xref: /openbmc/linux/arch/x86/kernel/irq.c (revision 84d517f3)
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 #if IS_ENABLED(CONFIG_HYPERV) || defined(CONFIG_XEN)
129 	seq_printf(p, "%*s: ", prec, "THR");
130 	for_each_online_cpu(j)
131 		seq_printf(p, "%10u ", irq_stats(j)->irq_hv_callback_count);
132 	seq_printf(p, "  Hypervisor callback interrupts\n");
133 #endif
134 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
135 #if defined(CONFIG_X86_IO_APIC)
136 	seq_printf(p, "%*s: %10u\n", prec, "MIS", atomic_read(&irq_mis_count));
137 #endif
138 	return 0;
139 }
140 
141 /*
142  * /proc/stat helpers
143  */
144 u64 arch_irq_stat_cpu(unsigned int cpu)
145 {
146 	u64 sum = irq_stats(cpu)->__nmi_count;
147 
148 #ifdef CONFIG_X86_LOCAL_APIC
149 	sum += irq_stats(cpu)->apic_timer_irqs;
150 	sum += irq_stats(cpu)->irq_spurious_count;
151 	sum += irq_stats(cpu)->apic_perf_irqs;
152 	sum += irq_stats(cpu)->apic_irq_work_irqs;
153 	sum += irq_stats(cpu)->icr_read_retry_count;
154 #endif
155 	if (x86_platform_ipi_callback)
156 		sum += irq_stats(cpu)->x86_platform_ipis;
157 #ifdef CONFIG_SMP
158 	sum += irq_stats(cpu)->irq_resched_count;
159 	sum += irq_stats(cpu)->irq_call_count;
160 #endif
161 #ifdef CONFIG_X86_THERMAL_VECTOR
162 	sum += irq_stats(cpu)->irq_thermal_count;
163 #endif
164 #ifdef CONFIG_X86_MCE_THRESHOLD
165 	sum += irq_stats(cpu)->irq_threshold_count;
166 #endif
167 #ifdef CONFIG_X86_MCE
168 	sum += per_cpu(mce_exception_count, cpu);
169 	sum += per_cpu(mce_poll_count, cpu);
170 #endif
171 	return sum;
172 }
173 
174 u64 arch_irq_stat(void)
175 {
176 	u64 sum = atomic_read(&irq_err_count);
177 	return sum;
178 }
179 
180 
181 /*
182  * do_IRQ handles all normal device IRQ's (the special
183  * SMP cross-CPU interrupts have their own specific
184  * handlers).
185  */
186 __visible unsigned int __irq_entry do_IRQ(struct pt_regs *regs)
187 {
188 	struct pt_regs *old_regs = set_irq_regs(regs);
189 
190 	/* high bit used in ret_from_ code  */
191 	unsigned vector = ~regs->orig_ax;
192 	unsigned irq;
193 
194 	irq_enter();
195 	exit_idle();
196 
197 	irq = __this_cpu_read(vector_irq[vector]);
198 
199 	if (!handle_irq(irq, regs)) {
200 		ack_APIC_irq();
201 
202 		if (irq != VECTOR_RETRIGGERED) {
203 			pr_emerg_ratelimited("%s: %d.%d No irq handler for vector (irq %d)\n",
204 					     __func__, smp_processor_id(),
205 					     vector, irq);
206 		} else {
207 			__this_cpu_write(vector_irq[vector], VECTOR_UNDEFINED);
208 		}
209 	}
210 
211 	irq_exit();
212 
213 	set_irq_regs(old_regs);
214 	return 1;
215 }
216 
217 /*
218  * Handler for X86_PLATFORM_IPI_VECTOR.
219  */
220 void __smp_x86_platform_ipi(void)
221 {
222 	inc_irq_stat(x86_platform_ipis);
223 
224 	if (x86_platform_ipi_callback)
225 		x86_platform_ipi_callback();
226 }
227 
228 __visible void smp_x86_platform_ipi(struct pt_regs *regs)
229 {
230 	struct pt_regs *old_regs = set_irq_regs(regs);
231 
232 	entering_ack_irq();
233 	__smp_x86_platform_ipi();
234 	exiting_irq();
235 	set_irq_regs(old_regs);
236 }
237 
238 #ifdef CONFIG_HAVE_KVM
239 /*
240  * Handler for POSTED_INTERRUPT_VECTOR.
241  */
242 __visible void smp_kvm_posted_intr_ipi(struct pt_regs *regs)
243 {
244 	struct pt_regs *old_regs = set_irq_regs(regs);
245 
246 	ack_APIC_irq();
247 
248 	irq_enter();
249 
250 	exit_idle();
251 
252 	inc_irq_stat(kvm_posted_intr_ipis);
253 
254 	irq_exit();
255 
256 	set_irq_regs(old_regs);
257 }
258 #endif
259 
260 __visible void smp_trace_x86_platform_ipi(struct pt_regs *regs)
261 {
262 	struct pt_regs *old_regs = set_irq_regs(regs);
263 
264 	entering_ack_irq();
265 	trace_x86_platform_ipi_entry(X86_PLATFORM_IPI_VECTOR);
266 	__smp_x86_platform_ipi();
267 	trace_x86_platform_ipi_exit(X86_PLATFORM_IPI_VECTOR);
268 	exiting_irq();
269 	set_irq_regs(old_regs);
270 }
271 
272 EXPORT_SYMBOL_GPL(vector_used_by_percpu_irq);
273 
274 #ifdef CONFIG_HOTPLUG_CPU
275 
276 /* These two declarations are only used in check_irq_vectors_for_cpu_disable()
277  * below, which is protected by stop_machine().  Putting them on the stack
278  * results in a stack frame overflow.  Dynamically allocating could result in a
279  * failure so declare these two cpumasks as global.
280  */
281 static struct cpumask affinity_new, online_new;
282 
283 /*
284  * This cpu is going to be removed and its vectors migrated to the remaining
285  * online cpus.  Check to see if there are enough vectors in the remaining cpus.
286  * This function is protected by stop_machine().
287  */
288 int check_irq_vectors_for_cpu_disable(void)
289 {
290 	int irq, cpu;
291 	unsigned int this_cpu, vector, this_count, count;
292 	struct irq_desc *desc;
293 	struct irq_data *data;
294 
295 	this_cpu = smp_processor_id();
296 	cpumask_copy(&online_new, cpu_online_mask);
297 	cpu_clear(this_cpu, online_new);
298 
299 	this_count = 0;
300 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
301 		irq = __this_cpu_read(vector_irq[vector]);
302 		if (irq >= 0) {
303 			desc = irq_to_desc(irq);
304 			data = irq_desc_get_irq_data(desc);
305 			cpumask_copy(&affinity_new, data->affinity);
306 			cpu_clear(this_cpu, affinity_new);
307 
308 			/* Do not count inactive or per-cpu irqs. */
309 			if (!irq_has_action(irq) || irqd_is_per_cpu(data))
310 				continue;
311 
312 			/*
313 			 * A single irq may be mapped to multiple
314 			 * cpu's vector_irq[] (for example IOAPIC cluster
315 			 * mode).  In this case we have two
316 			 * possibilities:
317 			 *
318 			 * 1) the resulting affinity mask is empty; that is
319 			 * this the down'd cpu is the last cpu in the irq's
320 			 * affinity mask, or
321 			 *
322 			 * 2) the resulting affinity mask is no longer
323 			 * a subset of the online cpus but the affinity
324 			 * mask is not zero; that is the down'd cpu is the
325 			 * last online cpu in a user set affinity mask.
326 			 */
327 			if (cpumask_empty(&affinity_new) ||
328 			    !cpumask_subset(&affinity_new, &online_new))
329 				this_count++;
330 		}
331 	}
332 
333 	count = 0;
334 	for_each_online_cpu(cpu) {
335 		if (cpu == this_cpu)
336 			continue;
337 		for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
338 		     vector++) {
339 			if (per_cpu(vector_irq, cpu)[vector] < 0)
340 				count++;
341 		}
342 	}
343 
344 	if (count < this_count) {
345 		pr_warn("CPU %d disable failed: CPU has %u vectors assigned and there are only %u available.\n",
346 			this_cpu, this_count, count);
347 		return -ERANGE;
348 	}
349 	return 0;
350 }
351 
352 /* A cpu has been removed from cpu_online_mask.  Reset irq affinities. */
353 void fixup_irqs(void)
354 {
355 	unsigned int irq, vector;
356 	static int warned;
357 	struct irq_desc *desc;
358 	struct irq_data *data;
359 	struct irq_chip *chip;
360 
361 	for_each_irq_desc(irq, desc) {
362 		int break_affinity = 0;
363 		int set_affinity = 1;
364 		const struct cpumask *affinity;
365 
366 		if (!desc)
367 			continue;
368 		if (irq == 2)
369 			continue;
370 
371 		/* interrupt's are disabled at this point */
372 		raw_spin_lock(&desc->lock);
373 
374 		data = irq_desc_get_irq_data(desc);
375 		affinity = data->affinity;
376 		if (!irq_has_action(irq) || irqd_is_per_cpu(data) ||
377 		    cpumask_subset(affinity, cpu_online_mask)) {
378 			raw_spin_unlock(&desc->lock);
379 			continue;
380 		}
381 
382 		/*
383 		 * Complete the irq move. This cpu is going down and for
384 		 * non intr-remapping case, we can't wait till this interrupt
385 		 * arrives at this cpu before completing the irq move.
386 		 */
387 		irq_force_complete_move(irq);
388 
389 		if (cpumask_any_and(affinity, cpu_online_mask) >= nr_cpu_ids) {
390 			break_affinity = 1;
391 			affinity = cpu_online_mask;
392 		}
393 
394 		chip = irq_data_get_irq_chip(data);
395 		if (!irqd_can_move_in_process_context(data) && chip->irq_mask)
396 			chip->irq_mask(data);
397 
398 		if (chip->irq_set_affinity)
399 			chip->irq_set_affinity(data, affinity, true);
400 		else if (!(warned++))
401 			set_affinity = 0;
402 
403 		/*
404 		 * We unmask if the irq was not marked masked by the
405 		 * core code. That respects the lazy irq disable
406 		 * behaviour.
407 		 */
408 		if (!irqd_can_move_in_process_context(data) &&
409 		    !irqd_irq_masked(data) && chip->irq_unmask)
410 			chip->irq_unmask(data);
411 
412 		raw_spin_unlock(&desc->lock);
413 
414 		if (break_affinity && set_affinity)
415 			pr_notice("Broke affinity for irq %i\n", irq);
416 		else if (!set_affinity)
417 			pr_notice("Cannot set affinity for irq %i\n", irq);
418 	}
419 
420 	/*
421 	 * We can remove mdelay() and then send spuriuous interrupts to
422 	 * new cpu targets for all the irqs that were handled previously by
423 	 * this cpu. While it works, I have seen spurious interrupt messages
424 	 * (nothing wrong but still...).
425 	 *
426 	 * So for now, retain mdelay(1) and check the IRR and then send those
427 	 * interrupts to new targets as this cpu is already offlined...
428 	 */
429 	mdelay(1);
430 
431 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
432 		unsigned int irr;
433 
434 		if (__this_cpu_read(vector_irq[vector]) <= VECTOR_UNDEFINED)
435 			continue;
436 
437 		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
438 		if (irr  & (1 << (vector % 32))) {
439 			irq = __this_cpu_read(vector_irq[vector]);
440 
441 			desc = irq_to_desc(irq);
442 			data = irq_desc_get_irq_data(desc);
443 			chip = irq_data_get_irq_chip(data);
444 			raw_spin_lock(&desc->lock);
445 			if (chip->irq_retrigger) {
446 				chip->irq_retrigger(data);
447 				__this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
448 			}
449 			raw_spin_unlock(&desc->lock);
450 		}
451 		if (__this_cpu_read(vector_irq[vector]) != VECTOR_RETRIGGERED)
452 			__this_cpu_write(vector_irq[vector], VECTOR_UNDEFINED);
453 	}
454 }
455 #endif
456