[dpdk-dev] [PATCH 07/10] eal: add core list input format

Bruce Richardson bruce.richardson at intel.com
Mon Nov 24 14:28:22 CET 2014


On Mon, Nov 24, 2014 at 02:19:16PM +0100, Thomas Monjalon wrote:
> Hi Bruce and Neil,
> 
> 2014-11-24 11:28, Bruce Richardson:
> > On Sat, Nov 22, 2014 at 08:35:17PM -0500, Neil Horman wrote:
> > > On Sat, Nov 22, 2014 at 10:43:39PM +0100, Thomas Monjalon wrote:
> > > > From: Didier Pallard <didier.pallard at 6wind.com>
> > > > 
> > > > In current version, used cores can only be specified using a bitmask.
> > > > It will now be possible to specify cores in 2 different ways:
> > > > - Using a bitmask (-c [0x]nnn): bitmask must be in hex format
> > > > - Using a list in following format: -l <c1>[-c2][,c3[-c4],...]
> > > > 
> > > > The letter -l can stand for lcore or list.
> > > > 
> > > > -l 0-7,16-23,31 being equivalent to -c 0x80FF00FF
> > > 
> > > Do you want to burn an option letter on that?  It seems like it might be better
> > > to search the string for 0x and base the selection of bitmap of list parsing
> > > based on its presence or absence.
> 
> It was the initial proposal (in April):
> 	http://dpdk.org/ml/archives/dev/2014-April/002173.html
> And I liked keeping only 1 option;
> 	http://dpdk.org/ml/archives/dev/2014-May/002722.html
> But Anatoly raised the compatibility problem:
> 	http://dpdk.org/ml/archives/dev/2014-May/002723.html
> Then there was no other comment so Didier and I reworked a separate option. 
> 
> > The existing coremask parsing always assumes a hex coremask, so just looking
> > for a 0x will not work. I prefer this scheme of using a new flag for this method
> > of specifying the cores to use. 
> > 
> > If you don't want to use up a single-letter option, two alternatives:
> > 1) use a long option instead.
> > 2) if the -c parameter includes a "-" or a ",", treat it as a new-style option,
> > otherwise treat as old. The only abiguity here would be for specifying a single
> > core value 1-9 e.g. is "-c 6" a mask with two bits, or a single-core to run on.
> > [0 is obviously a named core as it's an invalid mask, and A-F are obviously
> > masks.] If we did want this scheme, I would suggest that we allow trailing
> > commas in the list specifier, so we can force users to clear ambiguity by
> > either writing "0x6" or "6," i.e. disallow ambiguous values to avoid problems.
> > However, this is probably more work that it's worth to avoid using up a letter
> > option.
> > 
> > I'd prefer any of these options to breaking backward compatibility in this case.
> 
> We need a consensus here.
> Who is supporting a "burn" of an one-letter option with clear usage?
> Who is supporting a "re-merge" of the 2 syntaxes with more complicated rules
> (list syntax is triggered by presence of "-" or ",")? 
>

Burn!


More information about the dev mailing list