Skip to contentSkip to conversation

Switch APIC (+nmi, +oprofile) to driver model

DISCUSSION Mikael Pettersson 1 participant 1 message Feb 9, 2003

lore
Share & tools
lore
#0 lore raw
Quoted 6 lines · John Levon
Quoted 2 lines · Mikael Pettersson

I don't like the idea of registering apic_driver when !cpu_has_apic, but it might be needed for the local-APIC NMI watchdog.

Really ?

Pavel's patch changes nmi.c to include:

extern struct sys_device device_apic;

static int __init init_nmi_devicefs(void)

{
	driver_register(&nmi_driver);

	device_nmi.parent = &device_apic.dev;
        return device_register(&device_nmi);
}

device_initcall(init_nmi_devicefs);

so nmi.c will unconditionally register a device with the local APIC's device as parent. I strongly suspect this will break if &device_apic hasn't itself been device_register():d.

The NMI driver's suspend/resume procedures check nmi_active before doing anything, so registering the NMI device unconditionally is _probably_ safe. (Although I personally think it should be conditional.)

Quoted 9 lines

as long as it's exported to modules. I'd probably prefer to just have :

	disable_nmi_watchdog();
	...
	enable_nmi_watchdog();

and have those do the right thing depending on a (nmi.c local) nmi_watchdog.

Yeah, that's nicer. disable_nmi_watchdog() could easily stash away the previous state, and enable_nmi_watchdog() could check that copy and conditionally call setup_apic_nmi_watchdog().

/Mikael