xref: /openbmc/linux/arch/x86/kernel/irq.c (revision 8fdff1dc)
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 atomic_t irq_err_count;
22 
23 /* Function pointer for generic interrupt vector handling */
24 void (*x86_platform_ipi_callback)(void) = NULL;
25 
26 /*
27  * 'what should we do if we get a hw irq event on an illegal vector'.
28  * each architecture has to answer this themselves.
29  */
30 void ack_bad_irq(unsigned int irq)
31 {
32 	if (printk_ratelimit())
33 		pr_err("unexpected IRQ trap at vector %02x\n", irq);
34 
35 	/*
36 	 * Currently unexpected vectors happen only on SMP and APIC.
37 	 * We _must_ ack these because every local APIC has only N
38 	 * irq slots per priority level, and a 'hanging, unacked' IRQ
39 	 * holds up an irq slot - in excessive cases (when multiple
40 	 * unexpected vectors occur) that might lock up the APIC
41 	 * completely.
42 	 * But only ack when the APIC is enabled -AK
43 	 */
44 	ack_APIC_irq();
45 }
46 
47 #define irq_stats(x)		(&per_cpu(irq_stat, x))
48 /*
49  * /proc/interrupts printing for arch specific interrupts
50  */
51 int arch_show_interrupts(struct seq_file *p, int prec)
52 {
53 	int j;
54 
55 	seq_printf(p, "%*s: ", prec, "NMI");
56 	for_each_online_cpu(j)
57 		seq_printf(p, "%10u ", irq_stats(j)->__nmi_count);
58 	seq_printf(p, "  Non-maskable interrupts\n");
59 #ifdef CONFIG_X86_LOCAL_APIC
60 	seq_printf(p, "%*s: ", prec, "LOC");
61 	for_each_online_cpu(j)
62 		seq_printf(p, "%10u ", irq_stats(j)->apic_timer_irqs);
63 	seq_printf(p, "  Local timer interrupts\n");
64 
65 	seq_printf(p, "%*s: ", prec, "SPU");
66 	for_each_online_cpu(j)
67 		seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count);
68 	seq_printf(p, "  Spurious interrupts\n");
69 	seq_printf(p, "%*s: ", prec, "PMI");
70 	for_each_online_cpu(j)
71 		seq_printf(p, "%10u ", irq_stats(j)->apic_perf_irqs);
72 	seq_printf(p, "  Performance monitoring interrupts\n");
73 	seq_printf(p, "%*s: ", prec, "IWI");
74 	for_each_online_cpu(j)
75 		seq_printf(p, "%10u ", irq_stats(j)->apic_irq_work_irqs);
76 	seq_printf(p, "  IRQ work interrupts\n");
77 	seq_printf(p, "%*s: ", prec, "RTR");
78 	for_each_online_cpu(j)
79 		seq_printf(p, "%10u ", irq_stats(j)->icr_read_retry_count);
80 	seq_printf(p, "  APIC ICR read retries\n");
81 #endif
82 	if (x86_platform_ipi_callback) {
83 		seq_printf(p, "%*s: ", prec, "PLT");
84 		for_each_online_cpu(j)
85 			seq_printf(p, "%10u ", irq_stats(j)->x86_platform_ipis);
86 		seq_printf(p, "  Platform interrupts\n");
87 	}
88 #ifdef CONFIG_SMP
89 	seq_printf(p, "%*s: ", prec, "RES");
90 	for_each_online_cpu(j)
91 		seq_printf(p, "%10u ", irq_stats(j)->irq_resched_count);
92 	seq_printf(p, "  Rescheduling interrupts\n");
93 	seq_printf(p, "%*s: ", prec, "CAL");
94 	for_each_online_cpu(j)
95 		seq_printf(p, "%10u ", irq_stats(j)->irq_call_count -
96 					irq_stats(j)->irq_tlb_count);
97 	seq_printf(p, "  Function call interrupts\n");
98 	seq_printf(p, "%*s: ", prec, "TLB");
99 	for_each_online_cpu(j)
100 		seq_printf(p, "%10u ", irq_stats(j)->irq_tlb_count);
101 	seq_printf(p, "  TLB shootdowns\n");
102 #endif
103 #ifdef CONFIG_X86_THERMAL_VECTOR
104 	seq_printf(p, "%*s: ", prec, "TRM");
105 	for_each_online_cpu(j)
106 		seq_printf(p, "%10u ", irq_stats(j)->irq_thermal_count);
107 	seq_printf(p, "  Thermal event interrupts\n");
108 #endif
109 #ifdef CONFIG_X86_MCE_THRESHOLD
110 	seq_printf(p, "%*s: ", prec, "THR");
111 	for_each_online_cpu(j)
112 		seq_printf(p, "%10u ", irq_stats(j)->irq_threshold_count);
113 	seq_printf(p, "  Threshold APIC interrupts\n");
114 #endif
115 #ifdef CONFIG_X86_MCE
116 	seq_printf(p, "%*s: ", prec, "MCE");
117 	for_each_online_cpu(j)
118 		seq_printf(p, "%10u ", per_cpu(mce_exception_count, j));
119 	seq_printf(p, "  Machine check exceptions\n");
120 	seq_printf(p, "%*s: ", prec, "MCP");
121 	for_each_online_cpu(j)
122 		seq_printf(p, "%10u ", per_cpu(mce_poll_count, j));
123 	seq_printf(p, "  Machine check polls\n");
124 #endif
125 	seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
126 #if defined(CONFIG_X86_IO_APIC)
127 	seq_printf(p, "%*s: %10u\n", prec, "MIS", atomic_read(&irq_mis_count));
128 #endif
129 	return 0;
130 }
131 
132 /*
133  * /proc/stat helpers
134  */
135 u64 arch_irq_stat_cpu(unsigned int cpu)
136 {
137 	u64 sum = irq_stats(cpu)->__nmi_count;
138 
139 #ifdef CONFIG_X86_LOCAL_APIC
140 	sum += irq_stats(cpu)->apic_timer_irqs;
141 	sum += irq_stats(cpu)->irq_spurious_count;
142 	sum += irq_stats(cpu)->apic_perf_irqs;
143 	sum += irq_stats(cpu)->apic_irq_work_irqs;
144 	sum += irq_stats(cpu)->icr_read_retry_count;
145 #endif
146 	if (x86_platform_ipi_callback)
147 		sum += irq_stats(cpu)->x86_platform_ipis;
148 #ifdef CONFIG_SMP
149 	sum += irq_stats(cpu)->irq_resched_count;
150 	sum += irq_stats(cpu)->irq_call_count;
151 #endif
152 #ifdef CONFIG_X86_THERMAL_VECTOR
153 	sum += irq_stats(cpu)->irq_thermal_count;
154 #endif
155 #ifdef CONFIG_X86_MCE_THRESHOLD
156 	sum += irq_stats(cpu)->irq_threshold_count;
157 #endif
158 #ifdef CONFIG_X86_MCE
159 	sum += per_cpu(mce_exception_count, cpu);
160 	sum += per_cpu(mce_poll_count, cpu);
161 #endif
162 	return sum;
163 }
164 
165 u64 arch_irq_stat(void)
166 {
167 	u64 sum = atomic_read(&irq_err_count);
168 
169 #ifdef CONFIG_X86_IO_APIC
170 	sum += atomic_read(&irq_mis_count);
171 #endif
172 	return sum;
173 }
174 
175 
176 /*
177  * do_IRQ handles all normal device IRQ's (the special
178  * SMP cross-CPU interrupts have their own specific
179  * handlers).
180  */
181 unsigned int __irq_entry do_IRQ(struct pt_regs *regs)
182 {
183 	struct pt_regs *old_regs = set_irq_regs(regs);
184 
185 	/* high bit used in ret_from_ code  */
186 	unsigned vector = ~regs->orig_ax;
187 	unsigned irq;
188 
189 	irq_enter();
190 	exit_idle();
191 
192 	irq = __this_cpu_read(vector_irq[vector]);
193 
194 	if (!handle_irq(irq, regs)) {
195 		ack_APIC_irq();
196 
197 		if (printk_ratelimit())
198 			pr_emerg("%s: %d.%d No irq handler for vector (irq %d)\n",
199 				__func__, smp_processor_id(), vector, irq);
200 	}
201 
202 	irq_exit();
203 
204 	set_irq_regs(old_regs);
205 	return 1;
206 }
207 
208 /*
209  * Handler for X86_PLATFORM_IPI_VECTOR.
210  */
211 void smp_x86_platform_ipi(struct pt_regs *regs)
212 {
213 	struct pt_regs *old_regs = set_irq_regs(regs);
214 
215 	ack_APIC_irq();
216 
217 	irq_enter();
218 
219 	exit_idle();
220 
221 	inc_irq_stat(x86_platform_ipis);
222 
223 	if (x86_platform_ipi_callback)
224 		x86_platform_ipi_callback();
225 
226 	irq_exit();
227 
228 	set_irq_regs(old_regs);
229 }
230 
231 EXPORT_SYMBOL_GPL(vector_used_by_percpu_irq);
232 
233 #ifdef CONFIG_HOTPLUG_CPU
234 /* A cpu has been removed from cpu_online_mask.  Reset irq affinities. */
235 void fixup_irqs(void)
236 {
237 	unsigned int irq, vector;
238 	static int warned;
239 	struct irq_desc *desc;
240 	struct irq_data *data;
241 	struct irq_chip *chip;
242 
243 	for_each_irq_desc(irq, desc) {
244 		int break_affinity = 0;
245 		int set_affinity = 1;
246 		const struct cpumask *affinity;
247 
248 		if (!desc)
249 			continue;
250 		if (irq == 2)
251 			continue;
252 
253 		/* interrupt's are disabled at this point */
254 		raw_spin_lock(&desc->lock);
255 
256 		data = irq_desc_get_irq_data(desc);
257 		affinity = data->affinity;
258 		if (!irq_has_action(irq) || irqd_is_per_cpu(data) ||
259 		    cpumask_subset(affinity, cpu_online_mask)) {
260 			raw_spin_unlock(&desc->lock);
261 			continue;
262 		}
263 
264 		/*
265 		 * Complete the irq move. This cpu is going down and for
266 		 * non intr-remapping case, we can't wait till this interrupt
267 		 * arrives at this cpu before completing the irq move.
268 		 */
269 		irq_force_complete_move(irq);
270 
271 		if (cpumask_any_and(affinity, cpu_online_mask) >= nr_cpu_ids) {
272 			break_affinity = 1;
273 			affinity = cpu_online_mask;
274 		}
275 
276 		chip = irq_data_get_irq_chip(data);
277 		if (!irqd_can_move_in_process_context(data) && chip->irq_mask)
278 			chip->irq_mask(data);
279 
280 		if (chip->irq_set_affinity)
281 			chip->irq_set_affinity(data, affinity, true);
282 		else if (!(warned++))
283 			set_affinity = 0;
284 
285 		/*
286 		 * We unmask if the irq was not marked masked by the
287 		 * core code. That respects the lazy irq disable
288 		 * behaviour.
289 		 */
290 		if (!irqd_can_move_in_process_context(data) &&
291 		    !irqd_irq_masked(data) && chip->irq_unmask)
292 			chip->irq_unmask(data);
293 
294 		raw_spin_unlock(&desc->lock);
295 
296 		if (break_affinity && set_affinity)
297 			pr_notice("Broke affinity for irq %i\n", irq);
298 		else if (!set_affinity)
299 			pr_notice("Cannot set affinity for irq %i\n", irq);
300 	}
301 
302 	/*
303 	 * We can remove mdelay() and then send spuriuous interrupts to
304 	 * new cpu targets for all the irqs that were handled previously by
305 	 * this cpu. While it works, I have seen spurious interrupt messages
306 	 * (nothing wrong but still...).
307 	 *
308 	 * So for now, retain mdelay(1) and check the IRR and then send those
309 	 * interrupts to new targets as this cpu is already offlined...
310 	 */
311 	mdelay(1);
312 
313 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
314 		unsigned int irr;
315 
316 		if (__this_cpu_read(vector_irq[vector]) < 0)
317 			continue;
318 
319 		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
320 		if (irr  & (1 << (vector % 32))) {
321 			irq = __this_cpu_read(vector_irq[vector]);
322 
323 			desc = irq_to_desc(irq);
324 			data = irq_desc_get_irq_data(desc);
325 			chip = irq_data_get_irq_chip(data);
326 			raw_spin_lock(&desc->lock);
327 			if (chip->irq_retrigger)
328 				chip->irq_retrigger(data);
329 			raw_spin_unlock(&desc->lock);
330 		}
331 		__this_cpu_write(vector_irq[vector], -1);
332 	}
333 }
334 #endif
335