, desc->last_unhandled + HZ/10))
274 			desc->irqs_unhandled = 1;
275 		else
276 			desc->irqs_unhandled++;
277 		desc->last_unhandled = jiffies;
278 		if (unlikely(action_ret != IRQ_NONE))
279 			report_bad_irq(irq, desc, action_ret);
280 	}
281 
282 	if (unlikely(try_misrouted_irq(irq, desc, action_ret))) {
283 		int ok = misrouted_irq(irq);
284 		if (action_ret == IRQ_NONE)
285 			desc->irqs_unhandled -= ok;
286 	}
287 
288 	desc->irq_count++;
289 	if (likely(desc->irq_count < 100000))
290 		return;
291 
292 	desc->irq_count = 0;
293 	if (unlikely(desc->irqs_unhandled > 99900)) {
294 		/*
295 		 * The interrupt is stuck
296 		 */
297 		__report_bad_irq(irq, desc, action_ret);
298 		/*
299 		 * Now kill the IRQ
300 		 */
301 		printk(KERN_EMERG "Disabling IRQ #%d\n", irq);
302 		desc->istate |= IRQS_SPURIOUS_DISABLED;
303 		desc->depth++;
304 		irq_disable(desc);
305 
306 		mod_timer(&poll_spurious_irq_timer,
307 			  jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
308 	}
309 	desc->irqs_unhandled = 0;
310 }
311 
312 int noirqdebug __read_mostly;
313 
314 int noirqdebug_setup(char *str)
315 {
316 	noirqdebug = 1;
317 	printk(KERN_INFO "IRQ lockup detection disabled\n");
318 
319 	return 1;
320 }
321 
322 __setup("noirqdebug", noirqdebug_setup);
323 module_param(noirqdebug, bool, 0644);
324 MODULE_PARM_DESC(noirqdebug, "Disable irq lockup detection when true");
325 
326 static int __init irqfixup_setup(char *str)
327 {
328 	irqfixup = 1;
329 	printk(KERN_WARNING "Misrouted IRQ fixup support enabled.\n");
330 	printk(KERN_WARNING "This may impact system performance.\n");
331 
332 	return 1;
333 }
334 
335 __setup("irqfixup", irqfixup_setup);
336 module_param(irqfixup, int, 0644);
337 
338 static int __init irqpoll_setup(char *str)
339 {
340 	irqfixup = 2;
341 	printk(KERN_WARNING "Misrouted IRQ fixup and polling support "
342 				"enabled\n");
343 	printk(KERN_WARNING "This may significantly impact system "
344 				"performance\n");
345 	return 1;
346 }
347 
348 __setup("irqpoll", irqpoll_setup);
349