Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / arch / ia64 / kernel / crash.c
diff --git a/kernel/arch/ia64/kernel/crash.c b/kernel/arch/ia64/kernel/crash.c
new file mode 100644 (file)
index 0000000..2955f35
--- /dev/null
@@ -0,0 +1,286 @@
+/*
+ * arch/ia64/kernel/crash.c
+ *
+ * Architecture specific (ia64) functions for kexec based crash dumps.
+ *
+ * Created by: Khalid Aziz <khalid.aziz@hp.com>
+ * Copyright (C) 2005 Hewlett-Packard Development Company, L.P.
+ * Copyright (C) 2005 Intel Corp       Zou Nan hai <nanhai.zou@intel.com>
+ *
+ */
+#include <linux/smp.h>
+#include <linux/delay.h>
+#include <linux/crash_dump.h>
+#include <linux/bootmem.h>
+#include <linux/kexec.h>
+#include <linux/elfcore.h>
+#include <linux/sysctl.h>
+#include <linux/init.h>
+#include <linux/kdebug.h>
+
+#include <asm/mca.h>
+
+int kdump_status[NR_CPUS];
+static atomic_t kdump_cpu_frozen;
+atomic_t kdump_in_progress;
+static int kdump_freeze_monarch;
+static int kdump_on_init = 1;
+static int kdump_on_fatal_mca = 1;
+
+static inline Elf64_Word
+*append_elf_note(Elf64_Word *buf, char *name, unsigned type, void *data,
+               size_t data_len)
+{
+       struct elf_note *note = (struct elf_note *)buf;
+       note->n_namesz = strlen(name) + 1;
+       note->n_descsz = data_len;
+       note->n_type   = type;
+       buf += (sizeof(*note) + 3)/4;
+       memcpy(buf, name, note->n_namesz);
+       buf += (note->n_namesz + 3)/4;
+       memcpy(buf, data, data_len);
+       buf += (data_len + 3)/4;
+       return buf;
+}
+
+static void
+final_note(void *buf)
+{
+       memset(buf, 0, sizeof(struct elf_note));
+}
+
+extern void ia64_dump_cpu_regs(void *);
+
+static DEFINE_PER_CPU(struct elf_prstatus, elf_prstatus);
+
+void
+crash_save_this_cpu(void)
+{
+       void *buf;
+       unsigned long cfm, sof, sol;
+
+       int cpu = smp_processor_id();
+       struct elf_prstatus *prstatus = &per_cpu(elf_prstatus, cpu);
+
+       elf_greg_t *dst = (elf_greg_t *)&(prstatus->pr_reg);
+       memset(prstatus, 0, sizeof(*prstatus));
+       prstatus->pr_pid = current->pid;
+
+       ia64_dump_cpu_regs(dst);
+       cfm = dst[43];
+       sol = (cfm >> 7) & 0x7f;
+       sof = cfm & 0x7f;
+       dst[46] = (unsigned long)ia64_rse_skip_regs((unsigned long *)dst[46],
+                       sof - sol);
+
+       buf = (u64 *) per_cpu_ptr(crash_notes, cpu);
+       if (!buf)
+               return;
+       buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS, prstatus,
+                       sizeof(*prstatus));
+       final_note(buf);
+}
+
+#ifdef CONFIG_SMP
+static int
+kdump_wait_cpu_freeze(void)
+{
+       int cpu_num = num_online_cpus() - 1;
+       int timeout = 1000;
+       while(timeout-- > 0) {
+               if (atomic_read(&kdump_cpu_frozen) == cpu_num)
+                       return 0;
+               udelay(1000);
+       }
+       return 1;
+}
+#endif
+
+void
+machine_crash_shutdown(struct pt_regs *pt)
+{
+       /* This function is only called after the system
+        * has paniced or is otherwise in a critical state.
+        * The minimum amount of code to allow a kexec'd kernel
+        * to run successfully needs to happen here.
+        *
+        * In practice this means shooting down the other cpus in
+        * an SMP system.
+        */
+       kexec_disable_iosapic();
+#ifdef CONFIG_SMP
+       /*
+        * If kdump_on_init is set and an INIT is asserted here, kdump will
+        * be started again via INIT monarch.
+        */
+       local_irq_disable();
+       ia64_set_psr_mc();      /* mask MCA/INIT */
+       if (atomic_inc_return(&kdump_in_progress) != 1)
+               unw_init_running(kdump_cpu_freeze, NULL);
+
+       /*
+        * Now this cpu is ready for kdump.
+        * Stop all others by IPI or INIT.  They could receive INIT from
+        * outside and might be INIT monarch, but only thing they have to
+        * do is falling into kdump_cpu_freeze().
+        *
+        * If an INIT is asserted here:
+        * - All receivers might be slaves, since some of cpus could already
+        *   be frozen and INIT might be masked on monarch.  In this case,
+        *   all slaves will be frozen soon since kdump_in_progress will let
+        *   them into DIE_INIT_SLAVE_LEAVE.
+        * - One might be a monarch, but INIT rendezvous will fail since
+        *   at least this cpu already have INIT masked so it never join
+        *   to the rendezvous.  In this case, all slaves and monarch will
+        *   be frozen soon with no wait since the INIT rendezvous is skipped
+        *   by kdump_in_progress.
+        */
+       kdump_smp_send_stop();
+       /* not all cpu response to IPI, send INIT to freeze them */
+       if (kdump_wait_cpu_freeze()) {
+               kdump_smp_send_init();
+               /* wait again, don't go ahead if possible */
+               kdump_wait_cpu_freeze();
+       }
+#endif
+}
+
+static void
+machine_kdump_on_init(void)
+{
+       crash_save_vmcoreinfo();
+       local_irq_disable();
+       kexec_disable_iosapic();
+       machine_kexec(ia64_kimage);
+}
+
+void
+kdump_cpu_freeze(struct unw_frame_info *info, void *arg)
+{
+       int cpuid;
+
+       local_irq_disable();
+       cpuid = smp_processor_id();
+       crash_save_this_cpu();
+       current->thread.ksp = (__u64)info->sw - 16;
+
+       ia64_set_psr_mc();      /* mask MCA/INIT and stop reentrance */
+
+       atomic_inc(&kdump_cpu_frozen);
+       kdump_status[cpuid] = 1;
+       mb();
+       for (;;)
+               cpu_relax();
+}
+
+static int
+kdump_init_notifier(struct notifier_block *self, unsigned long val, void *data)
+{
+       struct ia64_mca_notify_die *nd;
+       struct die_args *args = data;
+
+       if (atomic_read(&kdump_in_progress)) {
+               switch (val) {
+               case DIE_INIT_MONARCH_LEAVE:
+                       if (!kdump_freeze_monarch)
+                               break;
+                       /* fall through */
+               case DIE_INIT_SLAVE_LEAVE:
+               case DIE_INIT_MONARCH_ENTER:
+               case DIE_MCA_RENDZVOUS_LEAVE:
+                       unw_init_running(kdump_cpu_freeze, NULL);
+                       break;
+               }
+       }
+
+       if (!kdump_on_init && !kdump_on_fatal_mca)
+               return NOTIFY_DONE;
+
+       if (!ia64_kimage) {
+               if (val == DIE_INIT_MONARCH_LEAVE)
+                       ia64_mca_printk(KERN_NOTICE
+                                       "%s: kdump not configured\n",
+                                       __func__);
+               return NOTIFY_DONE;
+       }
+
+       if (val != DIE_INIT_MONARCH_LEAVE &&
+           val != DIE_INIT_MONARCH_PROCESS &&
+           val != DIE_MCA_MONARCH_LEAVE)
+               return NOTIFY_DONE;
+
+       nd = (struct ia64_mca_notify_die *)args->err;
+
+       switch (val) {
+       case DIE_INIT_MONARCH_PROCESS:
+               /* Reason code 1 means machine check rendezvous*/
+               if (kdump_on_init && (nd->sos->rv_rc != 1)) {
+                       if (atomic_inc_return(&kdump_in_progress) != 1)
+                               kdump_freeze_monarch = 1;
+               }
+               break;
+       case DIE_INIT_MONARCH_LEAVE:
+               /* Reason code 1 means machine check rendezvous*/
+               if (kdump_on_init && (nd->sos->rv_rc != 1))
+                       machine_kdump_on_init();
+               break;
+       case DIE_MCA_MONARCH_LEAVE:
+               /* *(nd->data) indicate if MCA is recoverable */
+               if (kdump_on_fatal_mca && !(*(nd->data))) {
+                       if (atomic_inc_return(&kdump_in_progress) == 1)
+                               machine_kdump_on_init();
+                       /* We got fatal MCA while kdump!? No way!! */
+               }
+               break;
+       }
+       return NOTIFY_DONE;
+}
+
+#ifdef CONFIG_SYSCTL
+static struct ctl_table kdump_ctl_table[] = {
+       {
+               .procname = "kdump_on_init",
+               .data = &kdump_on_init,
+               .maxlen = sizeof(int),
+               .mode = 0644,
+               .proc_handler = proc_dointvec,
+       },
+       {
+               .procname = "kdump_on_fatal_mca",
+               .data = &kdump_on_fatal_mca,
+               .maxlen = sizeof(int),
+               .mode = 0644,
+               .proc_handler = proc_dointvec,
+       },
+       { }
+};
+
+static struct ctl_table sys_table[] = {
+       {
+         .procname = "kernel",
+         .mode = 0555,
+         .child = kdump_ctl_table,
+       },
+       { }
+};
+#endif
+
+static int
+machine_crash_setup(void)
+{
+       /* be notified before default_monarch_init_process */
+       static struct notifier_block kdump_init_notifier_nb = {
+               .notifier_call = kdump_init_notifier,
+               .priority = 1,
+       };
+       int ret;
+       if((ret = register_die_notifier(&kdump_init_notifier_nb)) != 0)
+               return ret;
+#ifdef CONFIG_SYSCTL
+       register_sysctl_table(sys_table);
+#endif
+       return 0;
+}
+
+__initcall(machine_crash_setup);
+