xref: /openbmc/linux/arch/x86/kernel/irq.c (revision 36db6e8484ed455bbb320d89a119378897ae991c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Common interrupt code for 32 and 64 bit
4  */
5 #include <linux/cpu.h>
6 #include <linux/interrupt.h>
7 #include <linux/kernel_stat.h>
8 #include <linux/of.h>
9 #include <linux/seq_file.h>
10 #include <linux/smp.h>
11 #include <linux/ftrace.h>
12 #include <linux/delay.h>
13 #include <linux/export.h>
14 #include <linux/irq.h>
15 
16 #include <asm/irq_stack.h>
17 #include <asm/apic.h>
18 #include <asm/io_apic.h>
19 #include <asm/irq.h>
20 #include <asm/mce.h>
21 #include <asm/hw_irq.h>
22 #include <asm/desc.h>
23 #include <asm/traps.h>
24 #include <asm/thermal.h>
25 
26 #if defined(CONFIG_X86_LOCAL_APIC) || defined(CONFIG_X86_THERMAL_VECTOR)
27 #define CREATE_TRACE_POINTS
28 #include <asm/trace/irq_vectors.h>
29 #endif
30 
31 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
32 EXPORT_PER_CPU_SYMBOL(irq_stat);
33 
34 atomic_t irq_err_count;
35 
36 /*
37  * 'what should we do if we get a hw irq event on an illegal vector'.
38  * each architecture has to answer this themselves.
39  */
ack_bad_irq(unsigned int irq)40 void ack_bad_irq(unsigned int irq)
41 {
42 	if (printk_ratelimit())
43 		pr_err("unexpected IRQ trap at vector %02x\n", irq);
44 
45 	/*
46 	 * Currently unexpected vectors happen only on SMP and APIC.
47 	 * We _must_ ack these because every local APIC has only N
48 	 * irq slots per priority level, and a 'hanging, unacked' IRQ
49 	 * holds up an irq slot - in excessive cases (when multiple
50 	 * unexpected vectors occur) that might lock up the APIC
51 	 * completely.
52 	 * But only ack when the APIC is enabled -AK
53 	 */
54 	apic_eoi();
55 }
56 
57 #define irq_stats(x)		(&per_cpu(irq_stat, x))
58 /*
59  * /proc/interrupts printing for arch specific interrupts
60  */
arch_show_interrupts(struct seq_file * p,int prec)61 int arch_show_interrupts(struct seq_file *p, int prec)
62 {
63 	int j;
64 
65 	seq_printf(p, "%*s: ", prec, "NMI");
66 	for_each_online_cpu(j)
67 		seq_printf(p, "%10u ", irq_stats(j)->__nmi_count);
68 	seq_puts(p, "  Non-maskable interrupts\n");
69 #ifdef CONFIG_X86_LOCAL_APIC
70 	seq_printf(p, "%*s: ", prec, "LOC");
71 	for_each_online_cpu(j)
72 		seq_printf(p, "%10u ", irq_stats(j)->apic_timer_irqs);
73 	seq_puts(p, "  Local timer interrupts\n");
74 
75 	seq_printf(p, "%*s: ", prec, "SPU");
76 	for_each_online_cpu(j)
77 		seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count);
78 	seq_puts(p, "  Spurious interrupts\n");
79 	seq_printf(p, "%*s: ", prec, "PMI");
80 	for_each_online_cpu(j)
81 		seq_printf(p, "%10u ", irq_stats(j)->apic_perf_irqs);
82 	seq_puts(p, "  Performance monitoring interrupts\n");
83 	seq_printf(p, "%*s: ", prec, "IWI");
84 	for_each_online_cpu(j)
85 		seq_printf(p, "%10u ", irq_stats(j)->apic_irq_work_irqs);
86 	seq_puts(p, "  IRQ work interrupts\n");
87 	seq_printf(p, "%*s: ", prec, "RTR");
88 	for_each_online_cpu(j)
89 		seq_printf(p, "%10u ", irq_stats(j)->icr_read_retry_count);
90 	seq_puts(p, "  APIC ICR read retries\n");
91 	if (x86_platform_ipi_callback) {
92 		seq_printf(p, "%*s: ", prec, "PLT");
93 		for_each_online_cpu(j)
94 			seq_printf(p, "%10u ", irq_stats(j)->x86_platform_ipis);
95 		seq_puts(p, "  Platform interrupts\n");
96 	}
97 #endif
98 #ifdef CONFIG_SMP
99 	seq_printf(p, "%*s: ", prec, "RES");
100 	for_each_online_cpu(j)
101 		seq_printf(p, "%10u ", irq_stats(j)->irq_resched_count);
102 	seq_puts(p, "  Rescheduling interrupts\n");
103 	seq_printf(p, "%*s: ", prec, "CAL");
104 	for_each_online_cpu(j)
105 		seq_printf(p, "%10u ", irq_stats(j)->irq_call_count);
106 	seq_puts(p, "  Function call interrupts\n");
107 	seq_printf(p, "%*s: ", prec, "TLB");
108 	for_each_online_cpu(j)
109 		seq_printf(p, "%10u ", irq_stats(j)->irq_tlb_count);
110 	seq_puts(p, "  TLB shootdowns\n");
111 #endif
112 #ifdef CONFIG_X86_THERMAL_VECTOR
113 	seq_printf(p, "%*s: ", prec, "TRM");
114 	for_each_online_cpu(j)
115 		seq_printf(p, "%10u ", irq_stats(j)->irq_thermal_count);
116 	seq_puts(p, "  Thermal event interrupts\n");
117 #endif
118 #ifdef CONFIG_X86_MCE_THRESHOLD
119 	seq_printf(p, "%*s: ", prec, "THR");
120 	for_each_online_cpu(j)
121 		seq_printf(p, "%10u ", irq_stats(j)->irq_threshold_count);
122 	seq_puts(p, "  Threshold APIC interrupts\n");
123 #endif
124 #ifdef CONFIG_X86_MCE_AMD
125 	seq_printf(p, "%*s: ", prec, "DFR");
126 	for_each_online_cpu(j)
127 		seq_printf(p, "%10u ", irq_stats(j)->irq_deferred_error_count);
128 	seq_puts(p, "  Deferred Error APIC interrupts\n");
129 #endif
130 #ifdef CONFIG_X86_MCE
131 	seq_printf(p, "%*s: ", prec, "MCE");
132 	for_each_online_cpu(j)
133 		seq_printf(p, "%10u ", per_cpu(mce_exception_count, j));
134 	seq_puts(p, "  Machine check exceptions\n");
135 	seq_printf(p, "%*s: ", prec, "MCP");
136 	for_each_online_cpu(j)
137 		seq_printf(p, "%10u ", per_cpu(mce_poll_count, j));
138 	seq_puts(p, "  Machine check polls\n");
139 #endif
140 #ifdef CONFIG_X86_HV_CALLBACK_VECTOR
141 	if (test_bit(HYPERVISOR_CALLBACK_VECTOR, system_vectors)) {
142 		seq_printf(p, "%*s: ", prec, "HYP");
143 		for_each_online_cpu(j)
144 			seq_printf(p, "%10u ",
145 				   irq_stats(j)->irq_hv_callback_count);
146 		seq_puts(p, "  Hypervisor callback interrupts\n");
147 	}
148 #endif
149 #if IS_ENABLED(CONFIG_HYPERV)
150 	if (test_bit(HYPERV_REENLIGHTENMENT_VECTOR, system_vectors)) {
151 		seq_printf(p, "%*s: ", prec, "HRE");
152 		for_each_online_cpu(j)
153 			seq_printf(p, "%10u ",
154 				   irq_stats(j)->irq_hv_reenlightenment_count);
155 		seq_puts(p, "  Hyper-V reenlightenment interrupts\n");
156 	}
157 	if (test_bit(HYPERV_STIMER0_VECTOR, system_vectors)) {
158 		seq_printf(p, "%*s: ", prec, "HVS");
159 		for_each_online_cpu(j)
160 			seq_printf(p, "%10u ",
161 				   irq_stats(j)->hyperv_stimer0_count);
162 		seq_puts(p, "  Hyper-V stimer0 interrupts\n");
163 	}
164 #endif
165 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
166 #if defined(CONFIG_X86_IO_APIC)
167 	seq_printf(p, "%*s: %10u\n", prec, "MIS", atomic_read(&irq_mis_count));
168 #endif
169 #ifdef CONFIG_HAVE_KVM
170 	seq_printf(p, "%*s: ", prec, "PIN");
171 	for_each_online_cpu(j)
172 		seq_printf(p, "%10u ", irq_stats(j)->kvm_posted_intr_ipis);
173 	seq_puts(p, "  Posted-interrupt notification event\n");
174 
175 	seq_printf(p, "%*s: ", prec, "NPI");
176 	for_each_online_cpu(j)
177 		seq_printf(p, "%10u ",
178 			   irq_stats(j)->kvm_posted_intr_nested_ipis);
179 	seq_puts(p, "  Nested posted-interrupt event\n");
180 
181 	seq_printf(p, "%*s: ", prec, "PIW");
182 	for_each_online_cpu(j)
183 		seq_printf(p, "%10u ",
184 			   irq_stats(j)->kvm_posted_intr_wakeup_ipis);
185 	seq_puts(p, "  Posted-interrupt wakeup event\n");
186 #endif
187 	return 0;
188 }
189 
190 /*
191  * /proc/stat helpers
192  */
arch_irq_stat_cpu(unsigned int cpu)193 u64 arch_irq_stat_cpu(unsigned int cpu)
194 {
195 	u64 sum = irq_stats(cpu)->__nmi_count;
196 
197 #ifdef CONFIG_X86_LOCAL_APIC
198 	sum += irq_stats(cpu)->apic_timer_irqs;
199 	sum += irq_stats(cpu)->irq_spurious_count;
200 	sum += irq_stats(cpu)->apic_perf_irqs;
201 	sum += irq_stats(cpu)->apic_irq_work_irqs;
202 	sum += irq_stats(cpu)->icr_read_retry_count;
203 	if (x86_platform_ipi_callback)
204 		sum += irq_stats(cpu)->x86_platform_ipis;
205 #endif
206 #ifdef CONFIG_SMP
207 	sum += irq_stats(cpu)->irq_resched_count;
208 	sum += irq_stats(cpu)->irq_call_count;
209 #endif
210 #ifdef CONFIG_X86_THERMAL_VECTOR
211 	sum += irq_stats(cpu)->irq_thermal_count;
212 #endif
213 #ifdef CONFIG_X86_MCE_THRESHOLD
214 	sum += irq_stats(cpu)->irq_threshold_count;
215 #endif
216 #ifdef CONFIG_X86_HV_CALLBACK_VECTOR
217 	sum += irq_stats(cpu)->irq_hv_callback_count;
218 #endif
219 #if IS_ENABLED(CONFIG_HYPERV)
220 	sum += irq_stats(cpu)->irq_hv_reenlightenment_count;
221 	sum += irq_stats(cpu)->hyperv_stimer0_count;
222 #endif
223 #ifdef CONFIG_X86_MCE
224 	sum += per_cpu(mce_exception_count, cpu);
225 	sum += per_cpu(mce_poll_count, cpu);
226 #endif
227 	return sum;
228 }
229 
arch_irq_stat(void)230 u64 arch_irq_stat(void)
231 {
232 	u64 sum = atomic_read(&irq_err_count);
233 	return sum;
234 }
235 
handle_irq(struct irq_desc * desc,struct pt_regs * regs)236 static __always_inline void handle_irq(struct irq_desc *desc,
237 				       struct pt_regs *regs)
238 {
239 	if (IS_ENABLED(CONFIG_X86_64))
240 		generic_handle_irq_desc(desc);
241 	else
242 		__handle_irq(desc, regs);
243 }
244 
245 /*
246  * common_interrupt() handles all normal device IRQ's (the special SMP
247  * cross-CPU interrupts have their own entry points).
248  */
DEFINE_IDTENTRY_IRQ(common_interrupt)249 DEFINE_IDTENTRY_IRQ(common_interrupt)
250 {
251 	struct pt_regs *old_regs = set_irq_regs(regs);
252 	struct irq_desc *desc;
253 
254 	/* entry code tells RCU that we're not quiescent.  Check it. */
255 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "IRQ failed to wake up RCU");
256 
257 	desc = __this_cpu_read(vector_irq[vector]);
258 	if (likely(!IS_ERR_OR_NULL(desc))) {
259 		handle_irq(desc, regs);
260 	} else {
261 		apic_eoi();
262 
263 		if (desc == VECTOR_UNUSED) {
264 			pr_emerg_ratelimited("%s: %d.%u No irq handler for vector\n",
265 					     __func__, smp_processor_id(),
266 					     vector);
267 		} else {
268 			__this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
269 		}
270 	}
271 
272 	set_irq_regs(old_regs);
273 }
274 
275 #ifdef CONFIG_X86_LOCAL_APIC
276 /* Function pointer for generic interrupt vector handling */
277 void (*x86_platform_ipi_callback)(void) = NULL;
278 /*
279  * Handler for X86_PLATFORM_IPI_VECTOR.
280  */
DEFINE_IDTENTRY_SYSVEC(sysvec_x86_platform_ipi)281 DEFINE_IDTENTRY_SYSVEC(sysvec_x86_platform_ipi)
282 {
283 	struct pt_regs *old_regs = set_irq_regs(regs);
284 
285 	apic_eoi();
286 	trace_x86_platform_ipi_entry(X86_PLATFORM_IPI_VECTOR);
287 	inc_irq_stat(x86_platform_ipis);
288 	if (x86_platform_ipi_callback)
289 		x86_platform_ipi_callback();
290 	trace_x86_platform_ipi_exit(X86_PLATFORM_IPI_VECTOR);
291 	set_irq_regs(old_regs);
292 }
293 #endif
294 
295 #ifdef CONFIG_HAVE_KVM
dummy_handler(void)296 static void dummy_handler(void) {}
297 static void (*kvm_posted_intr_wakeup_handler)(void) = dummy_handler;
298 
kvm_set_posted_intr_wakeup_handler(void (* handler)(void))299 void kvm_set_posted_intr_wakeup_handler(void (*handler)(void))
300 {
301 	if (handler)
302 		kvm_posted_intr_wakeup_handler = handler;
303 	else {
304 		kvm_posted_intr_wakeup_handler = dummy_handler;
305 		synchronize_rcu();
306 	}
307 }
308 EXPORT_SYMBOL_GPL(kvm_set_posted_intr_wakeup_handler);
309 
310 /*
311  * Handler for POSTED_INTERRUPT_VECTOR.
312  */
DEFINE_IDTENTRY_SYSVEC_SIMPLE(sysvec_kvm_posted_intr_ipi)313 DEFINE_IDTENTRY_SYSVEC_SIMPLE(sysvec_kvm_posted_intr_ipi)
314 {
315 	apic_eoi();
316 	inc_irq_stat(kvm_posted_intr_ipis);
317 }
318 
319 /*
320  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
321  */
DEFINE_IDTENTRY_SYSVEC(sysvec_kvm_posted_intr_wakeup_ipi)322 DEFINE_IDTENTRY_SYSVEC(sysvec_kvm_posted_intr_wakeup_ipi)
323 {
324 	apic_eoi();
325 	inc_irq_stat(kvm_posted_intr_wakeup_ipis);
326 	kvm_posted_intr_wakeup_handler();
327 }
328 
329 /*
330  * Handler for POSTED_INTERRUPT_NESTED_VECTOR.
331  */
DEFINE_IDTENTRY_SYSVEC_SIMPLE(sysvec_kvm_posted_intr_nested_ipi)332 DEFINE_IDTENTRY_SYSVEC_SIMPLE(sysvec_kvm_posted_intr_nested_ipi)
333 {
334 	apic_eoi();
335 	inc_irq_stat(kvm_posted_intr_nested_ipis);
336 }
337 #endif
338 
339 
340 #ifdef CONFIG_HOTPLUG_CPU
341 /* A cpu has been removed from cpu_online_mask.  Reset irq affinities. */
fixup_irqs(void)342 void fixup_irqs(void)
343 {
344 	unsigned int irr, vector;
345 	struct irq_desc *desc;
346 	struct irq_data *data;
347 	struct irq_chip *chip;
348 
349 	irq_migrate_all_off_this_cpu();
350 
351 	/*
352 	 * We can remove mdelay() and then send spurious interrupts to
353 	 * new cpu targets for all the irqs that were handled previously by
354 	 * this cpu. While it works, I have seen spurious interrupt messages
355 	 * (nothing wrong but still...).
356 	 *
357 	 * So for now, retain mdelay(1) and check the IRR and then send those
358 	 * interrupts to new targets as this cpu is already offlined...
359 	 */
360 	mdelay(1);
361 
362 	/*
363 	 * We can walk the vector array of this cpu without holding
364 	 * vector_lock because the cpu is already marked !online, so
365 	 * nothing else will touch it.
366 	 */
367 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
368 		if (IS_ERR_OR_NULL(__this_cpu_read(vector_irq[vector])))
369 			continue;
370 
371 		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
372 		if (irr  & (1 << (vector % 32))) {
373 			desc = __this_cpu_read(vector_irq[vector]);
374 
375 			raw_spin_lock(&desc->lock);
376 			data = irq_desc_get_irq_data(desc);
377 			chip = irq_data_get_irq_chip(data);
378 			if (chip->irq_retrigger) {
379 				chip->irq_retrigger(data);
380 				__this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
381 			}
382 			raw_spin_unlock(&desc->lock);
383 		}
384 		if (__this_cpu_read(vector_irq[vector]) != VECTOR_RETRIGGERED)
385 			__this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
386 	}
387 }
388 #endif
389 
390 #ifdef CONFIG_X86_THERMAL_VECTOR
smp_thermal_vector(void)391 static void smp_thermal_vector(void)
392 {
393 	if (x86_thermal_enabled())
394 		intel_thermal_interrupt();
395 	else
396 		pr_err("CPU%d: Unexpected LVT thermal interrupt!\n",
397 		       smp_processor_id());
398 }
399 
DEFINE_IDTENTRY_SYSVEC(sysvec_thermal)400 DEFINE_IDTENTRY_SYSVEC(sysvec_thermal)
401 {
402 	trace_thermal_apic_entry(THERMAL_APIC_VECTOR);
403 	inc_irq_stat(irq_thermal_count);
404 	smp_thermal_vector();
405 	trace_thermal_apic_exit(THERMAL_APIC_VECTOR);
406 	apic_eoi();
407 }
408 #endif
409