Re: [PATCH] soc-family: fix SOC_FAMILY override order

[Denys Dmytriyenko <denis@denix.org>]

On Mon, Mar 11, 2013 at 06:03:25PM +0000, Maupin, Chase wrote:
> > -----Original Message-----
> > From: otavio.salvador@gmail.com
> > [mailto:otavio.salvador@gmail.com] On Behalf Of Otavio Salvador
> > Sent: Monday, March 11, 2013 11:24 AM
> > To: Maupin, Chase
> > Cc: Denys Dmytriyenko; Patches and discussions about the oe-core
> > layer
> > Subject: Re: [OE-core] [PATCH] soc-family: fix SOC_FAMILY
> > override order
> > 
> > On Mon, Mar 11, 2013 at 11:49 AM, Maupin, Chase
> > <chase.maupin@ti.com> wrote:
> > >> -----Original Message-----
> > >> From: otavio.salvador@gmail.com
> > >> [mailto:otavio.salvador@gmail.com] On Behalf Of Otavio
> > Salvador
> > >> Sent: Saturday, March 09, 2013 6:11 AM
> > >> To: Maupin, Chase
> > >> Cc: Denys Dmytriyenko; Patches and discussions about the oe-
> > core
> > >> layer
> > >> Subject: Re: [OE-core] [PATCH] soc-family: fix SOC_FAMILY
> > >> override order
> > >>
> > >> On Fri, Mar 8, 2013 at 7:16 PM, Maupin, Chase
> > >> <chase.maupin@ti.com> wrote:
> > >> >> -----Original Message-----
> > >> >> From: otavio.salvador@gmail.com
> > >> >> [mailto:otavio.salvador@gmail.com] On Behalf Of Otavio
> > >> Salvador
> > >> >> Sent: Friday, March 08, 2013 2:27 PM
> > >> >> To: Denys Dmytriyenko
> > >> >> Cc: Maupin, Chase; Patches and discussions about the oe-
> > core
> > >> >> layer
> > >> >> Subject: Re: [OE-core] [PATCH] soc-family: fix SOC_FAMILY
> > >> >> override order
> > >> >>
> > >> >> On Fri, Mar 8, 2013 at 5:06 PM, Denys Dmytriyenko
> > >> >> <denis@denix.org> wrote:
> > >> >> > On Fri, Mar 08, 2013 at 03:52:57PM -0300, Otavio Salvador
> > >> >> wrote:
> > >> >> >> On Fri, Mar 8, 2013 at 2:51 PM, Chase Maupin
> > >> >> <Chase.Maupin@ti.com> wrote:
> > >> >> >> > * the current order has SOC_FAMILY settings, which are
> > >> >> generic
> > >> >> >> >   settings for a group of devices, overriding the
> > machine
> > >> >> specific
> > >> >> >> >   settings.  For example:
> > >> >> >> >
> > >> >> >> >   KERNEL_DEVICETREE_ti33x = "xxxx"
> > >> >> >> >   KERNEL_DEVICETREE_beaglebone = "yyyy"
> > >> >> >> >
> > >> >> >> >   Should yield "yyyy" when building for the beaglebone
> > >> >> because
> > >> >> >> >   that is a more specific device than ti33x.  However,
> > >> >> without this
> > >> >> >> >   change the result is that the value is set to "xxxx"
> > >> >> meaning the
> > >> >> >> >   more generic setting overrides the more specific
> > >> setting.
> > >> >> >> >
> > >> >> >> > Signed-off-by: Chase Maupin <Chase.Maupin@ti.com>
> > >> >> >>
> > >> >> >> Maybe while on that you could look at supporting xx:yy
> > as
> > >> SoC
> > >> >> family?
> > >> >> >> like am37xx:am3715 ?
> > >> >> >
> > >> >> > Did you mean am3517? That's a slightly different variant
> > of
> > >> >> am35x/omap35x SoC.
> > >> >>
> > >> >> Yes; sorry ... what I meant was 'am35xx:am3517'
> > >> >>
> > >> >> > But if you really meant am3715 (as well as am3705, am3725
> > >> and
> > >> >> am3730), then
> > >> >> > those are variants of am37x SoC, just with some
> > subsystems,
> > >> >> like SGX or DSP,
> > >> >> > being absent or present. Having those variants handled by
> > >> >> SOC_FAMILY would be
> > >> >> > an overkill. Instead, we've started using
> > MACHINE_FEATURES
> > >> to
> > >> >> distinguish
> > >> >> > between those variants of the same SoC, by checking for
> > >> "sgx"
> > >> >> and/or "dsp"
> > >> >> > flags there and pulling in needed software components
> > >> >> accordingly.
> > >> >>
> > >> >> My main concern here is that COMPATIBLE_MACHINE also parses
> > >> >> SOC_FAMILY
> > >> >> however if you have two (as the 'am35xx:am3517') it is
> > going
> > >> to
> > >> >> fail;
> > >> >> it could split it and parse it individually.
> > >> >
> > >> > Sorry, I'm not sure if I'm following here.  Are you saying
> > you
> > >> would find it useful to have support for a MACHINE to have
> > more
> > >> than one SOC_FAMILY?  In the example above I would have
> > expected
> > >> that you would have a machine config file for am3517 which has
> > an
> > >> SOC_FAMILY of am35xx.  Why would you specify two SOC_FAMILY
> > >> values per machine?
> > >>
> > >> We can have more generic to more specific combinations.
> > >>
> > >> > Or are you trying to build something like omap3->am35xx-
> > >am3517
> > >> where you can use omap3 as a more generic setting but still
> > use
> > >> am35xx for a slightly more restrictive group that is still
> > >> grouping like parts, and finally you use am3517 for the exact
> > >> part?
> > >>
> > >> Exactly so we avoid duplication stuff to boards or SoCs.
> > Another
> > >> example of use: imx -> mx6q -> mx6.
> > >
> > > I see.  This could be of some use and I'll play with it.  This
> > should not be required though for this patch since right now I
> > want to fix the order issue.  Any objection to the patch as is?
> > 
> > No; not really. I just wanted to ask if you could look at it as
> > well
> > so we can have it working. It does make things much easier for
> > all us.
> 
> Sure.  Btw, I noticed a weirdness when looking at this where the 
> COMPATIBLE_MACHINE being evaluated evaulated/matched only has to be a 
> substring of the SOC family to work.  For example if I have:
> 
> MACHINE = omap5-evm
> SOC_FAMILY = omap-a15
> COMPATIBLE_MACHINE = omap
> 
> Then this will work because "omap" (the COMPATIBLE_MACHINE) is a substring 
> of SOC_FAMILY (and technically also a substring of omap5-evm)
> 
> Even setting COMPATIBLE_MACHINE to omap- will work because that is a 
> substring of omap-a15.  So the "match" operation is not really precise 
> enough.  I'm wondering if this needs to be changed to do an exact match.

COMPATIBLE_MACHINE is a regular expression and after thinking about it some 
more and digging around, I believe it was implemented that way on purpose, to 
be as inclusive as possible, so you can do things like "nokia(800|770)" or 
what I do in Arago - "(?!arago)". Or in case of COMPATIBLE_HOST it could be
"^(i.86|x86_64|powerpc).*-linux". So, if you need an exact match, then specify 
it with the regular expression syntax:

COMPATIBLE_MACHINE = "^omap$"

The above won't match either of the examples you provided above, unless 
MACHINE or SOC_FAMILY are specifically set to "omap".

For multiple entries in COMPATIBLE_MACHINE, use "^(omap|davinci)$"

-- 
Denys