On Mon, 9 Feb 2015, Tony Lindgren wrote:
> * Paul Walmsley <paul@xxxxxxxxx> [150209 08:04]:
> > On Mon, 9 Feb 2015, Peter Ujfalusi wrote:
> >
> > > On 02/06/2015 09:26 PM, Peter Ujfalusi wrote:
> > > >> Yeah, I've never really bothered with data too much, its a debug
> > > >> feature. So lock_class_key is 8 bytes, and strictly speaking you could
> > > >> union them over other fields, all we really need is unique addresses, we
> > > >> don't actually use the storage.
> > > >
> > > > True. our omap2plus defconfig does not have LOCKDEP enabled so it should not
> > > > add anything to the data when running default kernel.
> > > > I'll test the lockdep_set_class() method you suggested on Monday (not
> > > > tomorrow), but still as first thing.
> > > > If it is working as expected I'll send a patch with you as author.
> > >
> > > With omap2plus_defconfig my build produces (vmlinux size):
> > > Base: 99905522
> > > with my series: 99908385 (base + 2863)
> > > with Peter Zijlstra's patch: 99910625 (base + 5103)
> > >
> > > The reason for this is that we will only have
> > > struct lock_class_key { };
> > > in case of !CONFIG_LOCKDEP. On ARM however CONFIG_LOCKDEP is enabled by
> > > default, while the CONFIG_DEBUG_LOCKDEP is disabled.
> > >
> > > So it does add more data to our default omap2plus config.
> > >
> > > Tony: do you have preference on the way we fix this issue?
> > >
> > > As I recall there is a plan to remove the hwmod static database and move it or
> > > generate it from DT? Not sure when and how this will be done, but will it
> > > affect the lockdep_set_class() way?
> >
> > Well I guess we could see what Tony says, but you do realize that the
> > difference in sizes that you posted above is about .003% of the total
> > binary size, right?
> >
> > If there's one thing we can say about the last few years of ARM kernel
> > development, it's that those kind of size increases are utterly dwarfed by
> > other changes in the kernel. So I'd say, post a patch based on PeterZ's
> > fix and be done with it...
>
> Well the thing to consider here is what Peter U is saying about
> having struct omap_hwmod allocated based on the data from .dts
> files. If the fix makes the dynamic allocation harder to do later on,
> we should probably avoid it. If it's relatively easy to do later on,
> then I don't have a problem with it.
The future destination for that code that makes the most sense to me is
for it to become integrated with the OMAP Sonics & Arteris bus drivers and
DT data. So I wouldn't worry too much about it; I don't think the
lockdep fix will affect that at all.
- Paul
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
