Add a low impact health check triggered by the system NMI command that essentially checks which CPUs are responding to external NMI's. Signed-off-by: Mike Travis Acked-by: Dimitri Sivanich Reviewed-by: Russ Anderson Reviewed-by: Alex Thorlton --- arch/x86/platform/uv/uv_nmi.c | 23 ++++++++++++++++++++++- 1 file changed, 22 insertions(+), 1 deletion(-) --- linux.orig/arch/x86/platform/uv/uv_nmi.c +++ linux/arch/x86/platform/uv/uv_nmi.c @@ -177,6 +177,7 @@ module_param_named(debug, uv_nmi_debug, * "kdump" - do crash dump * "kdb" - enter KDB (default) * "kgdb" - enter KGDB + * "health" - check if CPUs respond to NMI */ static char uv_nmi_action[8] = "kdb"; module_param_string(action, uv_nmi_action, sizeof(uv_nmi_action), 0644); @@ -573,6 +574,22 @@ static void uv_nmi_sync_exit(int master) } } +/* Current "health" check is which CPU's are responsive */ +static void uv_nmi_action_health(int cpu, struct pt_regs *regs, int master) +{ + if (master) { + int in = atomic_read(&uv_nmi_cpus_in_nmi); + int out = num_online_cpus() - in; + + pr_alert("UV: NMI CPU health check (non-responding:%d)\n", out); + atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT); + } else { + while (!atomic_read(&uv_nmi_slave_continue)) + cpu_relax(); + } + uv_nmi_sync_exit(master); +} + /* Walk through cpu list and dump state of each */ static void uv_nmi_dump_state(int cpu, struct pt_regs *regs, int master) { @@ -748,8 +765,12 @@ int uv_handle_nmi(unsigned int reason, s /* Pause as all cpus enter the NMI handler */ uv_nmi_wait(master); + /* Health check for any processors that are 'stuck' */ + if (uv_nmi_action_is("health")) + uv_nmi_action_health(cpu, regs, master); + /* Dump state of each cpu */ - if (uv_nmi_action_is("ips") || uv_nmi_action_is("dump")) + else if (uv_nmi_action_is("ips") || uv_nmi_action_is("dump")) uv_nmi_dump_state(cpu, regs, master); /* Call KGDB/KDB if enabled */ --