arch/x86/kernel/irq.c

   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 #include <linux/irq.h>
  14
  15 #include <asm/apic.h>
  16 #include <asm/io_apic.h>
  17 #include <asm/irq.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
  25 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
  26 EXPORT_PER_CPU_SYMBOL(irq_stat);
  27
  28 DEFINE_PER_CPU(struct pt_regs *, irq_regs);
  29 EXPORT_PER_CPU_SYMBOL(irq_regs);
  30
  31 atomic_t irq_err_count;
  32
  33 /*
  34  * 'what should we do if we get a hw irq event on an illegal vector'.
  35  * each architecture has to answer this themselves.
  36  */
  37 void ack_bad_irq(unsigned int irq)
  38 {
  39         if (printk_ratelimit())
  40                 pr_err("unexpected IRQ trap at vector %02x\n", irq);
  41
  42         /*
  43          * Currently unexpected vectors happen only on SMP and APIC.
  44          * We _must_ ack these because every local APIC has only N
  45          * irq slots per priority level, and a 'hanging, unacked' IRQ
  46          * holds up an irq slot - in excessive cases (when multiple
  47          * unexpected vectors occur) that might lock up the APIC
  48          * completely.
  49          * But only ack when the APIC is enabled -AK
  50          */
  51         ack_APIC_irq();
  52 }
  53
  54 #define irq_stats(x)            (&per_cpu(irq_stat, x))
  55 /*
  56  * /proc/interrupts printing for arch specific interrupts
  57  */
  58 int arch_show_interrupts(struct seq_file *p, int prec)
  59 {
  60         int j;
  61
  62         seq_printf(p, "%*s: ", prec, "NMI");
  63         for_each_online_cpu(j)
  64                 seq_printf(p, "%10u ", irq_stats(j)->__nmi_count);
  65         seq_puts(p, "  Non-maskable interrupts\n");
  66 #ifdef CONFIG_X86_LOCAL_APIC
  67         seq_printf(p, "%*s: ", prec, "LOC");
  68         for_each_online_cpu(j)
  69                 seq_printf(p, "%10u ", irq_stats(j)->apic_timer_irqs);
  70         seq_puts(p, "  Local timer interrupts\n");
  71
  72         seq_printf(p, "%*s: ", prec, "SPU");
  73         for_each_online_cpu(j)
  74                 seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count);
  75         seq_puts(p, "  Spurious interrupts\n");
  76         seq_printf(p, "%*s: ", prec, "PMI");
  77         for_each_online_cpu(j)
  78                 seq_printf(p, "%10u ", irq_stats(j)->apic_perf_irqs);
  79         seq_puts(p, "  Performance monitoring interrupts\n");
  80         seq_printf(p, "%*s: ", prec, "IWI");
  81         for_each_online_cpu(j)
  82                 seq_printf(p, "%10u ", irq_stats(j)->apic_irq_work_irqs);
  83         seq_puts(p, "  IRQ work interrupts\n");
  84         seq_printf(p, "%*s: ", prec, "RTR");
  85         for_each_online_cpu(j)
  86                 seq_printf(p, "%10u ", irq_stats(j)->icr_read_retry_count);
  87         seq_puts(p, "  APIC ICR read retries\n");
  88         if (x86_platform_ipi_callback) {
  89                 seq_printf(p, "%*s: ", prec, "PLT");
  90                 for_each_online_cpu(j)
  91                         seq_printf(p, "%10u ", irq_stats(j)->x86_platform_ipis);
  92                 seq_puts(p, "  Platform interrupts\n");
  93         }
  94 #endif
  95 #ifdef CONFIG_SMP
  96         seq_printf(p, "%*s: ", prec, "RES");
  97         for_each_online_cpu(j)
  98                 seq_printf(p, "%10u ", irq_stats(j)->irq_resched_count);
  99         seq_puts(p, "  Rescheduling interrupts\n");
 100         seq_printf(p, "%*s: ", prec, "CAL");
 101         for_each_online_cpu(j)
 102                 seq_printf(p, "%10u ", irq_stats(j)->irq_call_count);
 103         seq_puts(p, "  Function call interrupts\n");
 104         seq_printf(p, "%*s: ", prec, "TLB");
 105         for_each_online_cpu(j)
 106                 seq_printf(p, "%10u ", irq_stats(j)->irq_tlb_count);
 107         seq_puts(p, "  TLB shootdowns\n");
 108 #endif
 109 #ifdef CONFIG_X86_THERMAL_VECTOR
 110         seq_printf(p, "%*s: ", prec, "TRM");
 111         for_each_online_cpu(j)
 112                 seq_printf(p, "%10u ", irq_stats(j)->irq_thermal_count);
 113         seq_puts(p, "  Thermal event interrupts\n");
 114 #endif
 115 #ifdef CONFIG_X86_MCE_THRESHOLD
 116         seq_printf(p, "%*s: ", prec, "THR");
 117         for_each_online_cpu(j)
 118                 seq_printf(p, "%10u ", irq_stats(j)->irq_threshold_count);
 119         seq_puts(p, "  Threshold APIC interrupts\n");
 120 #endif
 121 #ifdef CONFIG_X86_MCE_AMD
 122         seq_printf(p, "%*s: ", prec, "DFR");
 123         for_each_online_cpu(j)
 124                 seq_printf(p, "%10u ", irq_stats(j)->irq_deferred_error_count);
 125         seq_puts(p, "  Deferred Error APIC interrupts\n");
 126 #endif
 127 #ifdef CONFIG_X86_MCE
 128         seq_printf(p, "%*s: ", prec, "MCE");
 129         for_each_online_cpu(j)
 130                 seq_printf(p, "%10u ", per_cpu(mce_exception_count, j));
 131         seq_puts(p, "  Machine check exceptions\n");
 132         seq_printf(p, "%*s: ", prec, "MCP");
 133         for_each_online_cpu(j)
 134                 seq_printf(p, "%10u ", per_cpu(mce_poll_count, j));
 135         seq_puts(p, "  Machine check polls\n");
 136 #endif
 137 #if IS_ENABLED(CONFIG_HYPERV) || defined(CONFIG_XEN)
 138         if (test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors)) {
 139                 seq_printf(p, "%*s: ", prec, "HYP");
 140                 for_each_online_cpu(j)
 141                         seq_printf(p, "%10u ",
 142                                    irq_stats(j)->irq_hv_callback_count);
 143                 seq_puts(p, "  Hypervisor callback interrupts\n");
 144         }
 145 #endif
 146         seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
 147 #if defined(CONFIG_X86_IO_APIC)
 148         seq_printf(p, "%*s: %10u\n", prec, "MIS", atomic_read(&irq_mis_count));
 149 #endif
 150 #ifdef CONFIG_HAVE_KVM
 151         seq_printf(p, "%*s: ", prec, "PIN");
 152         for_each_online_cpu(j)
 153                 seq_printf(p, "%10u ", irq_stats(j)->kvm_posted_intr_ipis);
 154         seq_puts(p, "  Posted-interrupt notification event\n");
 155
 156         seq_printf(p, "%*s: ", prec, "NPI");
 157         for_each_online_cpu(j)
 158                 seq_printf(p, "%10u ",
 159                            irq_stats(j)->kvm_posted_intr_nested_ipis);
 160         seq_puts(p, "  Nested posted-interrupt event\n");
 161
 162         seq_printf(p, "%*s: ", prec, "PIW");
 163         for_each_online_cpu(j)
 164                 seq_printf(p, "%10u ",
 165                            irq_stats(j)->kvm_posted_intr_wakeup_ipis);
 166         seq_puts(p, "  Posted-interrupt wakeup event\n");
 167 #endif
 168         return 0;
 169 }
 170
 171 /*
 172  * /proc/stat helpers
 173  */
 174 u64 arch_irq_stat_cpu(unsigned int cpu)
 175 {
 176         u64 sum = irq_stats(cpu)->__nmi_count;
 177
 178 #ifdef CONFIG_X86_LOCAL_APIC
 179         sum += irq_stats(cpu)->apic_timer_irqs;
 180         sum += irq_stats(cpu)->irq_spurious_count;
 181         sum += irq_stats(cpu)->apic_perf_irqs;
 182         sum += irq_stats(cpu)->apic_irq_work_irqs;
 183         sum += irq_stats(cpu)->icr_read_retry_count;
 184         if (x86_platform_ipi_callback)
 185                 sum += irq_stats(cpu)->x86_platform_ipis;
 186 #endif
 187 #ifdef CONFIG_SMP
 188         sum += irq_stats(cpu)->irq_resched_count;
 189         sum += irq_stats(cpu)->irq_call_count;
 190 #endif
 191 #ifdef CONFIG_X86_THERMAL_VECTOR
 192         sum += irq_stats(cpu)->irq_thermal_count;
 193 #endif
 194 #ifdef CONFIG_X86_MCE_THRESHOLD
 195         sum += irq_stats(cpu)->irq_threshold_count;
 196 #endif
 197 #ifdef CONFIG_X86_MCE
 198         sum += per_cpu(mce_exception_count, cpu);
 199         sum += per_cpu(mce_poll_count, cpu);
 200 #endif
 201         return sum;
 202 }
 203
 204 u64 arch_irq_stat(void)
 205 {
 206         u64 sum = atomic_read(&irq_err_count);
 207         return sum;
 208 }
 209
 210
 211 /*
 212  * do_IRQ handles all normal device IRQ's (the special
 213  * SMP cross-CPU interrupts have their own specific
 214  * handlers).
 215  */
 216 __visible unsigned int __irq_entry do_IRQ(struct pt_regs *regs)
 217 {
 218         struct pt_regs *old_regs = set_irq_regs(regs);
 219         struct irq_desc * desc;
 220         /* high bit used in ret_from_ code  */
 221         unsigned vector = ~regs->orig_ax;
 222
 223         entering_irq();
 224
 225         /* entering_irq() tells RCU that we're not quiescent.  Check it. */
 226         RCU_LOCKDEP_WARN(!rcu_is_watching(), "IRQ failed to wake up RCU");
 227
 228         desc = __this_cpu_read(vector_irq[vector]);
 229
 230         if (!handle_irq(desc, regs)) {
 231                 ack_APIC_irq();
 232
 233                 if (desc != VECTOR_RETRIGGERED) {
 234                         pr_emerg_ratelimited("%s: %d.%d No irq handler for vector\n",
 235                                              __func__, smp_processor_id(),
 236                                              vector);
 237                 } else {
 238                         __this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
 239                 }
 240         }
 241
 242         exiting_irq();
 243
 244         set_irq_regs(old_regs);
 245         return 1;
 246 }
 247
 248 #ifdef CONFIG_X86_LOCAL_APIC
 249 /* Function pointer for generic interrupt vector handling */
 250 void (*x86_platform_ipi_callback)(void) = NULL;
 251 /*
 252  * Handler for X86_PLATFORM_IPI_VECTOR.
 253  */
 254 __visible void __irq_entry smp_x86_platform_ipi(struct pt_regs *regs)
 255 {
 256         struct pt_regs *old_regs = set_irq_regs(regs);
 257
 258         entering_ack_irq();
 259         trace_x86_platform_ipi_entry(X86_PLATFORM_IPI_VECTOR);
 260         inc_irq_stat(x86_platform_ipis);
 261         if (x86_platform_ipi_callback)
 262                 x86_platform_ipi_callback();
 263         trace_x86_platform_ipi_exit(X86_PLATFORM_IPI_VECTOR);
 264         exiting_irq();
 265         set_irq_regs(old_regs);
 266 }
 267 #endif
 268
 269 #ifdef CONFIG_HAVE_KVM
 270 static void dummy_handler(void) {}
 271 static void (*kvm_posted_intr_wakeup_handler)(void) = dummy_handler;
 272
 273 void kvm_set_posted_intr_wakeup_handler(void (*handler)(void))
 274 {
 275         if (handler)
 276                 kvm_posted_intr_wakeup_handler = handler;
 277         else {
 278                 kvm_posted_intr_wakeup_handler = dummy_handler;
 279                 synchronize_rcu();
 280         }
 281 }
 282 EXPORT_SYMBOL_GPL(kvm_set_posted_intr_wakeup_handler);
 283
 284 /*
 285  * Handler for POSTED_INTERRUPT_VECTOR.
 286  */
 287 __visible void smp_kvm_posted_intr_ipi(struct pt_regs *regs)
 288 {
 289         struct pt_regs *old_regs = set_irq_regs(regs);
 290
 291         entering_ack_irq();
 292         inc_irq_stat(kvm_posted_intr_ipis);
 293         exiting_irq();
 294         set_irq_regs(old_regs);
 295 }
 296
 297 /*
 298  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
 299  */
 300 __visible void smp_kvm_posted_intr_wakeup_ipi(struct pt_regs *regs)
 301 {
 302         struct pt_regs *old_regs = set_irq_regs(regs);
 303
 304         entering_ack_irq();
 305         inc_irq_stat(kvm_posted_intr_wakeup_ipis);
 306         kvm_posted_intr_wakeup_handler();
 307         exiting_irq();
 308         set_irq_regs(old_regs);
 309 }
 310
 311 /*
 312  * Handler for POSTED_INTERRUPT_NESTED_VECTOR.
 313  */
 314 __visible void smp_kvm_posted_intr_nested_ipi(struct pt_regs *regs)
 315 {
 316         struct pt_regs *old_regs = set_irq_regs(regs);
 317
 318         entering_ack_irq();
 319         inc_irq_stat(kvm_posted_intr_nested_ipis);
 320         exiting_irq();
 321         set_irq_regs(old_regs);
 322 }
 323 #endif
 324
 325
 326 #ifdef CONFIG_HOTPLUG_CPU
 327
 328 /* These two declarations are only used in check_irq_vectors_for_cpu_disable()
 329  * below, which is protected by stop_machine().  Putting them on the stack
 330  * results in a stack frame overflow.  Dynamically allocating could result in a
 331  * failure so declare these two cpumasks as global.
 332  */
 333 static struct cpumask affinity_new, online_new;
 334
 335 /*
 336  * This cpu is going to be removed and its vectors migrated to the remaining
 337  * online cpus.  Check to see if there are enough vectors in the remaining cpus.
 338  * This function is protected by stop_machine().
 339  */
 340 int check_irq_vectors_for_cpu_disable(void)
 341 {
 342         unsigned int this_cpu, vector, this_count, count;
 343         struct irq_desc *desc;
 344         struct irq_data *data;
 345         int cpu;
 346
 347         this_cpu = smp_processor_id();
 348         cpumask_copy(&online_new, cpu_online_mask);
 349         cpumask_clear_cpu(this_cpu, &online_new);
 350
 351         this_count = 0;
 352         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
 353                 desc = __this_cpu_read(vector_irq[vector]);
 354                 if (IS_ERR_OR_NULL(desc))
 355                         continue;
 356                 /*
 357                  * Protect against concurrent action removal, affinity
 358                  * changes etc.
 359                  */
 360                 raw_spin_lock(&desc->lock);
 361                 data = irq_desc_get_irq_data(desc);
 362                 cpumask_copy(&affinity_new,
 363                              irq_data_get_affinity_mask(data));
 364                 cpumask_clear_cpu(this_cpu, &affinity_new);
 365
 366                 /* Do not count inactive or per-cpu irqs. */
 367                 if (!irq_desc_has_action(desc) || irqd_is_per_cpu(data)) {
 368                         raw_spin_unlock(&desc->lock);
 369                         continue;
 370                 }
 371
 372                 raw_spin_unlock(&desc->lock);
 373                 /*
 374                  * A single irq may be mapped to multiple cpu's
 375                  * vector_irq[] (for example IOAPIC cluster mode).  In
 376                  * this case we have two possibilities:
 377                  *
 378                  * 1) the resulting affinity mask is empty; that is
 379                  * this the down'd cpu is the last cpu in the irq's
 380                  * affinity mask, or
 381                  *
 382                  * 2) the resulting affinity mask is no longer a
 383                  * subset of the online cpus but the affinity mask is
 384                  * not zero; that is the down'd cpu is the last online
 385                  * cpu in a user set affinity mask.
 386                  */
 387                 if (cpumask_empty(&affinity_new) ||
 388                     !cpumask_subset(&affinity_new, &online_new))
 389                         this_count++;
 390         }
 391         /* No need to check any further. */
 392         if (!this_count)
 393                 return 0;
 394
 395         count = 0;
 396         for_each_online_cpu(cpu) {
 397                 if (cpu == this_cpu)
 398                         continue;
 399                 /*
 400                  * We scan from FIRST_EXTERNAL_VECTOR to first system
 401                  * vector. If the vector is marked in the used vectors
 402                  * bitmap or an irq is assigned to it, we don't count
 403                  * it as available.
 404                  *
 405                  * As this is an inaccurate snapshot anyway, we can do
 406                  * this w/o holding vector_lock.
 407                  */
 408                 for (vector = FIRST_EXTERNAL_VECTOR;
 409                      vector < FIRST_SYSTEM_VECTOR; vector++) {
 410                         if (!test_bit(vector, used_vectors) &&
 411                             IS_ERR_OR_NULL(per_cpu(vector_irq, cpu)[vector])) {
 412                                 if (++count == this_count)
 413                                         return 0;
 414                         }
 415                 }
 416         }
 417
 418         if (count < this_count) {
 419                 pr_warn("CPU %d disable failed: CPU has %u vectors assigned and there are only %u available.\n",
 420                         this_cpu, this_count, count);
 421                 return -ERANGE;
 422         }
 423         return 0;
 424 }
 425
 426 /* A cpu has been removed from cpu_online_mask.  Reset irq affinities. */
 427 void fixup_irqs(void)
 428 {
 429         unsigned int irr, vector;
 430         struct irq_desc *desc;
 431         struct irq_data *data;
 432         struct irq_chip *chip;
 433
 434         irq_migrate_all_off_this_cpu();
 435
 436         /*
 437          * We can remove mdelay() and then send spuriuous interrupts to
 438          * new cpu targets for all the irqs that were handled previously by
 439          * this cpu. While it works, I have seen spurious interrupt messages
 440          * (nothing wrong but still...).
 441          *
 442          * So for now, retain mdelay(1) and check the IRR and then send those
 443          * interrupts to new targets as this cpu is already offlined...
 444          */
 445         mdelay(1);
 446
 447         /*
 448          * We can walk the vector array of this cpu without holding
 449          * vector_lock because the cpu is already marked !online, so
 450          * nothing else will touch it.
 451          */
 452         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
 453                 if (IS_ERR_OR_NULL(__this_cpu_read(vector_irq[vector])))
 454                         continue;
 455
 456                 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
 457                 if (irr  & (1 << (vector % 32))) {
 458                         desc = __this_cpu_read(vector_irq[vector]);
 459
 460                         raw_spin_lock(&desc->lock);
 461                         data = irq_desc_get_irq_data(desc);
 462                         chip = irq_data_get_irq_chip(data);
 463                         if (chip->irq_retrigger) {
 464                                 chip->irq_retrigger(data);
 465                                 __this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
 466                         }
 467                         raw_spin_unlock(&desc->lock);
 468                 }
 469                 if (__this_cpu_read(vector_irq[vector]) != VECTOR_RETRIGGERED)
 470                         __this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
 471         }
 472 }
 473 #endif