Re: [PATCH] NMI request/release

[John Levon]

On Mon, Oct 21, 2002 at 08:32:47PM -0500, Corey Minyard wrote:

> The attached patch implements a way to request to receive an NMI if it 
> comes from an otherwise unknown source.  I needed this for handling NMIs 
> with the IPMI watchdog.  This function was discussed a little a while 

Then NMI watchdog and oprofile should be changed to use this too. We
also need priority and/or equivalent of NOTIFY_STOP_MASK so we can break
out of calling all the handlers. Actually, why do you continue if one
of the handlers returns 1 anyway ?

> +	atomic_inc(&calling_nmi_handlers);

Isn't this going to cause cacheline ping pong ?

> +	curr = nmi_handler_list;
> +	while (curr) {
> +		handled |= curr->handler(curr->dev_id, regs);

dev_name is never used at all. What is it for ? Also, would be nice
to do an smp_processor_id() just once and pass that in to prevent
multiple calls to get_current().

Couldn't you modify the notifier code to do the xchg()s (though that's
not available on all CPU types ...)

> +#define HAVE_NMI_HANDLER	1

What uses this ?

> +	volatile struct nmi_handler *next;

Hmm ...

Is it not possible to use linux/rcupdate.h for this stuff ?

regards
john
-- 
"Lots of companies would love to be in our hole."
	- Scott McNealy