Re: [PATCH/RFC] Add node 'states' sysfs class attribute - V2

[Lee Schermerhorn <Lee.Schermerhorn@hp.com>]

On Tue, 2007-08-28 at 15:13 -0700, Nish Aravamudan wrote:
> On 8/28/07, Christoph Lameter <clameter@sgi.com> wrote:
> > On Tue, 28 Aug 2007, Lee Schermerhorn wrote:
> >
> > > I thought I'd give it a try, but thinking that /proc variables were
> > > discouraged, where else but sysfs to put them.  A class attribute
> > > to /sys/devices/system/node seemed like the appropriate place.
> >
> > Right. That is the right place.
> >
> > > I'm not wedded to this interface.  However, I realy don't think it's
> > > worth doing as multiple files.
> >
> > I think one single file per nodemask makes sense. Otherwise files become
> > difficult to parse. I just forgot....
> >
> > > its executed, in the grand scheme of things.  However, I must admit that
> > > I've become addicted to the ease with which one can write one-off
> > > scripts to query configuration/statistics, tune/modify behavior or
> > > trigger actions via just cat'ing from and/or echo'ing to a /proc or /sys
> > > file.
> > >
> > > So, where to go with this patch?  Drop it?  Leave it as is?  Move
> > > it /proc so that it can be a single file?   Make it multiple files in
> > > sysfs?  Putting it as politely as possible, the last is not my favorite
> > > option, but if folks think this info is useful and that's the way to go,
> > > so be it.  And what about mask vs list?  It's a 4 character change in
> > > the code to go either way.
> >
> > I would suggest to do the one file thing in sysfs and use the function
> > that already exists in the kernel to print the nice nodelists. Using the
> > nice function is just calling another function since the code is already
> > there.
> >
> > At some point we may even allow changing the nodemasks. One could imagine
> > that we would add nodemasks that allow use of hugepages on certain nodes
> > or the slab allocator to allocate on certain nodes.
> 
> Just to chime in here -- I've been on vacation for a bit recently -- I
> fully support the one-value per file rule for sysfs. I think it makes
> things a bit clearer. I like this attribute as well, and the idea of
> expanding it down the road is easiest if we use one file per-nodemask.

Welcome back, Nish.

OK, I relent.  I'll respin with one file per state.  I'll go with a
slight modification to the names suggested by Yasunori-san:

possible, online, has_memory, has_cpu

Some come, mon...

Lee

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>