Re: [PATCH v3] x86/mce: Honour bios-set CMCI threshold

[Naveen N. Rao]

On 10/17/2012 06:39 PM, Borislav Petkov wrote:
> On Wed, Oct 17, 2012 at 04:57:30PM +0530, Naveen N. Rao wrote:
> > On 10/17/2012 04:29 PM, Borislav Petkov wrote:
> > > > > +static struct dev_ext_attribute dev_attr_bios_cmci_threshold = {
> > > > > +	__ATTR(bios_cmci_threshold, 0444, device_show_int, NULL),
> > > > > +	&mce_bios_cmci_threshold
> > >
> > > Ok, I just noticed this (we must've missed it during review) but why is
> > > this read-only? If it has to be read-only, why do we have a node for
> > > this in sysfs instead of simply issuing the printk statements below and
> > > people who are interested in this, can grep dmesg?
> >
> > This was added so that user-space tools could find out if we're
> > using thresholds for CMCI.
>
> That I figured out.
>
> What I can't figure out is why userspace tools need to know that - the
> fact that some MCI_CTL2 has a 0 CMCI threshold because BIOS forgot to
> set it correctly? IOW, this is a rather evolved workaround for b0rked
> BIOS (the gazillionth BIOS f*ckup, btw if someone is counting :)) and,
> on top of that, we have a read-only, special sysfs node which is pretty
> useless to me.
>
> Why?

This is not about the CMCI threshold being 0, but rather about the 
threshold not being 1. Linux used to program a threshold of 1 always but 
with this boot option, a firmware-set non-zero threshold is honoured.

Userspace tools need this sysfs attribute so they know how to react on 
receipt of a corrected error event: whether this is the first event or 
if such events have already been threshold-ed.


Regards,
Naveen

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/