[ofa-general] Re: [PATCH] infiniband-diags/ibtracert: print lids in decimal form

[Hal Rosenstock]

On Mon, 2007-11-26 at 21:07 +0000, Sasha Khapyorsky wrote:
> On 13:41 Sat 17 Nov     , Jason Gunthorpe wrote:
> > On Thu, Nov 15, 2007 at 06:30:17AM -0800, Hal Rosenstock wrote:
> > > On Thu, 2007-11-15 at 12:37 +0200, Sasha Khapyorsky wrote:
> > > > As stated in bug#504 (https://bugs.openfabrics.org/show_bug.cgi?id=504)
> > > > lid output format unification is needed. Print LIDs as decimal in
> > > > ibtracert.
> > > 
> > > I'd prefer to see this done as some sort of option. Also, I think hex is
> > > better for MLIDs.
> > 
> > FWIW, we consistently use hex notation in our switch products in the
> > format '0x10/16' which specifies both the LMC and the LID in a compact
> > manner.
> > 
> > Since both GIDs, GUIDs and MACs are printed in hex, choosing decimal for lid
> > seems like an inconsistent choice to me. It also makes it harder to
> > read out the LMC bits.
> > 
> > Also, this same kind of unification is needed for GID's. They should
> > always be printed and accepted in IPv6 format, not 128 bit
> > decimal.
> 
> Not sure this was discussed explicitly (probably it should be). My
> feeling based on the feedback from people (emails, opened bugs, etc)
> was that desired *IDs formats are decimal for LIDs, LMC and hexadecimal
> for GUIDs, MLIDs.

Not sure about decimal for LMC (as it is a mask to apply to LID).

I also think that some prefer hex and would rather see some option for
decimal/hex.

Also, as I said before, I think that unicast LIDs need to be consistent
with the SM.

> I don't remember that GID was discussed at all and I think that Jason's
> proposition about ipv6 format is first here.
> 
> Any other opinions? Should *IDs formats be formalized at all? Thoughts?

ipv6 format for GIDs makes sense to me (also affects SM too).

-- Hal

> Sasha