Re: [RFC] Linux memory error handling

[Dave Hansen]

On Mon, 2005-06-20 at 15:42 -0500, Russ Anderson wrote:
> Is there a standard for how to name hardware entries in 
> /sys/devices/system (or sysfs in general)?

For system devices, no, I don't think so.

> Seems like his same general issue would apply to other hardware
> components, cpus, nodes, etc.  

I don't think it's true for anything other than memory.  Linux doesn't
manage CPUs or nodes any differently than it manages its internal
representations of them.

But, for memory at a level any less granular than pages, Linux and the
hardware seldom have the same view. 

> > One other minor thing.  You might want to think about referring to the
> > pieces of memory as things other than DIMMs.  On ppc64, for instance,
> > the hypervisor hands off memory in sections called LMBs (logical memory
> > blocks), and they're not directly related to any hardware DIMM.  The
> > same things will show up in other virtualized environments.
> 
> If we're talking about specific hardware entries, it seems like they
> should be called what they are.  If we're talking about abstractions,
> a more abstract name seems in order.  One of my concerns is mapping 
> failures back to hardware components, hence my bias for component names.

Even with generic names mapping back to components should be easy

Somethinng like memory/type could even contain the hardware type of each
ram entry.

> Would having /sys/.../memory/LMB on ppc64 to correspond to 
> /sys/.../memory/DIMM be a problem?

No, it wouldn't really be too much of a problem.  But, it's not a very
accurate description.  There is certainly hardware that has RAM which
does not have a single DIMM. :)

> RAM would be an alternative,
> but that could be confused with /sys/block/ram.  :-)

RAM would probably be fine.  There should be very few properties
that /sys/devices/system devices share with /sys/block, so it shouldn't
be too bad.

> In general, I'm more concerned with getting the necessary functionality
> in than what the specific entries are called, so I'll go along with
> the consensus.

I don't think there's much of a consensus.  I just want to make sure we
do something that works on all platforms consistently.  For instance,
I'd hate to have every userspace utilities that are trying to examine
memory look like this:

if [ `uname -a` == ppc64 ]; then
	UNIT=LMB
elif [ `uname -a` == ia64 ]; then
	UNIT=DIMM
etc...

FILE=/sys/devices/system/memory/$UNIT$NUMBER

-- Dave

-
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/