Re: [RFC][PATCH 4/5] Documentation: add node files to sysfs ABI

[Nishanth Aravamudan <nacc@us.ibm.com>]

On 19.04.2008 [19:24:21 -0700], Greg KH wrote:
> On Fri, Apr 18, 2008 at 10:27:30AM -0700, Nishanth Aravamudan wrote:
> > On 17.04.2008 [23:04:04 -0700], Nishanth Aravamudan wrote:
> > > On 17.04.2008 [16:39:56 -0700], Christoph Lameter wrote:
> > > > On Thu, 17 Apr 2008, Nishanth Aravamudan wrote:
> > > > 
> > > > > That seems fine to me. I will work on it. However, as I mentioned in a
> > > > > previous e-mail, the files in /sys/devices/system/node/node<nr>/
> > > > > already violate the "one value per file" rule in several instances. I'm
> > > > > guessing Greg won't want me moving the files and keeping that violation?
> > > > 
> > > > That violation is replicated in /proc/meminfo /proc/vmstat etc etc.
> > > 
> > > Right, but /proc doesn't have such a restriction (the "one value per
> > > file" rule). I'm not sure how the meminfo, etc. files in sysfs got put
> > > in past Greg, but that's how it is :)
> > 
> > Greg, can you give any insight here? Are we better off leaving the files
> > in question in /sys/devices/system/node/node<nr>/{meminfo,numastat,etc}
> > since they are part of the ABI there and already violate the rules for
> > sysfs? Or can we move them to /sys/kernel and continue to violate the
> > rules? In this case, I don't see any way to provide a "snapshot" of the
> > system's memory information without all the values being in one file?
> 
> Yeah, the "snapshot" issue is what allows those values all to be present
> at once.
> 
> As for where to place them, are there any tools out there that are
> expecting the current file locations?  If so, can they work if they are
> in both places?

I believe libnuma uses /sys/devices/system/node for some information. As
long as the files are in both places, nothing should be affected,
though. And we could deprecate the old files (if we decide to move them)
for the longer-term and update the necessary libraries.

> If you think they should be moved, I'll defer to your judgement, but
> it will be a bit harder, as you will be working with "raw" kobjects in
> that case, not the sysdev structures, which do make things a bit
> easier for you.

Yeah, I noticed that while fiddling around. Still possible, just not as
easy.

> sorry for the delay, am traveling...

No problem, thanks for the input!

-Nish

--
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>