On Thu, 2008-04-03 at 21:33 -0500, Corey Minyard wrote:
> Andrew Morton wrote:
> > On Thu, 03 Apr 2008 16:30:18 -0800 Carol Hebert <[EMAIL PROTECTED]> wrote:
> >
> >   
> >> Hi,
> >>
> >> This reverse enumeration scheme is actually different from how the ipmi
> >> driver has worked in the past.  The BMCs in multinode systems were
> >> historically named by the ipmi driver in the order in which they were
> >> probed.  As such, current multinode users are expecting ipmi0 to
> >> reference the BMC in their first node, ipmi1 to reference the BMC in
> >> their second node, etc. as was always the case in the past.  Without
> >> this patch, they will unknowingly be addressing the wrong BMC.  This
> >> could conceivably cause operations to be erroneously executed that could
> >> result in data loss, etc.
> >>     
> >
> > In which kernel version did this behavioural change occur?
> >
> > (this is a hard way to write a changelog!)
> >   
> The change occurred in 2.6.14, when the DMI scanning was moved out of 
> the IPMI driver and into the DMI code.  So it's been a while.
> 
> But the trouble is, no one cared since no one had more than one IPMI 
> controller in a system.  But now IBM has come up with the machines with 
> a boatload of them.  As far as I know, that IBM machine is the only one 
> that will care.
> 
> IMHO, the best way to handle this is to use udev to properly name the 
> devices.  That was not an option for Carol, though.  And having them in 
> the opposite order of the firmware violates the principle of least 
> surprise.  So I went ahead and took the patch.
> 
> -corey

Hi Corey, 

Thank you very much for the info on the timing of the 2.6.14 dmi code
change.  :-)  Although the list_add call in the dmi code originates from
the 2.6.14 kernel, I seem to believe that the issue with the reverse
naming started more recently -- namely that it began when the ipmi
driver interfaces changed from an array table to a linked list.  I may
be totally wrong (it's been a couple of months since I looked at that
code), but I thought I remembered that the array table set the ipmi
interfaces in the table in probe order but that the new(er) ipmi linked
list interface allowed the dmi code's stack-type paradigm to cause a
reversal of the interface naming order.

I'll try to find code to back up (or negate) what I'm saying -- or maybe
you can answer the question right off?

Either way, thank you very much for all your help,

--Carol 


-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace
_______________________________________________
Openipmi-developer mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openipmi-developer

Reply via email to