xref: /openbmc/linux/arch/x86/kernel/irq.c (revision c127f98ba9aba1818a6ca3a1da5a24653a10d966)
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/mce.h>
18 #include <asm/hw_irq.h>
19 #include <asm/desc.h>
20 
21 #define CREATE_TRACE_POINTS
22 #include <asm/trace/irq_vectors.h>
23 
24 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
25 EXPORT_PER_CPU_SYMBOL(irq_stat);
26 
27 DEFINE_PER_CPU(struct pt_regs *, irq_regs);
28 EXPORT_PER_CPU_SYMBOL(irq_regs);
29 
30 atomic_t irq_err_count;
31 
32 /*
33  * 'what should we do if we get a hw irq event on an illegal vector'.
34  * each architecture has to answer this themselves.
35  */
36 void ack_bad_irq(unsigned int irq)
37 {
38 	if (printk_ratelimit())
39 		pr_err("unexpected IRQ trap at vector %02x\n", irq);
40 
41 	/*
42 	 * Currently unexpected vectors happen only on SMP and APIC.
43 	 * We _must_ ack these because every local APIC has only N
44 	 * irq slots per priority level, and a 'hanging, unacked' IRQ
45 	 * holds up an irq slot - in excessive cases (when multiple
46 	 * unexpected vectors occur) that might lock up the APIC
47 	 * completely.
48 	 * But only ack when the APIC is enabled -AK
49 	 */
50 	ack_APIC_irq();
51 }
52 
53 #define irq_stats(x)		(&per_cpu(irq_stat, x))
54 /*
55  * /proc/interrupts printing for arch specific interrupts
56  */
57 int arch_show_interrupts(struct seq_file *p, int prec)
58 {
59 	int j;
60 
61 	seq_printf(p, "%*s: ", prec, "NMI");
62 	for_each_online_cpu(j)
63 		seq_printf(p, "%10u ", irq_stats(j)->__nmi_count);
64 	seq_puts(p, "  Non-maskable interrupts\n");
65 #ifdef CONFIG_X86_LOCAL_APIC
66 	seq_printf(p, "%*s: ", prec, "LOC");
67 	for_each_online_cpu(j)
68 		seq_printf(p, "%10u ", irq_stats(j)->apic_timer_irqs);
69 	seq_puts(p, "  Local timer interrupts\n");
70 
71 	seq_printf(p, "%*s: ", prec, "SPU");
72 	for_each_online_cpu(j)
73 		seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count);
74 	seq_puts(p, "  Spurious interrupts\n");
75 	seq_printf(p, "%*s: ", prec, "PMI");
76 	for_each_online_cpu(j)
77 		seq_printf(p, "%10u ", irq_stats(j)->apic_perf_irqs);
78 	seq_puts(p, "  Performance monitoring interrupts\n");
79 	seq_printf(p, "%*s: ", prec, "IWI");
80 	for_each_online_cpu(j)
81 		seq_printf(p, "%10u ", irq_stats(j)->apic_irq_work_irqs);
82 	seq_puts(p, "  IRQ work interrupts\n");
83 	seq_printf(p, "%*s: ", prec, "RTR");
84 	for_each_online_cpu(j)
85 		seq_printf(p, "%10u ", irq_stats(j)->icr_read_retry_count);
86 	seq_puts(p, "  APIC ICR read retries\n");
87 	if (x86_platform_ipi_callback) {
88 		seq_printf(p, "%*s: ", prec, "PLT");
89 		for_each_online_cpu(j)
90 			seq_printf(p, "%10u ", irq_stats(j)->x86_platform_ipis);
91 		seq_puts(p, "  Platform interrupts\n");
92 	}
93 #endif
94 #ifdef CONFIG_SMP
95 	seq_printf(p, "%*s: ", prec, "RES");
96 	for_each_online_cpu(j)
97 		seq_printf(p, "%10u ", irq_stats(j)->irq_resched_count);
98 	seq_puts(p, "  Rescheduling interrupts\n");
99 	seq_printf(p, "%*s: ", prec, "CAL");
100 	for_each_online_cpu(j)
101 		seq_printf(p, "%10u ", irq_stats(j)->irq_call_count);
102 	seq_puts(p, "  Function call interrupts\n");
103 	seq_printf(p, "%*s: ", prec, "TLB");
104 	for_each_online_cpu(j)
105 		seq_printf(p, "%10u ", irq_stats(j)->irq_tlb_count);
106 	seq_puts(p, "  TLB shootdowns\n");
107 #endif
108 #ifdef CONFIG_X86_THERMAL_VECTOR
109 	seq_printf(p, "%*s: ", prec, "TRM");
110 	for_each_online_cpu(j)
111 		seq_printf(p, "%10u ", irq_stats(j)->irq_thermal_count);
112 	seq_puts(p, "  Thermal event interrupts\n");
113 #endif
114 #ifdef CONFIG_X86_MCE_THRESHOLD
115 	seq_printf(p, "%*s: ", prec, "THR");
116 	for_each_online_cpu(j)
117 		seq_printf(p, "%10u ", irq_stats(j)->irq_threshold_count);
118 	seq_puts(p, "  Threshold APIC interrupts\n");
119 #endif
120 #ifdef CONFIG_X86_MCE_AMD
121 	seq_printf(p, "%*s: ", prec, "DFR");
122 	for_each_online_cpu(j)
123 		seq_printf(p, "%10u ", irq_stats(j)->irq_deferred_error_count);
124 	seq_puts(p, "  Deferred Error APIC interrupts\n");
125 #endif
126 #ifdef CONFIG_X86_MCE
127 	seq_printf(p, "%*s: ", prec, "MCE");
128 	for_each_online_cpu(j)
129 		seq_printf(p, "%10u ", per_cpu(mce_exception_count, j));
130 	seq_puts(p, "  Machine check exceptions\n");
131 	seq_printf(p, "%*s: ", prec, "MCP");
132 	for_each_online_cpu(j)
133 		seq_printf(p, "%10u ", per_cpu(mce_poll_count, j));
134 	seq_puts(p, "  Machine check polls\n");
135 #endif
136 #if IS_ENABLED(CONFIG_HYPERV) || defined(CONFIG_XEN)
137 	if (test_bit(HYPERVISOR_CALLBACK_VECTOR, system_vectors)) {
138 		seq_printf(p, "%*s: ", prec, "HYP");
139 		for_each_online_cpu(j)
140 			seq_printf(p, "%10u ",
141 				   irq_stats(j)->irq_hv_callback_count);
142 		seq_puts(p, "  Hypervisor callback interrupts\n");
143 	}
144 #endif
145 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
146 #if defined(CONFIG_X86_IO_APIC)
147 	seq_printf(p, "%*s: %10u\n", prec, "MIS", atomic_read(&irq_mis_count));
148 #endif
149 #ifdef CONFIG_HAVE_KVM
150 	seq_printf(p, "%*s: ", prec, "PIN");
151 	for_each_online_cpu(j)
152 		seq_printf(p, "%10u ", irq_stats(j)->kvm_posted_intr_ipis);
153 	seq_puts(p, "  Posted-interrupt notification event\n");
154 
155 	seq_printf(p, "%*s: ", prec, "NPI");
156 	for_each_online_cpu(j)
157 		seq_printf(p, "%10u ",
158 			   irq_stats(j)->kvm_posted_intr_nested_ipis);
159 	seq_puts(p, "  Nested posted-interrupt event\n");
160 
161 	seq_printf(p, "%*s: ", prec, "PIW");
162 	for_each_online_cpu(j)
163 		seq_printf(p, "%10u ",
164 			   irq_stats(j)->kvm_posted_intr_wakeup_ipis);
165 	seq_puts(p, "  Posted-interrupt wakeup event\n");
166 #endif
167 	return 0;
168 }
169 
170 /*
171  * /proc/stat helpers
172  */
173 u64 arch_irq_stat_cpu(unsigned int cpu)
174 {
175 	u64 sum = irq_stats(cpu)->__nmi_count;
176 
177 #ifdef CONFIG_X86_LOCAL_APIC
178 	sum += irq_stats(cpu)->apic_timer_irqs;
179 	sum += irq_stats(cpu)->irq_spurious_count;
180 	sum += irq_stats(cpu)->apic_perf_irqs;
181 	sum += irq_stats(cpu)->apic_irq_work_irqs;
182 	sum += irq_stats(cpu)->icr_read_retry_count;
183 	if (x86_platform_ipi_callback)
184 		sum += irq_stats(cpu)->x86_platform_ipis;
185 #endif
186 #ifdef CONFIG_SMP
187 	sum += irq_stats(cpu)->irq_resched_count;
188 	sum += irq_stats(cpu)->irq_call_count;
189 #endif
190 #ifdef CONFIG_X86_THERMAL_VECTOR
191 	sum += irq_stats(cpu)->irq_thermal_count;
192 #endif
193 #ifdef CONFIG_X86_MCE_THRESHOLD
194 	sum += irq_stats(cpu)->irq_threshold_count;
195 #endif
196 #ifdef CONFIG_X86_MCE
197 	sum += per_cpu(mce_exception_count, cpu);
198 	sum += per_cpu(mce_poll_count, cpu);
199 #endif
200 	return sum;
201 }
202 
203 u64 arch_irq_stat(void)
204 {
205 	u64 sum = atomic_read(&irq_err_count);
206 	return sum;
207 }
208 
209 
210 /*
211  * do_IRQ handles all normal device IRQ's (the special
212  * SMP cross-CPU interrupts have their own specific
213  * handlers).
214  */
215 __visible unsigned int __irq_entry do_IRQ(struct pt_regs *regs)
216 {
217 	struct pt_regs *old_regs = set_irq_regs(regs);
218 	struct irq_desc * desc;
219 	/* high bit used in ret_from_ code  */
220 	unsigned vector = ~regs->orig_ax;
221 
222 	entering_irq();
223 
224 	/* entering_irq() tells RCU that we're not quiescent.  Check it. */
225 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "IRQ failed to wake up RCU");
226 
227 	desc = __this_cpu_read(vector_irq[vector]);
228 
229 	if (!handle_irq(desc, regs)) {
230 		ack_APIC_irq();
231 
232 		if (desc != VECTOR_RETRIGGERED) {
233 			pr_emerg_ratelimited("%s: %d.%d No irq handler for vector\n",
234 					     __func__, smp_processor_id(),
235 					     vector);
236 		} else {
237 			__this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
238 		}
239 	}
240 
241 	exiting_irq();
242 
243 	set_irq_regs(old_regs);
244 	return 1;
245 }
246 
247 #ifdef CONFIG_X86_LOCAL_APIC
248 /* Function pointer for generic interrupt vector handling */
249 void (*x86_platform_ipi_callback)(void) = NULL;
250 /*
251  * Handler for X86_PLATFORM_IPI_VECTOR.
252  */
253 __visible void __irq_entry smp_x86_platform_ipi(struct pt_regs *regs)
254 {
255 	struct pt_regs *old_regs = set_irq_regs(regs);
256 
257 	entering_ack_irq();
258 	trace_x86_platform_ipi_entry(X86_PLATFORM_IPI_VECTOR);
259 	inc_irq_stat(x86_platform_ipis);
260 	if (x86_platform_ipi_callback)
261 		x86_platform_ipi_callback();
262 	trace_x86_platform_ipi_exit(X86_PLATFORM_IPI_VECTOR);
263 	exiting_irq();
264 	set_irq_regs(old_regs);
265 }
266 #endif
267 
268 #ifdef CONFIG_HAVE_KVM
269 static void dummy_handler(void) {}
270 static void (*kvm_posted_intr_wakeup_handler)(void) = dummy_handler;
271 
272 void kvm_set_posted_intr_wakeup_handler(void (*handler)(void))
273 {
274 	if (handler)
275 		kvm_posted_intr_wakeup_handler = handler;
276 	else
277 		kvm_posted_intr_wakeup_handler = dummy_handler;
278 }
279 EXPORT_SYMBOL_GPL(kvm_set_posted_intr_wakeup_handler);
280 
281 /*
282  * Handler for POSTED_INTERRUPT_VECTOR.
283  */
284 __visible void smp_kvm_posted_intr_ipi(struct pt_regs *regs)
285 {
286 	struct pt_regs *old_regs = set_irq_regs(regs);
287 
288 	entering_ack_irq();
289 	inc_irq_stat(kvm_posted_intr_ipis);
290 	exiting_irq();
291 	set_irq_regs(old_regs);
292 }
293 
294 /*
295  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
296  */
297 __visible void smp_kvm_posted_intr_wakeup_ipi(struct pt_regs *regs)
298 {
299 	struct pt_regs *old_regs = set_irq_regs(regs);
300 
301 	entering_ack_irq();
302 	inc_irq_stat(kvm_posted_intr_wakeup_ipis);
303 	kvm_posted_intr_wakeup_handler();
304 	exiting_irq();
305 	set_irq_regs(old_regs);
306 }
307 
308 /*
309  * Handler for POSTED_INTERRUPT_NESTED_VECTOR.
310  */
311 __visible void smp_kvm_posted_intr_nested_ipi(struct pt_regs *regs)
312 {
313 	struct pt_regs *old_regs = set_irq_regs(regs);
314 
315 	entering_ack_irq();
316 	inc_irq_stat(kvm_posted_intr_nested_ipis);
317 	exiting_irq();
318 	set_irq_regs(old_regs);
319 }
320 #endif
321 
322 
323 #ifdef CONFIG_HOTPLUG_CPU
324 /* A cpu has been removed from cpu_online_mask.  Reset irq affinities. */
325 void fixup_irqs(void)
326 {
327 	unsigned int irr, vector;
328 	struct irq_desc *desc;
329 	struct irq_data *data;
330 	struct irq_chip *chip;
331 
332 	irq_migrate_all_off_this_cpu();
333 
334 	/*
335 	 * We can remove mdelay() and then send spuriuous interrupts to
336 	 * new cpu targets for all the irqs that were handled previously by
337 	 * this cpu. While it works, I have seen spurious interrupt messages
338 	 * (nothing wrong but still...).
339 	 *
340 	 * So for now, retain mdelay(1) and check the IRR and then send those
341 	 * interrupts to new targets as this cpu is already offlined...
342 	 */
343 	mdelay(1);
344 
345 	/*
346 	 * We can walk the vector array of this cpu without holding
347 	 * vector_lock because the cpu is already marked !online, so
348 	 * nothing else will touch it.
349 	 */
350 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
351 		if (IS_ERR_OR_NULL(__this_cpu_read(vector_irq[vector])))
352 			continue;
353 
354 		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
355 		if (irr  & (1 << (vector % 32))) {
356 			desc = __this_cpu_read(vector_irq[vector]);
357 
358 			raw_spin_lock(&desc->lock);
359 			data = irq_desc_get_irq_data(desc);
360 			chip = irq_data_get_irq_chip(data);
361 			if (chip->irq_retrigger) {
362 				chip->irq_retrigger(data);
363 				__this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
364 			}
365 			raw_spin_unlock(&desc->lock);
366 		}
367 		if (__this_cpu_read(vector_irq[vector]) != VECTOR_RETRIGGERED)
368 			__this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
369 	}
370 }
371 #endif
372