arch/ia64/kernel/crash.c

   1 /*
   2  * arch/ia64/kernel/crash.c
   3  *
   4  * Architecture specific (ia64) functions for kexec based crash dumps.
   5  *
   6  * Created by: Khalid Aziz <khalid.aziz@hp.com>
   7  * Copyright (C) 2005 Hewlett-Packard Development Company, L.P.
   8  * Copyright (C) 2005 Intel Corp        Zou Nan hai <nanhai.zou@intel.com>
   9  *
  10  */
  11 #include <linux/smp.h>
  12 #include <linux/delay.h>
  13 #include <linux/crash_dump.h>
  14 #include <linux/bootmem.h>
  15 #include <linux/kexec.h>
  16 #include <linux/elfcore.h>
  17 #include <linux/sysctl.h>
  18 #include <linux/init.h>
  19 #include <linux/kdebug.h>
  20
  21 #include <asm/mca.h>
  22
  23 int kdump_status[NR_CPUS];
  24 static atomic_t kdump_cpu_frozen;
  25 atomic_t kdump_in_progress;
  26 static int kdump_freeze_monarch;
  27 static int kdump_on_init = 1;
  28 static int kdump_on_fatal_mca = 1;
  29
  30 static inline Elf64_Word
  31 *append_elf_note(Elf64_Word *buf, char *name, unsigned type, void *data,
  32                 size_t data_len)
  33 {
  34         struct elf_note *note = (struct elf_note *)buf;
  35         note->n_namesz = strlen(name) + 1;
  36         note->n_descsz = data_len;
  37         note->n_type   = type;
  38         buf += (sizeof(*note) + 3)/4;
  39         memcpy(buf, name, note->n_namesz);
  40         buf += (note->n_namesz + 3)/4;
  41         memcpy(buf, data, data_len);
  42         buf += (data_len + 3)/4;
  43         return buf;
  44 }
  45
  46 static void
  47 final_note(void *buf)
  48 {
  49         memset(buf, 0, sizeof(struct elf_note));
  50 }
  51
  52 extern void ia64_dump_cpu_regs(void *);
  53
  54 static DEFINE_PER_CPU(struct elf_prstatus, elf_prstatus);
  55
  56 void
  57 crash_save_this_cpu(void)
  58 {
  59         void *buf;
  60         unsigned long cfm, sof, sol;
  61
  62         int cpu = smp_processor_id();
  63         struct elf_prstatus *prstatus = &per_cpu(elf_prstatus, cpu);
  64
  65         elf_greg_t *dst = (elf_greg_t *)&(prstatus->pr_reg);
  66         memset(prstatus, 0, sizeof(*prstatus));
  67         prstatus->pr_pid = current->pid;
  68
  69         ia64_dump_cpu_regs(dst);
  70         cfm = dst[43];
  71         sol = (cfm >> 7) & 0x7f;
  72         sof = cfm & 0x7f;
  73         dst[46] = (unsigned long)ia64_rse_skip_regs((unsigned long *)dst[46],
  74                         sof - sol);
  75
  76         buf = (u64 *) per_cpu_ptr(crash_notes, cpu);
  77         if (!buf)
  78                 return;
  79         buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS, prstatus,
  80                         sizeof(*prstatus));
  81         final_note(buf);
  82 }
  83
  84 #ifdef CONFIG_SMP
  85 static int
  86 kdump_wait_cpu_freeze(void)
  87 {
  88         int cpu_num = num_online_cpus() - 1;
  89         int timeout = 1000;
  90         while(timeout-- > 0) {
  91                 if (atomic_read(&kdump_cpu_frozen) == cpu_num)
  92                         return 0;
  93                 udelay(1000);
  94         }
  95         return 1;
  96 }
  97 #endif
  98
  99 void
 100 machine_crash_shutdown(struct pt_regs *pt)
 101 {
 102         /* This function is only called after the system
 103          * has paniced or is otherwise in a critical state.
 104          * The minimum amount of code to allow a kexec'd kernel
 105          * to run successfully needs to happen here.
 106          *
 107          * In practice this means shooting down the other cpus in
 108          * an SMP system.
 109          */
 110         kexec_disable_iosapic();
 111 #ifdef CONFIG_SMP
 112         /*
 113          * If kdump_on_init is set and an INIT is asserted here, kdump will
 114          * be started again via INIT monarch.
 115          */
 116         local_irq_disable();
 117         ia64_set_psr_mc();      /* mask MCA/INIT */
 118         if (atomic_inc_return(&kdump_in_progress) != 1)
 119                 unw_init_running(kdump_cpu_freeze, NULL);
 120
 121         /*
 122          * Now this cpu is ready for kdump.
 123          * Stop all others by IPI or INIT.  They could receive INIT from
 124          * outside and might be INIT monarch, but only thing they have to
 125          * do is falling into kdump_cpu_freeze().
 126          *
 127          * If an INIT is asserted here:
 128          * - All receivers might be slaves, since some of cpus could already
 129          *   be frozen and INIT might be masked on monarch.  In this case,
 130          *   all slaves will be frozen soon since kdump_in_progress will let
 131          *   them into DIE_INIT_SLAVE_LEAVE.
 132          * - One might be a monarch, but INIT rendezvous will fail since
 133          *   at least this cpu already have INIT masked so it never join
 134          *   to the rendezvous.  In this case, all slaves and monarch will
 135          *   be frozen soon with no wait since the INIT rendezvous is skipped
 136          *   by kdump_in_progress.
 137          */
 138         kdump_smp_send_stop();
 139         /* not all cpu response to IPI, send INIT to freeze them */
 140         if (kdump_wait_cpu_freeze()) {
 141                 kdump_smp_send_init();
 142                 /* wait again, don't go ahead if possible */
 143                 kdump_wait_cpu_freeze();
 144         }
 145 #endif
 146 }
 147
 148 static void
 149 machine_kdump_on_init(void)
 150 {
 151         crash_save_vmcoreinfo();
 152         local_irq_disable();
 153         kexec_disable_iosapic();
 154         machine_kexec(ia64_kimage);
 155 }
 156
 157 void
 158 kdump_cpu_freeze(struct unw_frame_info *info, void *arg)
 159 {
 160         int cpuid;
 161
 162         local_irq_disable();
 163         cpuid = smp_processor_id();
 164         crash_save_this_cpu();
 165         current->thread.ksp = (__u64)info->sw - 16;
 166
 167         ia64_set_psr_mc();      /* mask MCA/INIT and stop reentrance */
 168
 169         atomic_inc(&kdump_cpu_frozen);
 170         kdump_status[cpuid] = 1;
 171         mb();
 172         for (;;)
 173                 cpu_relax();
 174 }
 175
 176 static int
 177 kdump_init_notifier(struct notifier_block *self, unsigned long val, void *data)
 178 {
 179         struct ia64_mca_notify_die *nd;
 180         struct die_args *args = data;
 181
 182         if (atomic_read(&kdump_in_progress)) {
 183                 switch (val) {
 184                 case DIE_INIT_MONARCH_LEAVE:
 185                         if (!kdump_freeze_monarch)
 186                                 break;
 187                         /* fall through */
 188                 case DIE_INIT_SLAVE_LEAVE:
 189                 case DIE_INIT_MONARCH_ENTER:
 190                 case DIE_MCA_RENDZVOUS_LEAVE:
 191                         unw_init_running(kdump_cpu_freeze, NULL);
 192                         break;
 193                 }
 194         }
 195
 196         if (!kdump_on_init && !kdump_on_fatal_mca)
 197                 return NOTIFY_DONE;
 198
 199         if (!ia64_kimage) {
 200                 if (val == DIE_INIT_MONARCH_LEAVE)
 201                         ia64_mca_printk(KERN_NOTICE
 202                                         "%s: kdump not configured\n",
 203                                         __func__);
 204                 return NOTIFY_DONE;
 205         }
 206
 207         if (val != DIE_INIT_MONARCH_LEAVE &&
 208             val != DIE_INIT_MONARCH_PROCESS &&
 209             val != DIE_MCA_MONARCH_LEAVE)
 210                 return NOTIFY_DONE;
 211
 212         nd = (struct ia64_mca_notify_die *)args->err;
 213
 214         switch (val) {
 215         case DIE_INIT_MONARCH_PROCESS:
 216                 /* Reason code 1 means machine check rendezvous*/
 217                 if (kdump_on_init && (nd->sos->rv_rc != 1)) {
 218                         if (atomic_inc_return(&kdump_in_progress) != 1)
 219                                 kdump_freeze_monarch = 1;
 220                 }
 221                 break;
 222         case DIE_INIT_MONARCH_LEAVE:
 223                 /* Reason code 1 means machine check rendezvous*/
 224                 if (kdump_on_init && (nd->sos->rv_rc != 1))
 225                         machine_kdump_on_init();
 226                 break;
 227         case DIE_MCA_MONARCH_LEAVE:
 228                 /* *(nd->data) indicate if MCA is recoverable */
 229                 if (kdump_on_fatal_mca && !(*(nd->data))) {
 230                         if (atomic_inc_return(&kdump_in_progress) == 1)
 231                                 machine_kdump_on_init();
 232                         /* We got fatal MCA while kdump!? No way!! */
 233                 }
 234                 break;
 235         }
 236         return NOTIFY_DONE;
 237 }
 238
 239 #ifdef CONFIG_SYSCTL
 240 static ctl_table kdump_ctl_table[] = {
 241         {
 242                 .ctl_name = CTL_UNNUMBERED,
 243                 .procname = "kdump_on_init",
 244                 .data = &kdump_on_init,
 245                 .maxlen = sizeof(int),
 246                 .mode = 0644,
 247                 .proc_handler = &proc_dointvec,
 248         },
 249         {
 250                 .ctl_name = CTL_UNNUMBERED,
 251                 .procname = "kdump_on_fatal_mca",
 252                 .data = &kdump_on_fatal_mca,
 253                 .maxlen = sizeof(int),
 254                 .mode = 0644,
 255                 .proc_handler = &proc_dointvec,
 256         },
 257         { .ctl_name = 0 }
 258 };
 259
 260 static ctl_table sys_table[] = {
 261         {
 262           .ctl_name = CTL_KERN,
 263           .procname = "kernel",
 264           .mode = 0555,
 265           .child = kdump_ctl_table,
 266         },
 267         { .ctl_name = 0 }
 268 };
 269 #endif
 270
 271 static int
 272 machine_crash_setup(void)
 273 {
 274         /* be notified before default_monarch_init_process */
 275         static struct notifier_block kdump_init_notifier_nb = {
 276                 .notifier_call = kdump_init_notifier,
 277                 .priority = 1,
 278         };
 279         int ret;
 280         if((ret = register_die_notifier(&kdump_init_notifier_nb)) != 0)
 281                 return ret;
 282 #ifdef CONFIG_SYSCTL
 283         register_sysctl_table(sys_table);
 284 #endif
 285         return 0;
 286 }
 287
 288 __initcall(machine_crash_setup);
 289