[PATCH v4 05/15] lmb: notify of any changes to the LMB memory map

[Tom Rini]

On Tue, Feb 11, 2025 at 04:41:09PM +0530, Sughosh Ganu wrote:
> On Tue, 11 Feb 2025 at 16:34, Alexander Dahl <ada at thorsis.com> wrote:
> >
> > Hello,
> >
> > Am Tue, Feb 11, 2025 at 04:23:13PM +0530 schrieb Sughosh Ganu:
> > > On Mon, 20 Jan 2025 at 15:43, Alexander Dahl <ada at thorsis.com> wrote:
> > > >
> > > > Hello,
> > > >
> > > > Am Mon, Jan 20, 2025 at 02:40:38PM +0530 schrieb Sughosh Ganu:
> > > > > On Fri, 17 Jan 2025 at 16:40, Alexander Dahl <ada at thorsis.com> wrote:
> > > >
> > > > […]
> > > >
> > > > > > This changeset breaks build with CONFIG_CC_OPTIMIZE_FOR_DEBUG=y for
> > > > > > me.  In a usual non-debug build, everything is fine.  However in a
> > > > > > debug build now I get this error:
> > > > > >
> > > > > >       UPD     include/generated/timestamp_autogenerated.h
> > > > > >       GEN     Makefile
> > > > > >       Using /mnt/data/adahl/src/u-boot as source for U-Boot
> > > > > >       CC      common/version.o
> > > > > >       AR      common/built-in.o
> > > > > >       LD      u-boot
> > > > > >     arm-v5te-linux-gnueabi-ld.bfd: lib/lmb.o: in function `lmb_map_update_notify':
> > > > > >     /mnt/data/adahl/src/u-boot/lib/lmb.c:458: undefined reference to `efi_add_memory_map_pg'
> > > > >
> > > > > Can you share the details of the toolchain that you are using, where
> > > > > can I download it from. And the target that you build which shows this
> > > > > up. Thanks.
> > > >
> > > > Although I'm actually building with ptxdist and
> > > > OSELAS.Toolchain-2023.07.1/arm-v5te-linux-gnueabi for a custom board,
> > > > you should be able to reproduce the behaviour easily with buildman on
> > > > master when building for boards sam9x60ek or sam9x60_curiosity like
> > > > this:
> > > >
> > > >     buildman -o ~/build/u-boot/buildman -e -a CC_OPTIMIZE_FOR_DEBUG sam9x60
> > > >
> > > > My output when running the above is this:
> > >
> > > Sorry, I haven't spent time debugging this issue, although this is on
> > > my todo list. Btw, since you already have a working setup, can you
> > > please reset your tree to the following commit and then check if that
> > > fixes your issue.
> > >
> > > 92e75ee47f1 ("board_r: Remove duplicate headers")
> >
> > Does not change anything.
> >
> > Note: my tree is not special in any way, it's a clean mainline u-boot
> > checkout and I'm building for a board with an in tree config.
> > Everyone should be able to reproduce this by using
> > configs/sam9x60_curiosity_mmc_defconfig, enabling
> > CC_OPTIMIZE_FOR_DEBUG and building with the default toolchain
> > downloaded by buildman, which is also mainline u-boot tooling.
> >
> > My assumption is: some efi code is pulled in unconditionally here,
> > but efi is not enabled on armv5 and it can not be enabled,  efi is not
> > needed on this target anyways.  Furthermore: I assume the code is
> > optimized out when doing a normal build without CC_OPTIMIZE_FOR_DEBUG,
> > which is basically building without optimizations and keeping debug
> > symbols.
> 
> Okay. Please give me some time. I will try to spend some time on the
> issue this week. Thanks for your quick response.

Yes, the problem is that CC_OPTIMIZE_FOR_DEBUG disables compiler
optimizations that we normally rely on to avoid having to #ifdef the
code. In this case, lmb_should_notify() causes this to normally be
optimized away as CONFIG_IS_ENABLED(EFI_LOADER) will be false and the
compiler will do the right thing. The "fix" is to instead make most of
the body of lmb_map_update_notify be guarded with #if
CONFIG_IS_ENABLED(EFI_LOADER) ... #endif

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20250211/0d5c71f3/attachment.sig>