arch/x86/kernel/irq.c

   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/apic.h>
  17 #include <asm/io_apic.h>
  18 #include <asm/irq.h>
  19 #include <asm/mce.h>
  20 #include <asm/hw_irq.h>
  21 #include <asm/desc.h>
  22
  23 #define CREATE_TRACE_POINTS
  24 #include <asm/trace/irq_vectors.h>
  25
  26 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
  27 EXPORT_PER_CPU_SYMBOL(irq_stat);
  28
  29 DEFINE_PER_CPU(struct pt_regs *, irq_regs);
  30 EXPORT_PER_CPU_SYMBOL(irq_regs);
  31
  32 atomic_t irq_err_count;
  33
  34 /*
  35  * 'what should we do if we get a hw irq event on an illegal vector'.
  36  * each architecture has to answer this themselves.
  37  */
  38 void ack_bad_irq(unsigned int irq)
  39 {
  40         if (printk_ratelimit())
  41                 pr_err("unexpected IRQ trap at vector %02x\n", irq);
  42
  43         /*
  44          * Currently unexpected vectors happen only on SMP and APIC.
  45          * We _must_ ack these because every local APIC has only N
  46          * irq slots per priority level, and a 'hanging, unacked' IRQ
  47          * holds up an irq slot - in excessive cases (when multiple
  48          * unexpected vectors occur) that might lock up the APIC
  49          * completely.
  50          * But only ack when the APIC is enabled -AK
  51          */
  52         ack_APIC_irq();
  53 }
  54
  55 #define irq_stats(x)            (&per_cpu(irq_stat, x))
  56 /*
  57  * /proc/interrupts printing for arch specific interrupts
  58  */
  59 int arch_show_interrupts(struct seq_file *p, int prec)
  60 {
  61         int j;
  62
  63         seq_printf(p, "%*s: ", prec, "NMI");
  64         for_each_online_cpu(j)
  65                 seq_printf(p, "%10u ", irq_stats(j)->__nmi_count);
  66         seq_puts(p, "  Non-maskable interrupts\n");
  67 #ifdef CONFIG_X86_LOCAL_APIC
  68         seq_printf(p, "%*s: ", prec, "LOC");
  69         for_each_online_cpu(j)
  70                 seq_printf(p, "%10u ", irq_stats(j)->apic_timer_irqs);
  71         seq_puts(p, "  Local timer interrupts\n");
  72
  73         seq_printf(p, "%*s: ", prec, "SPU");
  74         for_each_online_cpu(j)
  75                 seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count);
  76         seq_puts(p, "  Spurious interrupts\n");
  77         seq_printf(p, "%*s: ", prec, "PMI");
  78         for_each_online_cpu(j)
  79                 seq_printf(p, "%10u ", irq_stats(j)->apic_perf_irqs);
  80         seq_puts(p, "  Performance monitoring interrupts\n");
  81         seq_printf(p, "%*s: ", prec, "IWI");
  82         for_each_online_cpu(j)
  83                 seq_printf(p, "%10u ", irq_stats(j)->apic_irq_work_irqs);
  84         seq_puts(p, "  IRQ work interrupts\n");
  85         seq_printf(p, "%*s: ", prec, "RTR");
  86         for_each_online_cpu(j)
  87                 seq_printf(p, "%10u ", irq_stats(j)->icr_read_retry_count);
  88         seq_puts(p, "  APIC ICR read retries\n");
  89         if (x86_platform_ipi_callback) {
  90                 seq_printf(p, "%*s: ", prec, "PLT");
  91                 for_each_online_cpu(j)
  92                         seq_printf(p, "%10u ", irq_stats(j)->x86_platform_ipis);
  93                 seq_puts(p, "  Platform interrupts\n");
  94         }
  95 #endif
  96 #ifdef CONFIG_SMP
  97         seq_printf(p, "%*s: ", prec, "RES");
  98         for_each_online_cpu(j)
  99                 seq_printf(p, "%10u ", irq_stats(j)->irq_resched_count);
 100         seq_puts(p, "  Rescheduling interrupts\n");
 101         seq_printf(p, "%*s: ", prec, "CAL");
 102         for_each_online_cpu(j)
 103                 seq_printf(p, "%10u ", irq_stats(j)->irq_call_count);
 104         seq_puts(p, "  Function call interrupts\n");
 105         seq_printf(p, "%*s: ", prec, "TLB");
 106         for_each_online_cpu(j)
 107                 seq_printf(p, "%10u ", irq_stats(j)->irq_tlb_count);
 108         seq_puts(p, "  TLB shootdowns\n");
 109 #endif
 110 #ifdef CONFIG_X86_THERMAL_VECTOR
 111         seq_printf(p, "%*s: ", prec, "TRM");
 112         for_each_online_cpu(j)
 113                 seq_printf(p, "%10u ", irq_stats(j)->irq_thermal_count);
 114         seq_puts(p, "  Thermal event interrupts\n");
 115 #endif
 116 #ifdef CONFIG_X86_MCE_THRESHOLD
 117         seq_printf(p, "%*s: ", prec, "THR");
 118         for_each_online_cpu(j)
 119                 seq_printf(p, "%10u ", irq_stats(j)->irq_threshold_count);
 120         seq_puts(p, "  Threshold APIC interrupts\n");
 121 #endif
 122 #ifdef CONFIG_X86_MCE_AMD
 123         seq_printf(p, "%*s: ", prec, "DFR");
 124         for_each_online_cpu(j)
 125                 seq_printf(p, "%10u ", irq_stats(j)->irq_deferred_error_count);
 126         seq_puts(p, "  Deferred Error APIC interrupts\n");
 127 #endif
 128 #ifdef CONFIG_X86_MCE
 129         seq_printf(p, "%*s: ", prec, "MCE");
 130         for_each_online_cpu(j)
 131                 seq_printf(p, "%10u ", per_cpu(mce_exception_count, j));
 132         seq_puts(p, "  Machine check exceptions\n");
 133         seq_printf(p, "%*s: ", prec, "MCP");
 134         for_each_online_cpu(j)
 135                 seq_printf(p, "%10u ", per_cpu(mce_poll_count, j));
 136         seq_puts(p, "  Machine check polls\n");
 137 #endif
 138 #ifdef CONFIG_X86_HV_CALLBACK_VECTOR
 139         if (test_bit(HYPERVISOR_CALLBACK_VECTOR, system_vectors)) {
 140                 seq_printf(p, "%*s: ", prec, "HYP");
 141                 for_each_online_cpu(j)
 142                         seq_printf(p, "%10u ",
 143                                    irq_stats(j)->irq_hv_callback_count);
 144                 seq_puts(p, "  Hypervisor callback interrupts\n");
 145         }
 146 #endif
 147 #if IS_ENABLED(CONFIG_HYPERV)
 148         if (test_bit(HYPERV_REENLIGHTENMENT_VECTOR, system_vectors)) {
 149                 seq_printf(p, "%*s: ", prec, "HRE");
 150                 for_each_online_cpu(j)
 151                         seq_printf(p, "%10u ",
 152                                    irq_stats(j)->irq_hv_reenlightenment_count);
 153                 seq_puts(p, "  Hyper-V reenlightenment interrupts\n");
 154         }
 155         if (test_bit(HYPERV_STIMER0_VECTOR, system_vectors)) {
 156                 seq_printf(p, "%*s: ", prec, "HVS");
 157                 for_each_online_cpu(j)
 158                         seq_printf(p, "%10u ",
 159                                    irq_stats(j)->hyperv_stimer0_count);
 160                 seq_puts(p, "  Hyper-V stimer0 interrupts\n");
 161         }
 162 #endif
 163         seq_printf(p, "%*s: %10u\n", prec, "ERR", atomic_read(&irq_err_count));
 164 #if defined(CONFIG_X86_IO_APIC)
 165         seq_printf(p, "%*s: %10u\n", prec, "MIS", atomic_read(&irq_mis_count));
 166 #endif
 167 #ifdef CONFIG_HAVE_KVM
 168         seq_printf(p, "%*s: ", prec, "PIN");
 169         for_each_online_cpu(j)
 170                 seq_printf(p, "%10u ", irq_stats(j)->kvm_posted_intr_ipis);
 171         seq_puts(p, "  Posted-interrupt notification event\n");
 172
 173         seq_printf(p, "%*s: ", prec, "NPI");
 174         for_each_online_cpu(j)
 175                 seq_printf(p, "%10u ",
 176                            irq_stats(j)->kvm_posted_intr_nested_ipis);
 177         seq_puts(p, "  Nested posted-interrupt event\n");
 178
 179         seq_printf(p, "%*s: ", prec, "PIW");
 180         for_each_online_cpu(j)
 181                 seq_printf(p, "%10u ",
 182                            irq_stats(j)->kvm_posted_intr_wakeup_ipis);
 183         seq_puts(p, "  Posted-interrupt wakeup event\n");
 184 #endif
 185         return 0;
 186 }
 187
 188 /*
 189  * /proc/stat helpers
 190  */
 191 u64 arch_irq_stat_cpu(unsigned int cpu)
 192 {
 193         u64 sum = irq_stats(cpu)->__nmi_count;
 194
 195 #ifdef CONFIG_X86_LOCAL_APIC
 196         sum += irq_stats(cpu)->apic_timer_irqs;
 197         sum += irq_stats(cpu)->irq_spurious_count;
 198         sum += irq_stats(cpu)->apic_perf_irqs;
 199         sum += irq_stats(cpu)->apic_irq_work_irqs;
 200         sum += irq_stats(cpu)->icr_read_retry_count;
 201         if (x86_platform_ipi_callback)
 202                 sum += irq_stats(cpu)->x86_platform_ipis;
 203 #endif
 204 #ifdef CONFIG_SMP
 205         sum += irq_stats(cpu)->irq_resched_count;
 206         sum += irq_stats(cpu)->irq_call_count;
 207 #endif
 208 #ifdef CONFIG_X86_THERMAL_VECTOR
 209         sum += irq_stats(cpu)->irq_thermal_count;
 210 #endif
 211 #ifdef CONFIG_X86_MCE_THRESHOLD
 212         sum += irq_stats(cpu)->irq_threshold_count;
 213 #endif
 214 #ifdef CONFIG_X86_MCE
 215         sum += per_cpu(mce_exception_count, cpu);
 216         sum += per_cpu(mce_poll_count, cpu);
 217 #endif
 218         return sum;
 219 }
 220
 221 u64 arch_irq_stat(void)
 222 {
 223         u64 sum = atomic_read(&irq_err_count);
 224         return sum;
 225 }
 226
 227
 228 /*
 229  * do_IRQ handles all normal device IRQ's (the special
 230  * SMP cross-CPU interrupts have their own specific
 231  * handlers).
 232  */
 233 __visible unsigned int __irq_entry do_IRQ(struct pt_regs *regs)
 234 {
 235         struct pt_regs *old_regs = set_irq_regs(regs);
 236         struct irq_desc * desc;
 237         /* high bit used in ret_from_ code  */
 238         unsigned vector = ~regs->orig_ax;
 239
 240         entering_irq();
 241
 242         /* entering_irq() tells RCU that we're not quiescent.  Check it. */
 243         RCU_LOCKDEP_WARN(!rcu_is_watching(), "IRQ failed to wake up RCU");
 244
 245         desc = __this_cpu_read(vector_irq[vector]);
 246
 247         if (!handle_irq(desc, regs)) {
 248                 ack_APIC_irq();
 249
 250                 if (desc != VECTOR_RETRIGGERED && desc != VECTOR_SHUTDOWN) {
 251                         pr_emerg_ratelimited("%s: %d.%d No irq handler for vector\n",
 252                                              __func__, smp_processor_id(),
 253                                              vector);
 254                 } else {
 255                         __this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
 256                 }
 257         }
 258
 259         exiting_irq();
 260
 261         set_irq_regs(old_regs);
 262         return 1;
 263 }
 264
 265 #ifdef CONFIG_X86_LOCAL_APIC
 266 /* Function pointer for generic interrupt vector handling */
 267 void (*x86_platform_ipi_callback)(void) = NULL;
 268 /*
 269  * Handler for X86_PLATFORM_IPI_VECTOR.
 270  */
 271 __visible void __irq_entry smp_x86_platform_ipi(struct pt_regs *regs)
 272 {
 273         struct pt_regs *old_regs = set_irq_regs(regs);
 274
 275         entering_ack_irq();
 276         trace_x86_platform_ipi_entry(X86_PLATFORM_IPI_VECTOR);
 277         inc_irq_stat(x86_platform_ipis);
 278         if (x86_platform_ipi_callback)
 279                 x86_platform_ipi_callback();
 280         trace_x86_platform_ipi_exit(X86_PLATFORM_IPI_VECTOR);
 281         exiting_irq();
 282         set_irq_regs(old_regs);
 283 }
 284 #endif
 285
 286 #ifdef CONFIG_HAVE_KVM
 287 static void dummy_handler(void) {}
 288 static void (*kvm_posted_intr_wakeup_handler)(void) = dummy_handler;
 289
 290 void kvm_set_posted_intr_wakeup_handler(void (*handler)(void))
 291 {
 292         if (handler)
 293                 kvm_posted_intr_wakeup_handler = handler;
 294         else
 295                 kvm_posted_intr_wakeup_handler = dummy_handler;
 296 }
 297 EXPORT_SYMBOL_GPL(kvm_set_posted_intr_wakeup_handler);
 298
 299 /*
 300  * Handler for POSTED_INTERRUPT_VECTOR.
 301  */
 302 __visible void smp_kvm_posted_intr_ipi(struct pt_regs *regs)
 303 {
 304         struct pt_regs *old_regs = set_irq_regs(regs);
 305
 306         entering_ack_irq();
 307         inc_irq_stat(kvm_posted_intr_ipis);
 308         exiting_irq();
 309         set_irq_regs(old_regs);
 310 }
 311
 312 /*
 313  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
 314  */
 315 __visible void smp_kvm_posted_intr_wakeup_ipi(struct pt_regs *regs)
 316 {
 317         struct pt_regs *old_regs = set_irq_regs(regs);
 318
 319         entering_ack_irq();
 320         inc_irq_stat(kvm_posted_intr_wakeup_ipis);
 321         kvm_posted_intr_wakeup_handler();
 322         exiting_irq();
 323         set_irq_regs(old_regs);
 324 }
 325
 326 /*
 327  * Handler for POSTED_INTERRUPT_NESTED_VECTOR.
 328  */
 329 __visible void smp_kvm_posted_intr_nested_ipi(struct pt_regs *regs)
 330 {
 331         struct pt_regs *old_regs = set_irq_regs(regs);
 332
 333         entering_ack_irq();
 334         inc_irq_stat(kvm_posted_intr_nested_ipis);
 335         exiting_irq();
 336         set_irq_regs(old_regs);
 337 }
 338 #endif
 339
 340
 341 #ifdef CONFIG_HOTPLUG_CPU
 342 /* A cpu has been removed from cpu_online_mask.  Reset irq affinities. */
 343 void fixup_irqs(void)
 344 {
 345         unsigned int irr, vector;
 346         struct irq_desc *desc;
 347         struct irq_data *data;
 348         struct irq_chip *chip;
 349
 350         irq_migrate_all_off_this_cpu();
 351
 352         /*
 353          * We can remove mdelay() and then send spuriuous interrupts to
 354          * new cpu targets for all the irqs that were handled previously by
 355          * this cpu. While it works, I have seen spurious interrupt messages
 356          * (nothing wrong but still...).
 357          *
 358          * So for now, retain mdelay(1) and check the IRR and then send those
 359          * interrupts to new targets as this cpu is already offlined...
 360          */
 361         mdelay(1);
 362
 363         /*
 364          * We can walk the vector array of this cpu without holding
 365          * vector_lock because the cpu is already marked !online, so
 366          * nothing else will touch it.
 367          */
 368         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
 369                 if (IS_ERR_OR_NULL(__this_cpu_read(vector_irq[vector])))
 370                         continue;
 371
 372                 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
 373                 if (irr  & (1 << (vector % 32))) {
 374                         desc = __this_cpu_read(vector_irq[vector]);
 375
 376                         raw_spin_lock(&desc->lock);
 377                         data = irq_desc_get_irq_data(desc);
 378                         chip = irq_data_get_irq_chip(data);
 379                         if (chip->irq_retrigger) {
 380                                 chip->irq_retrigger(data);
 381                                 __this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
 382                         }
 383                         raw_spin_unlock(&desc->lock);
 384                 }
 385                 if (__this_cpu_read(vector_irq[vector]) != VECTOR_RETRIGGERED)
 386                         __this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
 387         }
 388 }
 389 #endif