Re: [PATCH] lds: align u-boot-nodtb with 8 bytes boundary

[Tom Rini <trini@konsulko.com>]

[-- Attachment #1: Type: text/plain, Size: 5507 bytes --]

On Thu, Dec 22, 2022 at 12:20:01AM +0100, Mark Kettenis wrote:
> > Date: Wed, 21 Dec 2022 18:09:10 -0500
> > From: Tom Rini <trini@konsulko.com>
> > 
> > On Wed, Dec 21, 2022 at 11:42:56PM +0100, Mark Kettenis wrote:
> > > > Date: Wed, 21 Dec 2022 16:56:37 -0500
> > > > From: Tom Rini <trini@konsulko.com>
> > > > 
> > > > On Sat, Dec 17, 2022 at 11:14:41PM +0100, Mark Kettenis wrote:
> > > > > > From: Eugen Hristev <eugen.hristev@microchip.com>
> > > > > > Date: Thu, 15 Dec 2022 13:58:25 +0200
> > > > > > 
> > > > > > Newer DTC require that the DTB start address is aligned at 8 bytes.
> > > > > > In the u-boot.bin case, the u-boot-nodtb.bin is concatenated with the
> > > > > > DTB, but there is no alignment/padding to the next 8byte aligned address.
> > > > > > This causes the board to fail booting, because the FDT will claim
> > > > > > that the DTB inside u-boot.bin is not a valid DTB, it will fail with
> > > > > > -FDT_ERR_ALIGNMENT.
> > > > > > To solve this, have the u-boot binary `_end` aligned with 8 bytes.
> > > > > > The objcopy in the Makefile will create the u-boot-nodtb.bin and it has to
> > > > > > be truncated to 8 bytes to correspond to the u-boot.map file which will
> > > > > > have the `_end` aligned to 8 bytes.
> > > > > > The lds files which use devicetrees have been changed to align the `_end`
> > > > > > tag with 8 bytes.
> > > > > > 
> > > > > > This patch is also a prerequisite to have the possibility to update the
> > > > > > dtc inside u-boot to newer versions (1.6.1+)
> > > > > > 
> > > > > > Signed-off-by: Eugen Hristev <eugen.hristev@microchip.com>
> > > > > > ---
> > > > > > Hi,
> > > > > > 
> > > > > > I could not test all affected boards, it's an impossible task.
> > > > > > I tried this on at91 boards which I have, and ran the CI on denx.
> > > > > > I cannot guarantee that no other boards are affected, so this patch is a bit
> > > > > > of an RFC.
> > > > > > If the u-boot-nodtb.bin does not have the size equal with the corresponding
> > > > > > one in u-boot.map, the binary_size_check will fail at build time with
> > > > > > something like this:
> > > > > > 
> > > > > > u-boot.map shows a binary size of 502684
> > > > > > but u-boot-nodtb.bin shows 502688
> > > > > > 
> > > > > > Thanks,
> > > > > > Eugen
> > > > > > 
> > > > > >  Makefile                                    | 2 ++
> > > > > >  arch/arm/cpu/armv8/u-boot.lds               | 4 ++--
> > > > > >  arch/arm/cpu/u-boot-spl.lds                 | 1 +
> > > > > >  arch/arm/cpu/u-boot.lds                     | 1 +
> > > > > >  arch/arm/lib/elf_arm_efi.lds                | 5 +++++
> > > > > >  arch/arm/mach-at91/arm926ejs/u-boot-spl.lds | 2 +-
> > > > > >  arch/arm/mach-at91/armv7/u-boot-spl.lds     | 2 +-
> > > > > >  arch/arm/mach-zynq/u-boot-spl.lds           | 2 +-
> > > > > >  arch/mips/cpu/u-boot.lds                    | 2 +-
> > > > > >  arch/sandbox/cpu/u-boot.lds                 | 6 ++++++
> > > > > >  arch/sh/cpu/u-boot.lds                      | 2 ++
> > > > > >  board/ti/am335x/u-boot.lds                  | 1 +
> > > > > >  tools/binman/test/u_boot_binman_embed.lds   | 2 +-
> > > > > >  13 files changed, 25 insertions(+), 7 deletions(-)
> > > > > > 
> > > > > > diff --git a/Makefile b/Makefile
> > > > > > index 9d84f96481..b4d387bcce 100644
> > > > > > --- a/Makefile
> > > > > > +++ b/Makefile
> > > > > > @@ -1317,6 +1317,8 @@ endif
> > > > > >  
> > > > > >  u-boot-nodtb.bin: u-boot FORCE
> > > > > >  	$(call if_changed,objcopy_uboot)
> > > > > > +# Make sure the size is 8 byte-aligned.
> > > > > > +	@truncate -s %8 $@
> > > > > 
> > > > > $ truncate
> > > > > ksh: truncate: not found
> > > > > 
> > > > > In other words: truncate(1) isn't a standard UNIX utility and not
> > > > > present on OpenBSD for example.  It isn't part of POSIX and therefore
> > > > > its usage is unportable.
> > > > > 
> > > > > Please find a different solution.
> > > > 
> > > > Ah yes. Can this perhaps be done with dd? A bit of looking around
> > > > suggests that this might be a portable way to truncate a file to say 512
> > > > bytes:
> > > > dd if=/dev/urandom of=testfile bs=1 count=500
> > > > dd if=/dev/null of=testfile bs=1 count=512
> > > > 
> > > > And then hexdump or whatever to see that the last 12 bytes are zero. Can
> > > > you please test this on OpenBSD?  We'll need some shell work to get
> > > > current byte size and make it 8-byte aligned, but that should be
> > > > portable at least.  Thanks!
> > > 
> > > Hi Tom,
> > > 
> > > That results on a file with zero bytes on OpenBSD... but also on
> > > Linux.  Did you intend to write something different?  I can't
> > > immediately come up with a way to do this with dd(1).
> > 
> > Ah, I typod that, sorry!  I meant (and just re-checked):
> > dd if=/dev/urandom of=testfile bs=1 count=500
> > dd if=/dev/null of=testfile bs=1 seek=512
> > 
> > Which here does:
> > $ dd if=/dev/urandom of=testfile bs=1 count=500
> > 500+0 records in
> > 500+0 records out
> > 500 bytes copied, 0.00641125 s, 78.0 kB/s
> > $ dd if=/dev/null of=testfile bs=1 seek=512
> > 0+0 records in
> > 0+0 records out
> > 0 bytes copied, 0.000358246 s, 0.0 kB/s
> > $ ls -l testfile
> > -rw-rw-r-- 1 trini trini 512 Dec 21 18:07 testfile
> > $ hexdump testfile | tail -n2
> > 00001f0 8cdd 300e 0000 0000 0000 0000 0000 0000
> > 0000200
> 
> That works here as well.

Great, thanks!

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]