xref: /openbmc/linux/arch/x86/kernel/irq.c (revision cd5d5810)
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 (printk_ratelimit())
197 			pr_emerg("%s: %d.%d No irq handler for vector (irq %d)\n",
198 				__func__, smp_processor_id(), vector, irq);
199 	}
200 
201 	irq_exit();
202 
203 	set_irq_regs(old_regs);
204 	return 1;
205 }
206 
207 /*
208  * Handler for X86_PLATFORM_IPI_VECTOR.
209  */
210 void __smp_x86_platform_ipi(void)
211 {
212 	inc_irq_stat(x86_platform_ipis);
213 
214 	if (x86_platform_ipi_callback)
215 		x86_platform_ipi_callback();
216 }
217 
218 __visible void smp_x86_platform_ipi(struct pt_regs *regs)
219 {
220 	struct pt_regs *old_regs = set_irq_regs(regs);
221 
222 	entering_ack_irq();
223 	__smp_x86_platform_ipi();
224 	exiting_irq();
225 	set_irq_regs(old_regs);
226 }
227 
228 #ifdef CONFIG_HAVE_KVM
229 /*
230  * Handler for POSTED_INTERRUPT_VECTOR.
231  */
232 __visible void smp_kvm_posted_intr_ipi(struct pt_regs *regs)
233 {
234 	struct pt_regs *old_regs = set_irq_regs(regs);
235 
236 	ack_APIC_irq();
237 
238 	irq_enter();
239 
240 	exit_idle();
241 
242 	inc_irq_stat(kvm_posted_intr_ipis);
243 
244 	irq_exit();
245 
246 	set_irq_regs(old_regs);
247 }
248 #endif
249 
250 __visible void smp_trace_x86_platform_ipi(struct pt_regs *regs)
251 {
252 	struct pt_regs *old_regs = set_irq_regs(regs);
253 
254 	entering_ack_irq();
255 	trace_x86_platform_ipi_entry(X86_PLATFORM_IPI_VECTOR);
256 	__smp_x86_platform_ipi();
257 	trace_x86_platform_ipi_exit(X86_PLATFORM_IPI_VECTOR);
258 	exiting_irq();
259 	set_irq_regs(old_regs);
260 }
261 
262 EXPORT_SYMBOL_GPL(vector_used_by_percpu_irq);
263 
264 #ifdef CONFIG_HOTPLUG_CPU
265 /* A cpu has been removed from cpu_online_mask.  Reset irq affinities. */
266 void fixup_irqs(void)
267 {
268 	unsigned int irq, vector;
269 	static int warned;
270 	struct irq_desc *desc;
271 	struct irq_data *data;
272 	struct irq_chip *chip;
273 
274 	for_each_irq_desc(irq, desc) {
275 		int break_affinity = 0;
276 		int set_affinity = 1;
277 		const struct cpumask *affinity;
278 
279 		if (!desc)
280 			continue;
281 		if (irq == 2)
282 			continue;
283 
284 		/* interrupt's are disabled at this point */
285 		raw_spin_lock(&desc->lock);
286 
287 		data = irq_desc_get_irq_data(desc);
288 		affinity = data->affinity;
289 		if (!irq_has_action(irq) || irqd_is_per_cpu(data) ||
290 		    cpumask_subset(affinity, cpu_online_mask)) {
291 			raw_spin_unlock(&desc->lock);
292 			continue;
293 		}
294 
295 		/*
296 		 * Complete the irq move. This cpu is going down and for
297 		 * non intr-remapping case, we can't wait till this interrupt
298 		 * arrives at this cpu before completing the irq move.
299 		 */
300 		irq_force_complete_move(irq);
301 
302 		if (cpumask_any_and(affinity, cpu_online_mask) >= nr_cpu_ids) {
303 			break_affinity = 1;
304 			affinity = cpu_online_mask;
305 		}
306 
307 		chip = irq_data_get_irq_chip(data);
308 		if (!irqd_can_move_in_process_context(data) && chip->irq_mask)
309 			chip->irq_mask(data);
310 
311 		if (chip->irq_set_affinity)
312 			chip->irq_set_affinity(data, affinity, true);
313 		else if (!(warned++))
314 			set_affinity = 0;
315 
316 		/*
317 		 * We unmask if the irq was not marked masked by the
318 		 * core code. That respects the lazy irq disable
319 		 * behaviour.
320 		 */
321 		if (!irqd_can_move_in_process_context(data) &&
322 		    !irqd_irq_masked(data) && chip->irq_unmask)
323 			chip->irq_unmask(data);
324 
325 		raw_spin_unlock(&desc->lock);
326 
327 		if (break_affinity && set_affinity)
328 			pr_notice("Broke affinity for irq %i\n", irq);
329 		else if (!set_affinity)
330 			pr_notice("Cannot set affinity for irq %i\n", irq);
331 	}
332 
333 	/*
334 	 * We can remove mdelay() and then send spuriuous interrupts to
335 	 * new cpu targets for all the irqs that were handled previously by
336 	 * this cpu. While it works, I have seen spurious interrupt messages
337 	 * (nothing wrong but still...).
338 	 *
339 	 * So for now, retain mdelay(1) and check the IRR and then send those
340 	 * interrupts to new targets as this cpu is already offlined...
341 	 */
342 	mdelay(1);
343 
344 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
345 		unsigned int irr;
346 
347 		if (__this_cpu_read(vector_irq[vector]) < 0)
348 			continue;
349 
350 		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
351 		if (irr  & (1 << (vector % 32))) {
352 			irq = __this_cpu_read(vector_irq[vector]);
353 
354 			desc = irq_to_desc(irq);
355 			data = irq_desc_get_irq_data(desc);
356 			chip = irq_data_get_irq_chip(data);
357 			raw_spin_lock(&desc->lock);
358 			if (chip->irq_retrigger)
359 				chip->irq_retrigger(data);
360 			raw_spin_unlock(&desc->lock);
361 		}
362 		__this_cpu_write(vector_irq[vector], -1);
363 	}
364 }
365 #endif
366