Re: [PATCH] beaglebone.conf: temporarily use generic am335x_evm_config for U-boot

[Denys Dmytriyenko <denys@ti.com>]

On Tue, Oct 16, 2018 at 02:19:36PM -0700, Khem Raj wrote:
> On Tue, Oct 16, 2018 at 12:20 PM Khem Raj <raj.khem@gmail.com> wrote:
> >
> > On Tue, Oct 16, 2018 at 12:00 PM Denys Dmytriyenko <denys@ti.com> wrote:
> > >
> > > On Tue, Oct 16, 2018 at 11:50:36AM -0700, Khem Raj wrote:
> > > > On Tue, Oct 16, 2018 at 11:29 AM Denys Dmytriyenko <denys@ti.com> wrote:
> > > > >
> > > > > On Tue, Oct 16, 2018 at 11:11:36AM -0700, Khem Raj wrote:
> > > > > > On Tue, Oct 16, 2018 at 9:42 AM Tom Rini <trini@konsulko.com> wrote:
> > > > > > >
> > > > > > > On Sun, Oct 14, 2018 at 10:07:45PM -0700, Khem Raj wrote:
> > > > > > > > On Sun, Oct 14, 2018 at 12:24 PM Denys Dmytriyenko <denys@ti.com> wrote:
> > > > > > > > >
> > > > > > > > > On Sat, Oct 13, 2018 at 01:17:12AM -0700, Khem Raj wrote:
> > > > > > > > > > On Fri, Oct 12, 2018 at 8:00 PM Denys Dmytriyenko <denys@ti.com> wrote:
> > > > > > > > > > >
> > > > > > > > > > > There have been reports recently that am335x_beaglebone_config generates bad SPL.
> > > > > > > > > > > Until that is debugged and fixed, use generic am335x_evm_config that covers all
> > > > > > > > > > > AM335x platforms, including BeagleBone variants.
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > it fails to link
> > > > > > > > > >
> > > > > > > > > > | arm-yoe-linux-gnueabi-ld.bfd: u-boot-spl section `.rodata' will not
> > > > > > > > > > fit in region `.sram'
> > > > > > > > > > | arm-yoe-linux-gnueabi-ld.bfd: region `.sram' overflowed by 5772 bytes
> > > > > > > > > > | make[2]: *** [/mnt/a/yoe/build/tmp/work/beaglebone-yoe-linux-gnueabi/u-boot-ti-staging/2018.01+gitAUTOINC+2cc52408bf-r24/git/scripts/Makefile.spl:349:
> > > > > > > > > > spl/u-boot-spl] Error 1
> > > > > > > > >
> > > > > > > > > FWIW, just built u-boot-ti-staging with gcc7 and gcc8 from oe-core, as well as
> > > > > > > > > Linaro gcc7 - no problems.
> > > > > > > >
> > > > > > > > My distro inherits poky policies, and on master it now inherits
> > > > > > > > hardening policies ( security flags) by defaults
> > > > > > > > do you happen to test poky ?
> > > > > > >
> > > > > > > I think we want to take a look at which of the security flags really
> > > > > > > make sense to use in this context.  Thanks!
> > > > > > >
> > > > > >
> > > > > > there could be more to it, since the distro uses thumb2 ISA by
> > > > > > default, I am not sure
> > > > > > if u-boot overrides that and builds using arm mode ISA always but
> > > > > > something to consider, I saw several reports about u-boot overflowing
> > > > > > sram sections and most of
> > > > > > the solutions were "oh it works for me" or at the best your toolchain
> > > > > > is different then mine. here is mine use it and move on.
> > > > >
> > > > > Khem,
> > > > >
> > > > > Well, FWIW, Tom and I are very familiar with this issue. As a matter of fact,
> > > > > I first encountered it almost 2 years ago and had to prove there's such an
> > > > > issue, because everyone was saying it works for them, something must be wrong
> > > > > with my OE builds... :)
> > > > >
> > > > > While .sram region is very limited, the issue is exacerbated by the fact that
> > > > > all debug symbols from macros like __FILE__ are ended up in that section as
> > > > > well. So, the longer your build path, the larger the section becomes. Once I
> > > > > had instructions to reproduce the failure here internally with a series of
> > > > > long-named nested directories like aaaaaa and bbbbbb, Nishanth started this
> > > > > thread on U-boot mailing list:
> > > > > https://lists.denx.de/pipermail/u-boot/2017-March/285031.html
> > > > >
> > > > > We've had the corresponding bug open internally all this time, while adding
> > > > > workarounds to limit .sram section size by other means, like disabling some
> > > > > options to reduce the code size. Your patch is one of those workarounds...
> > > > >
> > > > > But we've been patiently waiting for the following feature to come into gcc to
> > > > > fix the issue properly:
> > > > > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70268
> > > > >
> > > > > Since it's now part of gcc8, we should be able to use it. Not sure how to keep
> > > > > gcc7 backward compatibility though...
> > > >
> > > > dumping absolute file name strings into SPL seems a waste of space to
> > > > me, but I will leave that out for now. Moreover it exposes build paths
> > > > into binaries that user may not be interested to share
> > > >
> > > > -ffile-prefix-map has been in OE toolchains since gcc6 and I think we
> > > > are already using it for kernel builds. We can probably enhance uboot
> > > > recipes in OE-Core to
> > > > use this option if compiler supports it. That solves my problem.
> > >
> > > Yeah, extending that from kernel to u-boot would be nice.
> > > Unfortunately, our products use Linaro gcc7 on rocko for now. Planning to
> > > migrate to gcc8 on thud soon...
> >
> > we can check for the option before using it so atleast it will not
> > break older toolchains more than what they are already broken.
> 
> I added -ffile-prefix-map locally just with this option added it did
> not fix the problem.

I also tried adding -ffile-prefix-map locally and it only shaved off less than 
500 bytes. Still wondering what those extra 5KB of code are...