Loading...
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#include <asm/desc.h>
21
22#define CREATE_TRACE_POINTS
23#include <asm/trace/irq_vectors.h>
24
25atomic_t irq_err_count;
26
27/* Function pointer for generic interrupt vector handling */
28void (*x86_platform_ipi_callback)(void) = NULL;
29
30/*
31 * 'what should we do if we get a hw irq event on an illegal vector'.
32 * each architecture has to answer this themselves.
33 */
34void ack_bad_irq(unsigned int irq)
35{
36 if (printk_ratelimit())
37 pr_err("unexpected IRQ trap at vector %02x\n", irq);
38
39 /*
40 * Currently unexpected vectors happen only on SMP and APIC.
41 * We _must_ ack these because every local APIC has only N
42 * irq slots per priority level, and a 'hanging, unacked' IRQ
43 * holds up an irq slot - in excessive cases (when multiple
44 * unexpected vectors occur) that might lock up the APIC
45 * completely.
46 * But only ack when the APIC is enabled -AK
47 */
48 ack_APIC_irq();
49}
50
51#define irq_stats(x) (&per_cpu(irq_stat, x))
52/*
53 * /proc/interrupts printing for arch specific interrupts
54 */
55int arch_show_interrupts(struct seq_file *p, int prec)
56{
57 int j;
58
59 seq_printf(p, "%*s: ", prec, "NMI");
60 for_each_online_cpu(j)
61 seq_printf(p, "%10u ", irq_stats(j)->__nmi_count);
62 seq_printf(p, " Non-maskable interrupts\n");
63#ifdef CONFIG_X86_LOCAL_APIC
64 seq_printf(p, "%*s: ", prec, "LOC");
65 for_each_online_cpu(j)
66 seq_printf(p, "%10u ", irq_stats(j)->apic_timer_irqs);
67 seq_printf(p, " Local timer interrupts\n");
68
69 seq_printf(p, "%*s: ", prec, "SPU");
70 for_each_online_cpu(j)
71 seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count);
72 seq_printf(p, " Spurious interrupts\n");
73 seq_printf(p, "%*s: ", prec, "PMI");
74 for_each_online_cpu(j)
75 seq_printf(p, "%10u ", irq_stats(j)->apic_perf_irqs);
76 seq_printf(p, " Performance monitoring interrupts\n");
77 seq_printf(p, "%*s: ", prec, "IWI");
78 for_each_online_cpu(j)
79 seq_printf(p, "%10u ", irq_stats(j)->apic_irq_work_irqs);
80 seq_printf(p, " IRQ work interrupts\n");
81 seq_printf(p, "%*s: ", prec, "RTR");
82 for_each_online_cpu(j)
83 seq_printf(p, "%10u ", irq_stats(j)->icr_read_retry_count);
84 seq_printf(p, " APIC ICR read retries\n");
85#endif
86 if (x86_platform_ipi_callback) {
87 seq_printf(p, "%*s: ", prec, "PLT");
88 for_each_online_cpu(j)
89 seq_printf(p, "%10u ", irq_stats(j)->x86_platform_ipis);
90 seq_printf(p, " Platform interrupts\n");
91 }
92#ifdef CONFIG_SMP
93 seq_printf(p, "%*s: ", prec, "RES");
94 for_each_online_cpu(j)
95 seq_printf(p, "%10u ", irq_stats(j)->irq_resched_count);
96 seq_printf(p, " Rescheduling interrupts\n");
97 seq_printf(p, "%*s: ", prec, "CAL");
98 for_each_online_cpu(j)
99 seq_printf(p, "%10u ", irq_stats(j)->irq_call_count -
100 irq_stats(j)->irq_tlb_count);
101 seq_printf(p, " Function call interrupts\n");
102 seq_printf(p, "%*s: ", prec, "TLB");
103 for_each_online_cpu(j)
104 seq_printf(p, "%10u ", irq_stats(j)->irq_tlb_count);
105 seq_printf(p, " TLB shootdowns\n");
106#endif
107#ifdef CONFIG_X86_THERMAL_VECTOR
108 seq_printf(p, "%*s: ", prec, "TRM");
109 for_each_online_cpu(j)
110 seq_printf(p, "%10u ", irq_stats(j)->irq_thermal_count);
111 seq_printf(p, " Thermal event interrupts\n");
112#endif
113#ifdef CONFIG_X86_MCE_THRESHOLD
114 seq_printf(p, "%*s: ", prec, "THR");
115 for_each_online_cpu(j)
116 seq_printf(p, "%10u ", irq_stats(j)->irq_threshold_count);
117 seq_printf(p, " Threshold APIC interrupts\n");
118#endif
119#ifdef CONFIG_X86_MCE
120 seq_printf(p, "%*s: ", prec, "MCE");
121 for_each_online_cpu(j)
122 seq_printf(p, "%10u ", per_cpu(mce_exception_count, j));
123 seq_printf(p, " Machine check exceptions\n");
124 seq_printf(p, "%*s: ", prec, "MCP");
125 for_each_online_cpu(j)
126 seq_printf(p, "%10u ", per_cpu(mce_poll_count, j));
127 seq_printf(p, " Machine check polls\n");
128#endif
129#if IS_ENABLED(CONFIG_HYPERV) || defined(CONFIG_XEN)
130 seq_printf(p, "%*s: ", prec, "THR");
131 for_each_online_cpu(j)
132 seq_printf(p, "%10u ", irq_stats(j)->irq_hv_callback_count);
133 seq_printf(p, " Hypervisor callback interrupts\n");
134#endif
135 seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
136#if defined(CONFIG_X86_IO_APIC)
137 seq_printf(p, "%*s: %10u\n", prec, "MIS", atomic_read(&irq_mis_count));
138#endif
139 return 0;
140}
141
142/*
143 * /proc/stat helpers
144 */
145u64 arch_irq_stat_cpu(unsigned int cpu)
146{
147 u64 sum = irq_stats(cpu)->__nmi_count;
148
149#ifdef CONFIG_X86_LOCAL_APIC
150 sum += irq_stats(cpu)->apic_timer_irqs;
151 sum += irq_stats(cpu)->irq_spurious_count;
152 sum += irq_stats(cpu)->apic_perf_irqs;
153 sum += irq_stats(cpu)->apic_irq_work_irqs;
154 sum += irq_stats(cpu)->icr_read_retry_count;
155#endif
156 if (x86_platform_ipi_callback)
157 sum += irq_stats(cpu)->x86_platform_ipis;
158#ifdef CONFIG_SMP
159 sum += irq_stats(cpu)->irq_resched_count;
160 sum += irq_stats(cpu)->irq_call_count;
161#endif
162#ifdef CONFIG_X86_THERMAL_VECTOR
163 sum += irq_stats(cpu)->irq_thermal_count;
164#endif
165#ifdef CONFIG_X86_MCE_THRESHOLD
166 sum += irq_stats(cpu)->irq_threshold_count;
167#endif
168#ifdef CONFIG_X86_MCE
169 sum += per_cpu(mce_exception_count, cpu);
170 sum += per_cpu(mce_poll_count, cpu);
171#endif
172 return sum;
173}
174
175u64 arch_irq_stat(void)
176{
177 u64 sum = atomic_read(&irq_err_count);
178 return sum;
179}
180
181
182/*
183 * do_IRQ handles all normal device IRQ's (the special
184 * SMP cross-CPU interrupts have their own specific
185 * handlers).
186 */
187__visible unsigned int __irq_entry do_IRQ(struct pt_regs *regs)
188{
189 struct pt_regs *old_regs = set_irq_regs(regs);
190
191 /* high bit used in ret_from_ code */
192 unsigned vector = ~regs->orig_ax;
193 unsigned irq;
194
195 irq_enter();
196 exit_idle();
197
198 irq = __this_cpu_read(vector_irq[vector]);
199
200 if (!handle_irq(irq, regs)) {
201 ack_APIC_irq();
202
203 if (irq != VECTOR_RETRIGGERED) {
204 pr_emerg_ratelimited("%s: %d.%d No irq handler for vector (irq %d)\n",
205 __func__, smp_processor_id(),
206 vector, irq);
207 } else {
208 __this_cpu_write(vector_irq[vector], VECTOR_UNDEFINED);
209 }
210 }
211
212 irq_exit();
213
214 set_irq_regs(old_regs);
215 return 1;
216}
217
218/*
219 * Handler for X86_PLATFORM_IPI_VECTOR.
220 */
221void __smp_x86_platform_ipi(void)
222{
223 inc_irq_stat(x86_platform_ipis);
224
225 if (x86_platform_ipi_callback)
226 x86_platform_ipi_callback();
227}
228
229__visible void smp_x86_platform_ipi(struct pt_regs *regs)
230{
231 struct pt_regs *old_regs = set_irq_regs(regs);
232
233 entering_ack_irq();
234 __smp_x86_platform_ipi();
235 exiting_irq();
236 set_irq_regs(old_regs);
237}
238
239#ifdef CONFIG_HAVE_KVM
240/*
241 * Handler for POSTED_INTERRUPT_VECTOR.
242 */
243__visible void smp_kvm_posted_intr_ipi(struct pt_regs *regs)
244{
245 struct pt_regs *old_regs = set_irq_regs(regs);
246
247 ack_APIC_irq();
248
249 irq_enter();
250
251 exit_idle();
252
253 inc_irq_stat(kvm_posted_intr_ipis);
254
255 irq_exit();
256
257 set_irq_regs(old_regs);
258}
259#endif
260
261__visible void smp_trace_x86_platform_ipi(struct pt_regs *regs)
262{
263 struct pt_regs *old_regs = set_irq_regs(regs);
264
265 entering_ack_irq();
266 trace_x86_platform_ipi_entry(X86_PLATFORM_IPI_VECTOR);
267 __smp_x86_platform_ipi();
268 trace_x86_platform_ipi_exit(X86_PLATFORM_IPI_VECTOR);
269 exiting_irq();
270 set_irq_regs(old_regs);
271}
272
273EXPORT_SYMBOL_GPL(vector_used_by_percpu_irq);
274
275#ifdef CONFIG_HOTPLUG_CPU
276
277/* These two declarations are only used in check_irq_vectors_for_cpu_disable()
278 * below, which is protected by stop_machine(). Putting them on the stack
279 * results in a stack frame overflow. Dynamically allocating could result in a
280 * failure so declare these two cpumasks as global.
281 */
282static struct cpumask affinity_new, online_new;
283
284/*
285 * This cpu is going to be removed and its vectors migrated to the remaining
286 * online cpus. Check to see if there are enough vectors in the remaining cpus.
287 * This function is protected by stop_machine().
288 */
289int check_irq_vectors_for_cpu_disable(void)
290{
291 int irq, cpu;
292 unsigned int this_cpu, vector, this_count, count;
293 struct irq_desc *desc;
294 struct irq_data *data;
295
296 this_cpu = smp_processor_id();
297 cpumask_copy(&online_new, cpu_online_mask);
298 cpu_clear(this_cpu, online_new);
299
300 this_count = 0;
301 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
302 irq = __this_cpu_read(vector_irq[vector]);
303 if (irq >= 0) {
304 desc = irq_to_desc(irq);
305 data = irq_desc_get_irq_data(desc);
306 cpumask_copy(&affinity_new, data->affinity);
307 cpu_clear(this_cpu, affinity_new);
308
309 /* Do not count inactive or per-cpu irqs. */
310 if (!irq_has_action(irq) || irqd_is_per_cpu(data))
311 continue;
312
313 /*
314 * A single irq may be mapped to multiple
315 * cpu's vector_irq[] (for example IOAPIC cluster
316 * mode). In this case we have two
317 * possibilities:
318 *
319 * 1) the resulting affinity mask is empty; that is
320 * this the down'd cpu is the last cpu in the irq's
321 * affinity mask, or
322 *
323 * 2) the resulting affinity mask is no longer
324 * a subset of the online cpus but the affinity
325 * mask is not zero; that is the down'd cpu is the
326 * last online cpu in a user set affinity mask.
327 */
328 if (cpumask_empty(&affinity_new) ||
329 !cpumask_subset(&affinity_new, &online_new))
330 this_count++;
331 }
332 }
333
334 count = 0;
335 for_each_online_cpu(cpu) {
336 if (cpu == this_cpu)
337 continue;
338 /*
339 * We scan from FIRST_EXTERNAL_VECTOR to first system
340 * vector. If the vector is marked in the used vectors
341 * bitmap or an irq is assigned to it, we don't count
342 * it as available.
343 */
344 for (vector = FIRST_EXTERNAL_VECTOR;
345 vector < first_system_vector; vector++) {
346 if (!test_bit(vector, used_vectors) &&
347 per_cpu(vector_irq, cpu)[vector] < 0)
348 count++;
349 }
350 }
351
352 if (count < this_count) {
353 pr_warn("CPU %d disable failed: CPU has %u vectors assigned and there are only %u available.\n",
354 this_cpu, this_count, count);
355 return -ERANGE;
356 }
357 return 0;
358}
359
360/* A cpu has been removed from cpu_online_mask. Reset irq affinities. */
361void fixup_irqs(void)
362{
363 unsigned int irq, vector;
364 static int warned;
365 struct irq_desc *desc;
366 struct irq_data *data;
367 struct irq_chip *chip;
368
369 for_each_irq_desc(irq, desc) {
370 int break_affinity = 0;
371 int set_affinity = 1;
372 const struct cpumask *affinity;
373
374 if (!desc)
375 continue;
376 if (irq == 2)
377 continue;
378
379 /* interrupt's are disabled at this point */
380 raw_spin_lock(&desc->lock);
381
382 data = irq_desc_get_irq_data(desc);
383 affinity = data->affinity;
384 if (!irq_has_action(irq) || irqd_is_per_cpu(data) ||
385 cpumask_subset(affinity, cpu_online_mask)) {
386 raw_spin_unlock(&desc->lock);
387 continue;
388 }
389
390 /*
391 * Complete the irq move. This cpu is going down and for
392 * non intr-remapping case, we can't wait till this interrupt
393 * arrives at this cpu before completing the irq move.
394 */
395 irq_force_complete_move(irq);
396
397 if (cpumask_any_and(affinity, cpu_online_mask) >= nr_cpu_ids) {
398 break_affinity = 1;
399 affinity = cpu_online_mask;
400 }
401
402 chip = irq_data_get_irq_chip(data);
403 if (!irqd_can_move_in_process_context(data) && chip->irq_mask)
404 chip->irq_mask(data);
405
406 if (chip->irq_set_affinity)
407 chip->irq_set_affinity(data, affinity, true);
408 else if (!(warned++))
409 set_affinity = 0;
410
411 /*
412 * We unmask if the irq was not marked masked by the
413 * core code. That respects the lazy irq disable
414 * behaviour.
415 */
416 if (!irqd_can_move_in_process_context(data) &&
417 !irqd_irq_masked(data) && chip->irq_unmask)
418 chip->irq_unmask(data);
419
420 raw_spin_unlock(&desc->lock);
421
422 if (break_affinity && set_affinity)
423 pr_notice("Broke affinity for irq %i\n", irq);
424 else if (!set_affinity)
425 pr_notice("Cannot set affinity for irq %i\n", irq);
426 }
427
428 /*
429 * We can remove mdelay() and then send spuriuous interrupts to
430 * new cpu targets for all the irqs that were handled previously by
431 * this cpu. While it works, I have seen spurious interrupt messages
432 * (nothing wrong but still...).
433 *
434 * So for now, retain mdelay(1) and check the IRR and then send those
435 * interrupts to new targets as this cpu is already offlined...
436 */
437 mdelay(1);
438
439 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
440 unsigned int irr;
441
442 if (__this_cpu_read(vector_irq[vector]) <= VECTOR_UNDEFINED)
443 continue;
444
445 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
446 if (irr & (1 << (vector % 32))) {
447 irq = __this_cpu_read(vector_irq[vector]);
448
449 desc = irq_to_desc(irq);
450 data = irq_desc_get_irq_data(desc);
451 chip = irq_data_get_irq_chip(data);
452 raw_spin_lock(&desc->lock);
453 if (chip->irq_retrigger) {
454 chip->irq_retrigger(data);
455 __this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
456 }
457 raw_spin_unlock(&desc->lock);
458 }
459 if (__this_cpu_read(vector_irq[vector]) != VECTOR_RETRIGGERED)
460 __this_cpu_write(vector_irq[vector], VECTOR_UNDEFINED);
461 }
462}
463#endif
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
24DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
25EXPORT_PER_CPU_SYMBOL(irq_stat);
26
27DEFINE_PER_CPU(struct pt_regs *, irq_regs);
28EXPORT_PER_CPU_SYMBOL(irq_regs);
29
30atomic_t irq_err_count;
31
32/* Function pointer for generic interrupt vector handling */
33void (*x86_platform_ipi_callback)(void) = NULL;
34
35/*
36 * 'what should we do if we get a hw irq event on an illegal vector'.
37 * each architecture has to answer this themselves.
38 */
39void ack_bad_irq(unsigned int irq)
40{
41 if (printk_ratelimit())
42 pr_err("unexpected IRQ trap at vector %02x\n", irq);
43
44 /*
45 * Currently unexpected vectors happen only on SMP and APIC.
46 * We _must_ ack these because every local APIC has only N
47 * irq slots per priority level, and a 'hanging, unacked' IRQ
48 * holds up an irq slot - in excessive cases (when multiple
49 * unexpected vectors occur) that might lock up the APIC
50 * completely.
51 * But only ack when the APIC is enabled -AK
52 */
53 ack_APIC_irq();
54}
55
56#define irq_stats(x) (&per_cpu(irq_stat, x))
57/*
58 * /proc/interrupts printing for arch specific interrupts
59 */
60int arch_show_interrupts(struct seq_file *p, int prec)
61{
62 int j;
63
64 seq_printf(p, "%*s: ", prec, "NMI");
65 for_each_online_cpu(j)
66 seq_printf(p, "%10u ", irq_stats(j)->__nmi_count);
67 seq_puts(p, " Non-maskable interrupts\n");
68#ifdef CONFIG_X86_LOCAL_APIC
69 seq_printf(p, "%*s: ", prec, "LOC");
70 for_each_online_cpu(j)
71 seq_printf(p, "%10u ", irq_stats(j)->apic_timer_irqs);
72 seq_puts(p, " Local timer interrupts\n");
73
74 seq_printf(p, "%*s: ", prec, "SPU");
75 for_each_online_cpu(j)
76 seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count);
77 seq_puts(p, " Spurious interrupts\n");
78 seq_printf(p, "%*s: ", prec, "PMI");
79 for_each_online_cpu(j)
80 seq_printf(p, "%10u ", irq_stats(j)->apic_perf_irqs);
81 seq_puts(p, " Performance monitoring interrupts\n");
82 seq_printf(p, "%*s: ", prec, "IWI");
83 for_each_online_cpu(j)
84 seq_printf(p, "%10u ", irq_stats(j)->apic_irq_work_irqs);
85 seq_puts(p, " IRQ work interrupts\n");
86 seq_printf(p, "%*s: ", prec, "RTR");
87 for_each_online_cpu(j)
88 seq_printf(p, "%10u ", irq_stats(j)->icr_read_retry_count);
89 seq_puts(p, " APIC ICR read retries\n");
90#endif
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#ifdef CONFIG_SMP
98 seq_printf(p, "%*s: ", prec, "RES");
99 for_each_online_cpu(j)
100 seq_printf(p, "%10u ", irq_stats(j)->irq_resched_count);
101 seq_puts(p, " Rescheduling interrupts\n");
102 seq_printf(p, "%*s: ", prec, "CAL");
103 for_each_online_cpu(j)
104 seq_printf(p, "%10u ", irq_stats(j)->irq_call_count);
105 seq_puts(p, " Function call interrupts\n");
106 seq_printf(p, "%*s: ", prec, "TLB");
107 for_each_online_cpu(j)
108 seq_printf(p, "%10u ", irq_stats(j)->irq_tlb_count);
109 seq_puts(p, " TLB shootdowns\n");
110#endif
111#ifdef CONFIG_X86_THERMAL_VECTOR
112 seq_printf(p, "%*s: ", prec, "TRM");
113 for_each_online_cpu(j)
114 seq_printf(p, "%10u ", irq_stats(j)->irq_thermal_count);
115 seq_puts(p, " Thermal event interrupts\n");
116#endif
117#ifdef CONFIG_X86_MCE_THRESHOLD
118 seq_printf(p, "%*s: ", prec, "THR");
119 for_each_online_cpu(j)
120 seq_printf(p, "%10u ", irq_stats(j)->irq_threshold_count);
121 seq_puts(p, " Threshold APIC interrupts\n");
122#endif
123#ifdef CONFIG_X86_MCE_AMD
124 seq_printf(p, "%*s: ", prec, "DFR");
125 for_each_online_cpu(j)
126 seq_printf(p, "%10u ", irq_stats(j)->irq_deferred_error_count);
127 seq_puts(p, " Deferred Error APIC interrupts\n");
128#endif
129#ifdef CONFIG_X86_MCE
130 seq_printf(p, "%*s: ", prec, "MCE");
131 for_each_online_cpu(j)
132 seq_printf(p, "%10u ", per_cpu(mce_exception_count, j));
133 seq_puts(p, " Machine check exceptions\n");
134 seq_printf(p, "%*s: ", prec, "MCP");
135 for_each_online_cpu(j)
136 seq_printf(p, "%10u ", per_cpu(mce_poll_count, j));
137 seq_puts(p, " Machine check polls\n");
138#endif
139#if IS_ENABLED(CONFIG_HYPERV) || defined(CONFIG_XEN)
140 if (test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors)) {
141 seq_printf(p, "%*s: ", prec, "HYP");
142 for_each_online_cpu(j)
143 seq_printf(p, "%10u ",
144 irq_stats(j)->irq_hv_callback_count);
145 seq_puts(p, " Hypervisor callback interrupts\n");
146 }
147#endif
148 seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
149#if defined(CONFIG_X86_IO_APIC)
150 seq_printf(p, "%*s: %10u\n", prec, "MIS", atomic_read(&irq_mis_count));
151#endif
152#ifdef CONFIG_HAVE_KVM
153 seq_printf(p, "%*s: ", prec, "PIN");
154 for_each_online_cpu(j)
155 seq_printf(p, "%10u ", irq_stats(j)->kvm_posted_intr_ipis);
156 seq_puts(p, " Posted-interrupt notification event\n");
157
158 seq_printf(p, "%*s: ", prec, "PIW");
159 for_each_online_cpu(j)
160 seq_printf(p, "%10u ",
161 irq_stats(j)->kvm_posted_intr_wakeup_ipis);
162 seq_puts(p, " Posted-interrupt wakeup event\n");
163#endif
164 return 0;
165}
166
167/*
168 * /proc/stat helpers
169 */
170u64 arch_irq_stat_cpu(unsigned int cpu)
171{
172 u64 sum = irq_stats(cpu)->__nmi_count;
173
174#ifdef CONFIG_X86_LOCAL_APIC
175 sum += irq_stats(cpu)->apic_timer_irqs;
176 sum += irq_stats(cpu)->irq_spurious_count;
177 sum += irq_stats(cpu)->apic_perf_irqs;
178 sum += irq_stats(cpu)->apic_irq_work_irqs;
179 sum += irq_stats(cpu)->icr_read_retry_count;
180#endif
181 if (x86_platform_ipi_callback)
182 sum += irq_stats(cpu)->x86_platform_ipis;
183#ifdef CONFIG_SMP
184 sum += irq_stats(cpu)->irq_resched_count;
185 sum += irq_stats(cpu)->irq_call_count;
186#endif
187#ifdef CONFIG_X86_THERMAL_VECTOR
188 sum += irq_stats(cpu)->irq_thermal_count;
189#endif
190#ifdef CONFIG_X86_MCE_THRESHOLD
191 sum += irq_stats(cpu)->irq_threshold_count;
192#endif
193#ifdef CONFIG_X86_MCE
194 sum += per_cpu(mce_exception_count, cpu);
195 sum += per_cpu(mce_poll_count, cpu);
196#endif
197 return sum;
198}
199
200u64 arch_irq_stat(void)
201{
202 u64 sum = atomic_read(&irq_err_count);
203 return sum;
204}
205
206
207/*
208 * do_IRQ handles all normal device IRQ's (the special
209 * SMP cross-CPU interrupts have their own specific
210 * handlers).
211 */
212__visible unsigned int __irq_entry do_IRQ(struct pt_regs *regs)
213{
214 struct pt_regs *old_regs = set_irq_regs(regs);
215 struct irq_desc * desc;
216 /* high bit used in ret_from_ code */
217 unsigned vector = ~regs->orig_ax;
218
219 /*
220 * NB: Unlike exception entries, IRQ entries do not reliably
221 * handle context tracking in the low-level entry code. This is
222 * because syscall entries execute briefly with IRQs on before
223 * updating context tracking state, so we can take an IRQ from
224 * kernel mode with CONTEXT_USER. The low-level entry code only
225 * updates the context if we came from user mode, so we won't
226 * switch to CONTEXT_KERNEL. We'll fix that once the syscall
227 * code is cleaned up enough that we can cleanly defer enabling
228 * IRQs.
229 */
230
231 entering_irq();
232
233 /* entering_irq() tells RCU that we're not quiescent. Check it. */
234 RCU_LOCKDEP_WARN(!rcu_is_watching(), "IRQ failed to wake up RCU");
235
236 desc = __this_cpu_read(vector_irq[vector]);
237
238 if (!handle_irq(desc, regs)) {
239 ack_APIC_irq();
240
241 if (desc != VECTOR_RETRIGGERED) {
242 pr_emerg_ratelimited("%s: %d.%d No irq handler for vector\n",
243 __func__, smp_processor_id(),
244 vector);
245 } else {
246 __this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
247 }
248 }
249
250 exiting_irq();
251
252 set_irq_regs(old_regs);
253 return 1;
254}
255
256/*
257 * Handler for X86_PLATFORM_IPI_VECTOR.
258 */
259void __smp_x86_platform_ipi(void)
260{
261 inc_irq_stat(x86_platform_ipis);
262
263 if (x86_platform_ipi_callback)
264 x86_platform_ipi_callback();
265}
266
267__visible void smp_x86_platform_ipi(struct pt_regs *regs)
268{
269 struct pt_regs *old_regs = set_irq_regs(regs);
270
271 entering_ack_irq();
272 __smp_x86_platform_ipi();
273 exiting_irq();
274 set_irq_regs(old_regs);
275}
276
277#ifdef CONFIG_HAVE_KVM
278static void dummy_handler(void) {}
279static void (*kvm_posted_intr_wakeup_handler)(void) = dummy_handler;
280
281void kvm_set_posted_intr_wakeup_handler(void (*handler)(void))
282{
283 if (handler)
284 kvm_posted_intr_wakeup_handler = handler;
285 else
286 kvm_posted_intr_wakeup_handler = dummy_handler;
287}
288EXPORT_SYMBOL_GPL(kvm_set_posted_intr_wakeup_handler);
289
290/*
291 * Handler for POSTED_INTERRUPT_VECTOR.
292 */
293__visible void smp_kvm_posted_intr_ipi(struct pt_regs *regs)
294{
295 struct pt_regs *old_regs = set_irq_regs(regs);
296
297 entering_ack_irq();
298 inc_irq_stat(kvm_posted_intr_ipis);
299 exiting_irq();
300 set_irq_regs(old_regs);
301}
302
303/*
304 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
305 */
306__visible void smp_kvm_posted_intr_wakeup_ipi(struct pt_regs *regs)
307{
308 struct pt_regs *old_regs = set_irq_regs(regs);
309
310 entering_ack_irq();
311 inc_irq_stat(kvm_posted_intr_wakeup_ipis);
312 kvm_posted_intr_wakeup_handler();
313 exiting_irq();
314 set_irq_regs(old_regs);
315}
316#endif
317
318__visible void smp_trace_x86_platform_ipi(struct pt_regs *regs)
319{
320 struct pt_regs *old_regs = set_irq_regs(regs);
321
322 entering_ack_irq();
323 trace_x86_platform_ipi_entry(X86_PLATFORM_IPI_VECTOR);
324 __smp_x86_platform_ipi();
325 trace_x86_platform_ipi_exit(X86_PLATFORM_IPI_VECTOR);
326 exiting_irq();
327 set_irq_regs(old_regs);
328}
329
330EXPORT_SYMBOL_GPL(vector_used_by_percpu_irq);
331
332#ifdef CONFIG_HOTPLUG_CPU
333
334/* These two declarations are only used in check_irq_vectors_for_cpu_disable()
335 * below, which is protected by stop_machine(). Putting them on the stack
336 * results in a stack frame overflow. Dynamically allocating could result in a
337 * failure so declare these two cpumasks as global.
338 */
339static struct cpumask affinity_new, online_new;
340
341/*
342 * This cpu is going to be removed and its vectors migrated to the remaining
343 * online cpus. Check to see if there are enough vectors in the remaining cpus.
344 * This function is protected by stop_machine().
345 */
346int check_irq_vectors_for_cpu_disable(void)
347{
348 unsigned int this_cpu, vector, this_count, count;
349 struct irq_desc *desc;
350 struct irq_data *data;
351 int cpu;
352
353 this_cpu = smp_processor_id();
354 cpumask_copy(&online_new, cpu_online_mask);
355 cpumask_clear_cpu(this_cpu, &online_new);
356
357 this_count = 0;
358 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
359 desc = __this_cpu_read(vector_irq[vector]);
360 if (IS_ERR_OR_NULL(desc))
361 continue;
362 /*
363 * Protect against concurrent action removal, affinity
364 * changes etc.
365 */
366 raw_spin_lock(&desc->lock);
367 data = irq_desc_get_irq_data(desc);
368 cpumask_copy(&affinity_new,
369 irq_data_get_affinity_mask(data));
370 cpumask_clear_cpu(this_cpu, &affinity_new);
371
372 /* Do not count inactive or per-cpu irqs. */
373 if (!irq_desc_has_action(desc) || irqd_is_per_cpu(data)) {
374 raw_spin_unlock(&desc->lock);
375 continue;
376 }
377
378 raw_spin_unlock(&desc->lock);
379 /*
380 * A single irq may be mapped to multiple cpu's
381 * vector_irq[] (for example IOAPIC cluster mode). In
382 * this case we have two possibilities:
383 *
384 * 1) the resulting affinity mask is empty; that is
385 * this the down'd cpu is the last cpu in the irq's
386 * affinity mask, or
387 *
388 * 2) the resulting affinity mask is no longer a
389 * subset of the online cpus but the affinity mask is
390 * not zero; that is the down'd cpu is the last online
391 * cpu in a user set affinity mask.
392 */
393 if (cpumask_empty(&affinity_new) ||
394 !cpumask_subset(&affinity_new, &online_new))
395 this_count++;
396 }
397
398 count = 0;
399 for_each_online_cpu(cpu) {
400 if (cpu == this_cpu)
401 continue;
402 /*
403 * We scan from FIRST_EXTERNAL_VECTOR to first system
404 * vector. If the vector is marked in the used vectors
405 * bitmap or an irq is assigned to it, we don't count
406 * it as available.
407 *
408 * As this is an inaccurate snapshot anyway, we can do
409 * this w/o holding vector_lock.
410 */
411 for (vector = FIRST_EXTERNAL_VECTOR;
412 vector < first_system_vector; vector++) {
413 if (!test_bit(vector, used_vectors) &&
414 IS_ERR_OR_NULL(per_cpu(vector_irq, cpu)[vector]))
415 count++;
416 }
417 }
418
419 if (count < this_count) {
420 pr_warn("CPU %d disable failed: CPU has %u vectors assigned and there are only %u available.\n",
421 this_cpu, this_count, count);
422 return -ERANGE;
423 }
424 return 0;
425}
426
427/* A cpu has been removed from cpu_online_mask. Reset irq affinities. */
428void fixup_irqs(void)
429{
430 unsigned int irq, vector;
431 static int warned;
432 struct irq_desc *desc;
433 struct irq_data *data;
434 struct irq_chip *chip;
435 int ret;
436
437 for_each_irq_desc(irq, desc) {
438 int break_affinity = 0;
439 int set_affinity = 1;
440 const struct cpumask *affinity;
441
442 if (!desc)
443 continue;
444 if (irq == 2)
445 continue;
446
447 /* interrupt's are disabled at this point */
448 raw_spin_lock(&desc->lock);
449
450 data = irq_desc_get_irq_data(desc);
451 affinity = irq_data_get_affinity_mask(data);
452 if (!irq_has_action(irq) || irqd_is_per_cpu(data) ||
453 cpumask_subset(affinity, cpu_online_mask)) {
454 raw_spin_unlock(&desc->lock);
455 continue;
456 }
457
458 /*
459 * Complete the irq move. This cpu is going down and for
460 * non intr-remapping case, we can't wait till this interrupt
461 * arrives at this cpu before completing the irq move.
462 */
463 irq_force_complete_move(desc);
464
465 if (cpumask_any_and(affinity, cpu_online_mask) >= nr_cpu_ids) {
466 break_affinity = 1;
467 affinity = cpu_online_mask;
468 }
469
470 chip = irq_data_get_irq_chip(data);
471 /*
472 * The interrupt descriptor might have been cleaned up
473 * already, but it is not yet removed from the radix tree
474 */
475 if (!chip) {
476 raw_spin_unlock(&desc->lock);
477 continue;
478 }
479
480 if (!irqd_can_move_in_process_context(data) && chip->irq_mask)
481 chip->irq_mask(data);
482
483 if (chip->irq_set_affinity) {
484 ret = chip->irq_set_affinity(data, affinity, true);
485 if (ret == -ENOSPC)
486 pr_crit("IRQ %d set affinity failed because there are no available vectors. The device assigned to this IRQ is unstable.\n", irq);
487 } else {
488 if (!(warned++))
489 set_affinity = 0;
490 }
491
492 /*
493 * We unmask if the irq was not marked masked by the
494 * core code. That respects the lazy irq disable
495 * behaviour.
496 */
497 if (!irqd_can_move_in_process_context(data) &&
498 !irqd_irq_masked(data) && chip->irq_unmask)
499 chip->irq_unmask(data);
500
501 raw_spin_unlock(&desc->lock);
502
503 if (break_affinity && set_affinity)
504 pr_notice("Broke affinity for irq %i\n", irq);
505 else if (!set_affinity)
506 pr_notice("Cannot set affinity for irq %i\n", irq);
507 }
508
509 /*
510 * We can remove mdelay() and then send spuriuous interrupts to
511 * new cpu targets for all the irqs that were handled previously by
512 * this cpu. While it works, I have seen spurious interrupt messages
513 * (nothing wrong but still...).
514 *
515 * So for now, retain mdelay(1) and check the IRR and then send those
516 * interrupts to new targets as this cpu is already offlined...
517 */
518 mdelay(1);
519
520 /*
521 * We can walk the vector array of this cpu without holding
522 * vector_lock because the cpu is already marked !online, so
523 * nothing else will touch it.
524 */
525 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
526 unsigned int irr;
527
528 if (IS_ERR_OR_NULL(__this_cpu_read(vector_irq[vector])))
529 continue;
530
531 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
532 if (irr & (1 << (vector % 32))) {
533 desc = __this_cpu_read(vector_irq[vector]);
534
535 raw_spin_lock(&desc->lock);
536 data = irq_desc_get_irq_data(desc);
537 chip = irq_data_get_irq_chip(data);
538 if (chip->irq_retrigger) {
539 chip->irq_retrigger(data);
540 __this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
541 }
542 raw_spin_unlock(&desc->lock);
543 }
544 if (__this_cpu_read(vector_irq[vector]) != VECTOR_RETRIGGERED)
545 __this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
546 }
547}
548#endif