arch/x86/kernel/irq_32.c

   1 /*
   2  *      Copyright (C) 1992, 1998 Linus Torvalds, Ingo Molnar
   3  *
   4  * This file contains the lowest level x86-specific interrupt
   5  * entry, irq-stacks and irq statistics code. All the remaining
   6  * irq logic is done by the generic kernel/irq/ code and
   7  * by the x86-specific irq controller code. (e.g. i8259.c and
   8  * io_apic.c.)
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/seq_file.h>
  13 #include <linux/interrupt.h>
  14 #include <linux/kernel_stat.h>
  15 #include <linux/notifier.h>
  16 #include <linux/cpu.h>
  17 #include <linux/delay.h>
  18
  19 #include <asm/apic.h>
  20 #include <asm/uaccess.h>
  21
  22 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
  23 EXPORT_PER_CPU_SYMBOL(irq_stat);
  24
  25 DEFINE_PER_CPU(struct pt_regs *, irq_regs);
  26 EXPORT_PER_CPU_SYMBOL(irq_regs);
  27
  28 /*
  29  * 'what should we do if we get a hw irq event on an illegal vector'.
  30  * each architecture has to answer this themselves.
  31  */
  32 void ack_bad_irq(unsigned int irq)
  33 {
  34         printk(KERN_ERR "unexpected IRQ trap at vector %02x\n", irq);
  35
  36 #ifdef CONFIG_X86_LOCAL_APIC
  37         /*
  38          * Currently unexpected vectors happen only on SMP and APIC.
  39          * We _must_ ack these because every local APIC has only N
  40          * irq slots per priority level, and a 'hanging, unacked' IRQ
  41          * holds up an irq slot - in excessive cases (when multiple
  42          * unexpected vectors occur) that might lock up the APIC
  43          * completely.
  44          * But only ack when the APIC is enabled -AK
  45          */
  46         if (cpu_has_apic)
  47                 ack_APIC_irq();
  48 #endif
  49 }
  50
  51 #ifdef CONFIG_4KSTACKS
  52 /*
  53  * per-CPU IRQ handling contexts (thread information and stack)
  54  */
  55 union irq_ctx {
  56         struct thread_info      tinfo;
  57         u32                     stack[THREAD_SIZE/sizeof(u32)];
  58 };
  59
  60 static union irq_ctx *hardirq_ctx[NR_CPUS] __read_mostly;
  61 static union irq_ctx *softirq_ctx[NR_CPUS] __read_mostly;
  62 #endif
  63
  64 static void stack_overflow(void)
  65 {
  66         printk("low stack detected by irq handler\n");
  67         dump_stack();
  68 }
  69
  70 static inline void call_on_stack2(void *func, void *stack,
  71                            unsigned long arg1, unsigned long arg2)
  72 {
  73         unsigned long bx;
  74         asm volatile(
  75                         "       xchgl  %%ebx,%%esp    \n"
  76                         "       call   *%%edi         \n"
  77                         "       movl   %%ebx,%%esp    \n"
  78                         : "=a" (arg1), "=d" (arg2), "=b" (bx)
  79                         :  "0" (arg1),   "1" (arg2),  "2" (stack),
  80                            "D" (func)
  81                         : "memory", "cc", "ecx");
  82 }
  83
  84 /*
  85  * do_IRQ handles all normal device IRQ's (the special
  86  * SMP cross-CPU interrupts have their own specific
  87  * handlers).
  88  */
  89 unsigned int do_IRQ(struct pt_regs *regs)
  90 {
  91         struct pt_regs *old_regs;
  92         /* high bit used in ret_from_ code */
  93         int irq = ~regs->orig_ax;
  94         struct irq_desc *desc = irq_desc + irq;
  95 #ifdef CONFIG_4KSTACKS
  96         union irq_ctx *curctx, *irqctx;
  97         u32 *isp;
  98 #endif
  99         int overflow = 0;
 100
 101         if (unlikely((unsigned)irq >= NR_IRQS)) {
 102                 printk(KERN_EMERG "%s: cannot handle IRQ %d\n",
 103                                         __func__, irq);
 104                 BUG();
 105         }
 106
 107         old_regs = set_irq_regs(regs);
 108         irq_enter();
 109 #ifdef CONFIG_DEBUG_STACKOVERFLOW
 110         /* Debugging check for stack overflow: is there less than 1KB free? */
 111         {
 112                 long sp;
 113
 114                 __asm__ __volatile__("andl %%esp,%0" :
 115                                         "=r" (sp) : "0" (THREAD_SIZE - 1));
 116                 if (unlikely(sp < (sizeof(struct thread_info) + STACK_WARN)))
 117                         overflow = 1;
 118         }
 119 #endif
 120
 121 #ifdef CONFIG_4KSTACKS
 122
 123         curctx = (union irq_ctx *) current_thread_info();
 124         irqctx = hardirq_ctx[smp_processor_id()];
 125
 126         /*
 127          * this is where we switch to the IRQ stack. However, if we are
 128          * already using the IRQ stack (because we interrupted a hardirq
 129          * handler) we can't do that and just have to keep using the
 130          * current stack (which is the irq stack already after all)
 131          */
 132         if (curctx != irqctx) {
 133                 /* build the stack frame on the IRQ stack */
 134                 isp = (u32*) ((char*)irqctx + sizeof(*irqctx));
 135                 irqctx->tinfo.task = curctx->tinfo.task;
 136                 irqctx->tinfo.previous_esp = current_stack_pointer;
 137
 138                 /*
 139                  * Copy the softirq bits in preempt_count so that the
 140                  * softirq checks work in the hardirq context.
 141                  */
 142                 irqctx->tinfo.preempt_count =
 143                         (irqctx->tinfo.preempt_count & ~SOFTIRQ_MASK) |
 144                         (curctx->tinfo.preempt_count & SOFTIRQ_MASK);
 145
 146                 /* Execute warning on interrupt stack */
 147                 if (unlikely(overflow))
 148                         call_on_stack2(stack_overflow, isp, 0, 0);
 149
 150                 call_on_stack2(desc->handle_irq, isp, irq, (unsigned long)desc);
 151         } else
 152 #endif
 153         {
 154                 /* AK: Slightly bogus here */
 155                 if (overflow)
 156                         stack_overflow();
 157                 desc->handle_irq(irq, desc);
 158         }
 159
 160         irq_exit();
 161         set_irq_regs(old_regs);
 162         return 1;
 163 }
 164
 165 #ifdef CONFIG_4KSTACKS
 166
 167 static char softirq_stack[NR_CPUS * THREAD_SIZE]
 168                 __attribute__((__section__(".bss.page_aligned")));
 169
 170 static char hardirq_stack[NR_CPUS * THREAD_SIZE]
 171                 __attribute__((__section__(".bss.page_aligned")));
 172
 173 /*
 174  * allocate per-cpu stacks for hardirq and for softirq processing
 175  */
 176 void irq_ctx_init(int cpu)
 177 {
 178         union irq_ctx *irqctx;
 179
 180         if (hardirq_ctx[cpu])
 181                 return;
 182
 183         irqctx = (union irq_ctx*) &hardirq_stack[cpu*THREAD_SIZE];
 184         irqctx->tinfo.task              = NULL;
 185         irqctx->tinfo.exec_domain       = NULL;
 186         irqctx->tinfo.cpu               = cpu;
 187         irqctx->tinfo.preempt_count     = HARDIRQ_OFFSET;
 188         irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
 189
 190         hardirq_ctx[cpu] = irqctx;
 191
 192         irqctx = (union irq_ctx*) &softirq_stack[cpu*THREAD_SIZE];
 193         irqctx->tinfo.task              = NULL;
 194         irqctx->tinfo.exec_domain       = NULL;
 195         irqctx->tinfo.cpu               = cpu;
 196         irqctx->tinfo.preempt_count     = 0;
 197         irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
 198
 199         softirq_ctx[cpu] = irqctx;
 200
 201         printk("CPU %u irqstacks, hard=%p soft=%p\n",
 202                 cpu,hardirq_ctx[cpu],softirq_ctx[cpu]);
 203 }
 204
 205 void irq_ctx_exit(int cpu)
 206 {
 207         hardirq_ctx[cpu] = NULL;
 208 }
 209
 210 asmlinkage void do_softirq(void)
 211 {
 212         unsigned long flags;
 213         struct thread_info *curctx;
 214         union irq_ctx *irqctx;
 215         u32 *isp;
 216
 217         if (in_interrupt())
 218                 return;
 219
 220         local_irq_save(flags);
 221
 222         if (local_softirq_pending()) {
 223                 curctx = current_thread_info();
 224                 irqctx = softirq_ctx[smp_processor_id()];
 225                 irqctx->tinfo.task = curctx->task;
 226                 irqctx->tinfo.previous_esp = current_stack_pointer;
 227
 228                 /* build the stack frame on the softirq stack */
 229                 isp = (u32*) ((char*)irqctx + sizeof(*irqctx));
 230
 231                 asm volatile(
 232                         "       xchgl   %%ebx,%%esp     \n"
 233                         "       call    __do_softirq    \n"
 234                         "       movl    %%ebx,%%esp     \n"
 235                         : "=b"(isp)
 236                         : "0"(isp)
 237                         : "memory", "cc", "edx", "ecx", "eax"
 238                 );
 239                 /*
 240                  * Shouldnt happen, we returned above if in_interrupt():
 241                  */
 242                 WARN_ON_ONCE(softirq_count());
 243         }
 244
 245         local_irq_restore(flags);
 246 }
 247 #endif
 248
 249 /*
 250  * Interrupt statistics:
 251  */
 252
 253 atomic_t irq_err_count;
 254
 255 /*
 256  * /proc/interrupts printing:
 257  */
 258
 259 int show_interrupts(struct seq_file *p, void *v)
 260 {
 261         int i = *(loff_t *) v, j;
 262         struct irqaction * action;
 263         unsigned long flags;
 264
 265         if (i == 0) {
 266                 seq_printf(p, "           ");
 267                 for_each_online_cpu(j)
 268                         seq_printf(p, "CPU%-8d",j);
 269                 seq_putc(p, '\n');
 270         }
 271
 272         if (i < NR_IRQS) {
 273                 unsigned any_count = 0;
 274
 275                 spin_lock_irqsave(&irq_desc[i].lock, flags);
 276 #ifndef CONFIG_SMP
 277                 any_count = kstat_irqs(i);
 278 #else
 279                 for_each_online_cpu(j)
 280                         any_count |= kstat_cpu(j).irqs[i];
 281 #endif
 282                 action = irq_desc[i].action;
 283                 if (!action && !any_count)
 284                         goto skip;
 285                 seq_printf(p, "%3d: ",i);
 286 #ifndef CONFIG_SMP
 287                 seq_printf(p, "%10u ", kstat_irqs(i));
 288 #else
 289                 for_each_online_cpu(j)
 290                         seq_printf(p, "%10u ", kstat_cpu(j).irqs[i]);
 291 #endif
 292                 seq_printf(p, " %8s", irq_desc[i].chip->name);
 293                 seq_printf(p, "-%-8s", irq_desc[i].name);
 294
 295                 if (action) {
 296                         seq_printf(p, "  %s", action->name);
 297                         while ((action = action->next) != NULL)
 298                                 seq_printf(p, ", %s", action->name);
 299                 }
 300
 301                 seq_putc(p, '\n');
 302 skip:
 303                 spin_unlock_irqrestore(&irq_desc[i].lock, flags);
 304         } else if (i == NR_IRQS) {
 305                 seq_printf(p, "NMI: ");
 306                 for_each_online_cpu(j)
 307                         seq_printf(p, "%10u ", nmi_count(j));
 308                 seq_printf(p, "  Non-maskable interrupts\n");
 309 #ifdef CONFIG_X86_LOCAL_APIC
 310                 seq_printf(p, "LOC: ");
 311                 for_each_online_cpu(j)
 312                         seq_printf(p, "%10u ",
 313                                 per_cpu(irq_stat,j).apic_timer_irqs);
 314                 seq_printf(p, "  Local timer interrupts\n");
 315 #endif
 316 #ifdef CONFIG_SMP
 317                 seq_printf(p, "RES: ");
 318                 for_each_online_cpu(j)
 319                         seq_printf(p, "%10u ",
 320                                 per_cpu(irq_stat,j).irq_resched_count);
 321                 seq_printf(p, "  Rescheduling interrupts\n");
 322                 seq_printf(p, "CAL: ");
 323                 for_each_online_cpu(j)
 324                         seq_printf(p, "%10u ",
 325                                 per_cpu(irq_stat,j).irq_call_count);
 326                 seq_printf(p, "  function call interrupts\n");
 327                 seq_printf(p, "TLB: ");
 328                 for_each_online_cpu(j)
 329                         seq_printf(p, "%10u ",
 330                                 per_cpu(irq_stat,j).irq_tlb_count);
 331                 seq_printf(p, "  TLB shootdowns\n");
 332 #endif
 333                 seq_printf(p, "TRM: ");
 334                 for_each_online_cpu(j)
 335                         seq_printf(p, "%10u ",
 336                                 per_cpu(irq_stat,j).irq_thermal_count);
 337                 seq_printf(p, "  Thermal event interrupts\n");
 338                 seq_printf(p, "SPU: ");
 339                 for_each_online_cpu(j)
 340                         seq_printf(p, "%10u ",
 341                                 per_cpu(irq_stat,j).irq_spurious_count);
 342                 seq_printf(p, "  Spurious interrupts\n");
 343                 seq_printf(p, "ERR: %10u\n", atomic_read(&irq_err_count));
 344 #if defined(CONFIG_X86_IO_APIC)
 345                 seq_printf(p, "MIS: %10u\n", atomic_read(&irq_mis_count));
 346 #endif
 347         }
 348         return 0;
 349 }
 350
 351 #ifdef CONFIG_HOTPLUG_CPU
 352 #include <mach_apic.h>
 353
 354 void fixup_irqs(cpumask_t map)
 355 {
 356         unsigned int irq;
 357         static int warned;
 358
 359         for (irq = 0; irq < NR_IRQS; irq++) {
 360                 cpumask_t mask;
 361                 if (irq == 2)
 362                         continue;
 363
 364                 cpus_and(mask, irq_desc[irq].affinity, map);
 365                 if (any_online_cpu(mask) == NR_CPUS) {
 366                         printk("Breaking affinity for irq %i\n", irq);
 367                         mask = map;
 368                 }
 369                 if (irq_desc[irq].chip->set_affinity)
 370                         irq_desc[irq].chip->set_affinity(irq, mask);
 371                 else if (irq_desc[irq].action && !(warned++))
 372                         printk("Cannot set affinity for irq %i\n", irq);
 373         }
 374
 375 #if 0
 376         barrier();
 377         /* Ingo Molnar says: "after the IO-APIC masks have been redirected
 378            [note the nop - the interrupt-enable boundary on x86 is two
 379            instructions from sti] - to flush out pending hardirqs and
 380            IPIs. After this point nothing is supposed to reach this CPU." */
 381         __asm__ __volatile__("sti; nop; cli");
 382         barrier();
 383 #else
 384         /* That doesn't seem sufficient.  Give it 1ms. */
 385         local_irq_enable();
 386         mdelay(1);
 387         local_irq_disable();
 388 #endif
 389 }
 390 #endif
 391