xref: /linux/arch/x86/kernel/irq.c (revision d91517839e5d95adc0cf4b28caa7af62a71de526)
1 /*
2  * Common interrupt code for 32 and 64 bit
3  */
4 #include <linux/cpu.h>
5 #include <linux/interrupt.h>
6 #include <linux/kernel_stat.h>
7 #include <linux/of.h>
8 #include <linux/seq_file.h>
9 #include <linux/smp.h>
10 #include <linux/ftrace.h>
11 #include <linux/delay.h>
12 #include <linux/export.h>
13 
14 #include <asm/apic.h>
15 #include <asm/io_apic.h>
16 #include <asm/irq.h>
17 #include <asm/idle.h>
18 #include <asm/mce.h>
19 #include <asm/hw_irq.h>
20 
21 #define CREATE_TRACE_POINTS
22 #include <asm/trace/irq_vectors.h>
23 
24 atomic_t irq_err_count;
25 
26 /* Function pointer for generic interrupt vector handling */
27 void (*x86_platform_ipi_callback)(void) = NULL;
28 
29 /*
30  * 'what should we do if we get a hw irq event on an illegal vector'.
31  * each architecture has to answer this themselves.
32  */
33 void ack_bad_irq(unsigned int irq)
34 {
35 	if (printk_ratelimit())
36 		pr_err("unexpected IRQ trap at vector %02x\n", irq);
37 
38 	/*
39 	 * Currently unexpected vectors happen only on SMP and APIC.
40 	 * We _must_ ack these because every local APIC has only N
41 	 * irq slots per priority level, and a 'hanging, unacked' IRQ
42 	 * holds up an irq slot - in excessive cases (when multiple
43 	 * unexpected vectors occur) that might lock up the APIC
44 	 * completely.
45 	 * But only ack when the APIC is enabled -AK
46 	 */
47 	ack_APIC_irq();
48 }
49 
50 #define irq_stats(x)		(&per_cpu(irq_stat, x))
51 /*
52  * /proc/interrupts printing for arch specific interrupts
53  */
54 int arch_show_interrupts(struct seq_file *p, int prec)
55 {
56 	int j;
57 
58 	seq_printf(p, "%*s: ", prec, "NMI");
59 	for_each_online_cpu(j)
60 		seq_printf(p, "%10u ", irq_stats(j)->__nmi_count);
61 	seq_printf(p, "  Non-maskable interrupts\n");
62 #ifdef CONFIG_X86_LOCAL_APIC
63 	seq_printf(p, "%*s: ", prec, "LOC");
64 	for_each_online_cpu(j)
65 		seq_printf(p, "%10u ", irq_stats(j)->apic_timer_irqs);
66 	seq_printf(p, "  Local timer interrupts\n");
67 
68 	seq_printf(p, "%*s: ", prec, "SPU");
69 	for_each_online_cpu(j)
70 		seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count);
71 	seq_printf(p, "  Spurious interrupts\n");
72 	seq_printf(p, "%*s: ", prec, "PMI");
73 	for_each_online_cpu(j)
74 		seq_printf(p, "%10u ", irq_stats(j)->apic_perf_irqs);
75 	seq_printf(p, "  Performance monitoring interrupts\n");
76 	seq_printf(p, "%*s: ", prec, "IWI");
77 	for_each_online_cpu(j)
78 		seq_printf(p, "%10u ", irq_stats(j)->apic_irq_work_irqs);
79 	seq_printf(p, "  IRQ work interrupts\n");
80 	seq_printf(p, "%*s: ", prec, "RTR");
81 	for_each_online_cpu(j)
82 		seq_printf(p, "%10u ", irq_stats(j)->icr_read_retry_count);
83 	seq_printf(p, "  APIC ICR read retries\n");
84 #endif
85 	if (x86_platform_ipi_callback) {
86 		seq_printf(p, "%*s: ", prec, "PLT");
87 		for_each_online_cpu(j)
88 			seq_printf(p, "%10u ", irq_stats(j)->x86_platform_ipis);
89 		seq_printf(p, "  Platform interrupts\n");
90 	}
91 #ifdef CONFIG_SMP
92 	seq_printf(p, "%*s: ", prec, "RES");
93 	for_each_online_cpu(j)
94 		seq_printf(p, "%10u ", irq_stats(j)->irq_resched_count);
95 	seq_printf(p, "  Rescheduling interrupts\n");
96 	seq_printf(p, "%*s: ", prec, "CAL");
97 	for_each_online_cpu(j)
98 		seq_printf(p, "%10u ", irq_stats(j)->irq_call_count -
99 					irq_stats(j)->irq_tlb_count);
100 	seq_printf(p, "  Function call interrupts\n");
101 	seq_printf(p, "%*s: ", prec, "TLB");
102 	for_each_online_cpu(j)
103 		seq_printf(p, "%10u ", irq_stats(j)->irq_tlb_count);
104 	seq_printf(p, "  TLB shootdowns\n");
105 #endif
106 #ifdef CONFIG_X86_THERMAL_VECTOR
107 	seq_printf(p, "%*s: ", prec, "TRM");
108 	for_each_online_cpu(j)
109 		seq_printf(p, "%10u ", irq_stats(j)->irq_thermal_count);
110 	seq_printf(p, "  Thermal event interrupts\n");
111 #endif
112 #ifdef CONFIG_X86_MCE_THRESHOLD
113 	seq_printf(p, "%*s: ", prec, "THR");
114 	for_each_online_cpu(j)
115 		seq_printf(p, "%10u ", irq_stats(j)->irq_threshold_count);
116 	seq_printf(p, "  Threshold APIC interrupts\n");
117 #endif
118 #ifdef CONFIG_X86_MCE
119 	seq_printf(p, "%*s: ", prec, "MCE");
120 	for_each_online_cpu(j)
121 		seq_printf(p, "%10u ", per_cpu(mce_exception_count, j));
122 	seq_printf(p, "  Machine check exceptions\n");
123 	seq_printf(p, "%*s: ", prec, "MCP");
124 	for_each_online_cpu(j)
125 		seq_printf(p, "%10u ", per_cpu(mce_poll_count, j));
126 	seq_printf(p, "  Machine check polls\n");
127 #endif
128 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
129 #if defined(CONFIG_X86_IO_APIC)
130 	seq_printf(p, "%*s: %10u\n", prec, "MIS", atomic_read(&irq_mis_count));
131 #endif
132 	return 0;
133 }
134 
135 /*
136  * /proc/stat helpers
137  */
138 u64 arch_irq_stat_cpu(unsigned int cpu)
139 {
140 	u64 sum = irq_stats(cpu)->__nmi_count;
141 
142 #ifdef CONFIG_X86_LOCAL_APIC
143 	sum += irq_stats(cpu)->apic_timer_irqs;
144 	sum += irq_stats(cpu)->irq_spurious_count;
145 	sum += irq_stats(cpu)->apic_perf_irqs;
146 	sum += irq_stats(cpu)->apic_irq_work_irqs;
147 	sum += irq_stats(cpu)->icr_read_retry_count;
148 #endif
149 	if (x86_platform_ipi_callback)
150 		sum += irq_stats(cpu)->x86_platform_ipis;
151 #ifdef CONFIG_SMP
152 	sum += irq_stats(cpu)->irq_resched_count;
153 	sum += irq_stats(cpu)->irq_call_count;
154 #endif
155 #ifdef CONFIG_X86_THERMAL_VECTOR
156 	sum += irq_stats(cpu)->irq_thermal_count;
157 #endif
158 #ifdef CONFIG_X86_MCE_THRESHOLD
159 	sum += irq_stats(cpu)->irq_threshold_count;
160 #endif
161 #ifdef CONFIG_X86_MCE
162 	sum += per_cpu(mce_exception_count, cpu);
163 	sum += per_cpu(mce_poll_count, cpu);
164 #endif
165 	return sum;
166 }
167 
168 u64 arch_irq_stat(void)
169 {
170 	u64 sum = atomic_read(&irq_err_count);
171 	return sum;
172 }
173 
174 
175 /*
176  * do_IRQ handles all normal device IRQ's (the special
177  * SMP cross-CPU interrupts have their own specific
178  * handlers).
179  */
180 __visible unsigned int __irq_entry do_IRQ(struct pt_regs *regs)
181 {
182 	struct pt_regs *old_regs = set_irq_regs(regs);
183 
184 	/* high bit used in ret_from_ code  */
185 	unsigned vector = ~regs->orig_ax;
186 	unsigned irq;
187 
188 	irq_enter();
189 	exit_idle();
190 
191 	irq = __this_cpu_read(vector_irq[vector]);
192 
193 	if (!handle_irq(irq, regs)) {
194 		ack_APIC_irq();
195 
196 		if (irq != VECTOR_RETRIGGERED) {
197 			pr_emerg_ratelimited("%s: %d.%d No irq handler for vector (irq %d)\n",
198 					     __func__, smp_processor_id(),
199 					     vector, irq);
200 		} else {
201 			__this_cpu_write(vector_irq[vector], VECTOR_UNDEFINED);
202 		}
203 	}
204 
205 	irq_exit();
206 
207 	set_irq_regs(old_regs);
208 	return 1;
209 }
210 
211 /*
212  * Handler for X86_PLATFORM_IPI_VECTOR.
213  */
214 void __smp_x86_platform_ipi(void)
215 {
216 	inc_irq_stat(x86_platform_ipis);
217 
218 	if (x86_platform_ipi_callback)
219 		x86_platform_ipi_callback();
220 }
221 
222 __visible void smp_x86_platform_ipi(struct pt_regs *regs)
223 {
224 	struct pt_regs *old_regs = set_irq_regs(regs);
225 
226 	entering_ack_irq();
227 	__smp_x86_platform_ipi();
228 	exiting_irq();
229 	set_irq_regs(old_regs);
230 }
231 
232 #ifdef CONFIG_HAVE_KVM
233 /*
234  * Handler for POSTED_INTERRUPT_VECTOR.
235  */
236 __visible void smp_kvm_posted_intr_ipi(struct pt_regs *regs)
237 {
238 	struct pt_regs *old_regs = set_irq_regs(regs);
239 
240 	ack_APIC_irq();
241 
242 	irq_enter();
243 
244 	exit_idle();
245 
246 	inc_irq_stat(kvm_posted_intr_ipis);
247 
248 	irq_exit();
249 
250 	set_irq_regs(old_regs);
251 }
252 #endif
253 
254 __visible void smp_trace_x86_platform_ipi(struct pt_regs *regs)
255 {
256 	struct pt_regs *old_regs = set_irq_regs(regs);
257 
258 	entering_ack_irq();
259 	trace_x86_platform_ipi_entry(X86_PLATFORM_IPI_VECTOR);
260 	__smp_x86_platform_ipi();
261 	trace_x86_platform_ipi_exit(X86_PLATFORM_IPI_VECTOR);
262 	exiting_irq();
263 	set_irq_regs(old_regs);
264 }
265 
266 EXPORT_SYMBOL_GPL(vector_used_by_percpu_irq);
267 
268 #ifdef CONFIG_HOTPLUG_CPU
269 /*
270  * This cpu is going to be removed and its vectors migrated to the remaining
271  * online cpus.  Check to see if there are enough vectors in the remaining cpus.
272  * This function is protected by stop_machine().
273  */
274 int check_irq_vectors_for_cpu_disable(void)
275 {
276 	int irq, cpu;
277 	unsigned int this_cpu, vector, this_count, count;
278 	struct irq_desc *desc;
279 	struct irq_data *data;
280 	struct cpumask affinity_new, online_new;
281 
282 	this_cpu = smp_processor_id();
283 	cpumask_copy(&online_new, cpu_online_mask);
284 	cpu_clear(this_cpu, online_new);
285 
286 	this_count = 0;
287 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
288 		irq = __this_cpu_read(vector_irq[vector]);
289 		if (irq >= 0) {
290 			desc = irq_to_desc(irq);
291 			data = irq_desc_get_irq_data(desc);
292 			cpumask_copy(&affinity_new, data->affinity);
293 			cpu_clear(this_cpu, affinity_new);
294 
295 			/* Do not count inactive or per-cpu irqs. */
296 			if (!irq_has_action(irq) || irqd_is_per_cpu(data))
297 				continue;
298 
299 			/*
300 			 * A single irq may be mapped to multiple
301 			 * cpu's vector_irq[] (for example IOAPIC cluster
302 			 * mode).  In this case we have two
303 			 * possibilities:
304 			 *
305 			 * 1) the resulting affinity mask is empty; that is
306 			 * this the down'd cpu is the last cpu in the irq's
307 			 * affinity mask, or
308 			 *
309 			 * 2) the resulting affinity mask is no longer
310 			 * a subset of the online cpus but the affinity
311 			 * mask is not zero; that is the down'd cpu is the
312 			 * last online cpu in a user set affinity mask.
313 			 */
314 			if (cpumask_empty(&affinity_new) ||
315 			    !cpumask_subset(&affinity_new, &online_new))
316 				this_count++;
317 		}
318 	}
319 
320 	count = 0;
321 	for_each_online_cpu(cpu) {
322 		if (cpu == this_cpu)
323 			continue;
324 		for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
325 		     vector++) {
326 			if (per_cpu(vector_irq, cpu)[vector] < 0)
327 				count++;
328 		}
329 	}
330 
331 	if (count < this_count) {
332 		pr_warn("CPU %d disable failed: CPU has %u vectors assigned and there are only %u available.\n",
333 			this_cpu, this_count, count);
334 		return -ERANGE;
335 	}
336 	return 0;
337 }
338 
339 /* A cpu has been removed from cpu_online_mask.  Reset irq affinities. */
340 void fixup_irqs(void)
341 {
342 	unsigned int irq, vector;
343 	static int warned;
344 	struct irq_desc *desc;
345 	struct irq_data *data;
346 	struct irq_chip *chip;
347 
348 	for_each_irq_desc(irq, desc) {
349 		int break_affinity = 0;
350 		int set_affinity = 1;
351 		const struct cpumask *affinity;
352 
353 		if (!desc)
354 			continue;
355 		if (irq == 2)
356 			continue;
357 
358 		/* interrupt's are disabled at this point */
359 		raw_spin_lock(&desc->lock);
360 
361 		data = irq_desc_get_irq_data(desc);
362 		affinity = data->affinity;
363 		if (!irq_has_action(irq) || irqd_is_per_cpu(data) ||
364 		    cpumask_subset(affinity, cpu_online_mask)) {
365 			raw_spin_unlock(&desc->lock);
366 			continue;
367 		}
368 
369 		/*
370 		 * Complete the irq move. This cpu is going down and for
371 		 * non intr-remapping case, we can't wait till this interrupt
372 		 * arrives at this cpu before completing the irq move.
373 		 */
374 		irq_force_complete_move(irq);
375 
376 		if (cpumask_any_and(affinity, cpu_online_mask) >= nr_cpu_ids) {
377 			break_affinity = 1;
378 			affinity = cpu_online_mask;
379 		}
380 
381 		chip = irq_data_get_irq_chip(data);
382 		if (!irqd_can_move_in_process_context(data) && chip->irq_mask)
383 			chip->irq_mask(data);
384 
385 		if (chip->irq_set_affinity)
386 			chip->irq_set_affinity(data, affinity, true);
387 		else if (!(warned++))
388 			set_affinity = 0;
389 
390 		/*
391 		 * We unmask if the irq was not marked masked by the
392 		 * core code. That respects the lazy irq disable
393 		 * behaviour.
394 		 */
395 		if (!irqd_can_move_in_process_context(data) &&
396 		    !irqd_irq_masked(data) && chip->irq_unmask)
397 			chip->irq_unmask(data);
398 
399 		raw_spin_unlock(&desc->lock);
400 
401 		if (break_affinity && set_affinity)
402 			pr_notice("Broke affinity for irq %i\n", irq);
403 		else if (!set_affinity)
404 			pr_notice("Cannot set affinity for irq %i\n", irq);
405 	}
406 
407 	/*
408 	 * We can remove mdelay() and then send spuriuous interrupts to
409 	 * new cpu targets for all the irqs that were handled previously by
410 	 * this cpu. While it works, I have seen spurious interrupt messages
411 	 * (nothing wrong but still...).
412 	 *
413 	 * So for now, retain mdelay(1) and check the IRR and then send those
414 	 * interrupts to new targets as this cpu is already offlined...
415 	 */
416 	mdelay(1);
417 
418 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
419 		unsigned int irr;
420 
421 		if (__this_cpu_read(vector_irq[vector]) <= VECTOR_UNDEFINED)
422 			continue;
423 
424 		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
425 		if (irr  & (1 << (vector % 32))) {
426 			irq = __this_cpu_read(vector_irq[vector]);
427 
428 			desc = irq_to_desc(irq);
429 			data = irq_desc_get_irq_data(desc);
430 			chip = irq_data_get_irq_chip(data);
431 			raw_spin_lock(&desc->lock);
432 			if (chip->irq_retrigger) {
433 				chip->irq_retrigger(data);
434 				__this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
435 			}
436 			raw_spin_unlock(&desc->lock);
437 		}
438 		if (__this_cpu_read(vector_irq[vector]) != VECTOR_RETRIGGERED)
439 			__this_cpu_write(vector_irq[vector], VECTOR_UNDEFINED);
440 	}
441 }
442 #endif
443