arch/x86/kernel/cpu/mcheck/mce-inject.c

   1 /*
   2  * Machine check injection support.
   3  * Copyright 2008 Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License
   7  * as published by the Free Software Foundation; version 2
   8  * of the License.
   9  *
  10  * Authors:
  11  * Andi Kleen
  12  * Ying Huang
  13  */
  14 #include <linux/uaccess.h>
  15 #include <linux/module.h>
  16 #include <linux/timer.h>
  17 #include <linux/kernel.h>
  18 #include <linux/string.h>
  19 #include <linux/fs.h>
  20 #include <linux/smp.h>
  21 #include <linux/notifier.h>
  22 #include <linux/kdebug.h>
  23 #include <linux/cpu.h>
  24 #include <linux/sched.h>
  25 #include <asm/mce.h>
  26 #include <asm/apic.h>
  27
  28 /* Update fake mce registers on current CPU. */
  29 static void inject_mce(struct mce *m)
  30 {
  31         struct mce *i = &per_cpu(injectm, m->extcpu);
  32
  33         /* Make sure noone reads partially written injectm */
  34         i->finished = 0;
  35         mb();
  36         m->finished = 0;
  37         /* First set the fields after finished */
  38         i->extcpu = m->extcpu;
  39         mb();
  40         /* Now write record in order, finished last (except above) */
  41         memcpy(i, m, sizeof(struct mce));
  42         /* Finally activate it */
  43         mb();
  44         i->finished = 1;
  45 }
  46
  47 static void raise_poll(struct mce *m)
  48 {
  49         unsigned long flags;
  50         mce_banks_t b;
  51
  52         memset(&b, 0xff, sizeof(mce_banks_t));
  53         local_irq_save(flags);
  54         machine_check_poll(0, &b);
  55         local_irq_restore(flags);
  56         m->finished = 0;
  57 }
  58
  59 static void raise_exception(struct mce *m, struct pt_regs *pregs)
  60 {
  61         struct pt_regs regs;
  62         unsigned long flags;
  63
  64         if (!pregs) {
  65                 memset(&regs, 0, sizeof(struct pt_regs));
  66                 regs.ip = m->ip;
  67                 regs.cs = m->cs;
  68                 pregs = &regs;
  69         }
  70         /* in mcheck exeception handler, irq will be disabled */
  71         local_irq_save(flags);
  72         do_machine_check(pregs, 0);
  73         local_irq_restore(flags);
  74         m->finished = 0;
  75 }
  76
  77 static cpumask_t mce_inject_cpumask;
  78
  79 static int mce_raise_notify(struct notifier_block *self,
  80                             unsigned long val, void *data)
  81 {
  82         struct die_args *args = (struct die_args *)data;
  83         int cpu = smp_processor_id();
  84         struct mce *m = &__get_cpu_var(injectm);
  85         if (val != DIE_NMI_IPI || !cpu_isset(cpu, mce_inject_cpumask))
  86                 return NOTIFY_DONE;
  87         cpu_clear(cpu, mce_inject_cpumask);
  88         if (m->inject_flags & MCJ_EXCEPTION)
  89                 raise_exception(m, args->regs);
  90         else if (m->status)
  91                 raise_poll(m);
  92         return NOTIFY_STOP;
  93 }
  94
  95 static struct notifier_block mce_raise_nb = {
  96         .notifier_call = mce_raise_notify,
  97         .priority = 1000,
  98 };
  99
 100 /* Inject mce on current CPU */
 101 static int raise_local(void)
 102 {
 103         struct mce *m = &__get_cpu_var(injectm);
 104         int context = MCJ_CTX(m->inject_flags);
 105         int ret = 0;
 106         int cpu = m->extcpu;
 107
 108         if (m->inject_flags & MCJ_EXCEPTION) {
 109                 printk(KERN_INFO "Triggering MCE exception on CPU %d\n", cpu);
 110                 switch (context) {
 111                 case MCJ_CTX_IRQ:
 112                         /*
 113                          * Could do more to fake interrupts like
 114                          * calling irq_enter, but the necessary
 115                          * machinery isn't exported currently.
 116                          */
 117                         /*FALL THROUGH*/
 118                 case MCJ_CTX_PROCESS:
 119                         raise_exception(m, NULL);
 120                         break;
 121                 default:
 122                         printk(KERN_INFO "Invalid MCE context\n");
 123                         ret = -EINVAL;
 124                 }
 125                 printk(KERN_INFO "MCE exception done on CPU %d\n", cpu);
 126         } else if (m->status) {
 127                 printk(KERN_INFO "Starting machine check poll CPU %d\n", cpu);
 128                 raise_poll(m);
 129                 mce_notify_irq();
 130                 printk(KERN_INFO "Machine check poll done on CPU %d\n", cpu);
 131         } else
 132                 m->finished = 0;
 133
 134         return ret;
 135 }
 136
 137 static void raise_mce(struct mce *m)
 138 {
 139         int context = MCJ_CTX(m->inject_flags);
 140
 141         inject_mce(m);
 142
 143         if (context == MCJ_CTX_RANDOM)
 144                 return;
 145
 146 #ifdef CONFIG_X86_LOCAL_APIC
 147         if (m->inject_flags & MCJ_NMI_BROADCAST) {
 148                 unsigned long start;
 149                 int cpu;
 150                 get_online_cpus();
 151                 mce_inject_cpumask = cpu_online_map;
 152                 cpu_clear(get_cpu(), mce_inject_cpumask);
 153                 for_each_online_cpu(cpu) {
 154                         struct mce *mcpu = &per_cpu(injectm, cpu);
 155                         if (!mcpu->finished ||
 156                             MCJ_CTX(mcpu->inject_flags) != MCJ_CTX_RANDOM)
 157                                 cpu_clear(cpu, mce_inject_cpumask);
 158                 }
 159                 if (!cpus_empty(mce_inject_cpumask))
 160                         apic->send_IPI_mask(&mce_inject_cpumask, NMI_VECTOR);
 161                 start = jiffies;
 162                 while (!cpus_empty(mce_inject_cpumask)) {
 163                         if (!time_before(jiffies, start + 2*HZ)) {
 164                                 printk(KERN_ERR
 165                                 "Timeout waiting for mce inject NMI %lx\n",
 166                                         *cpus_addr(mce_inject_cpumask));
 167                                 break;
 168                         }
 169                         cpu_relax();
 170                 }
 171                 raise_local();
 172                 put_cpu();
 173                 put_online_cpus();
 174         } else
 175 #endif
 176                 raise_local();
 177 }
 178
 179 /* Error injection interface */
 180 static ssize_t mce_write(struct file *filp, const char __user *ubuf,
 181                          size_t usize, loff_t *off)
 182 {
 183         struct mce m;
 184
 185         if (!capable(CAP_SYS_ADMIN))
 186                 return -EPERM;
 187         /*
 188          * There are some cases where real MSR reads could slip
 189          * through.
 190          */
 191         if (!boot_cpu_has(X86_FEATURE_MCE) || !boot_cpu_has(X86_FEATURE_MCA))
 192                 return -EIO;
 193
 194         if ((unsigned long)usize > sizeof(struct mce))
 195                 usize = sizeof(struct mce);
 196         if (copy_from_user(&m, ubuf, usize))
 197                 return -EFAULT;
 198
 199         if (m.extcpu >= num_possible_cpus() || !cpu_online(m.extcpu))
 200                 return -EINVAL;
 201
 202         /*
 203          * Need to give user space some time to set everything up,
 204          * so do it a jiffie or two later everywhere.
 205          */
 206         schedule_timeout(2);
 207         raise_mce(&m);
 208         return usize;
 209 }
 210
 211 static int inject_init(void)
 212 {
 213         printk(KERN_INFO "Machine check injector initialized\n");
 214         mce_chrdev_ops.write = mce_write;
 215         register_die_notifier(&mce_raise_nb);
 216         return 0;
 217 }
 218
 219 module_init(inject_init);
 220 /*
 221  * Cannot tolerate unloading currently because we cannot
 222  * guarantee all openers of mce_chrdev will get a reference to us.
 223  */
 224 MODULE_LICENSE("GPL");