[PATCH] arm: Remove rel.dyn from SPL linker scripts
Tom Rini
trini at konsulko.com
Thu Jan 29 17:52:10 CET 2026
On Fri, Jan 30, 2026 at 12:43:51AM +0800, Sune Brian wrote:
> On Fri, Jan 30, 2026 at 12:41 AM Tom Rini <trini at konsulko.com> wrote:
> >
> > On Fri, Jan 30, 2026 at 12:39:17AM +0800, Sune Brian wrote:
> > > On Fri, Jan 30, 2026 at 12:35 AM Tom Rini <trini at konsulko.com> wrote:
> > > >
> > > > On Fri, Jan 30, 2026 at 12:29:56AM +0800, Sune Brian wrote:
> > > > > On Fri, Jan 30, 2026 at 12:22 AM Tom Rini <trini at konsulko.com> wrote:
> > > > > >
> > > > > > On Fri, Jan 30, 2026 at 12:09:26AM +0800, Sune Brian wrote:
> > > > > > > On Fri, Jan 30, 2026 at 12:08 AM Tom Rini <trini at konsulko.com> wrote:
> > > > > > > >
> > > > > > > > On Fri, Jan 30, 2026 at 12:07:25AM +0800, Sune Brian wrote:
> > > > > > > > > On Fri, Jan 30, 2026 at 12:03 AM Tom Rini <trini at konsulko.com> wrote:
> > > > > > > > > >
> > > > > > > > > > On Thu, Jan 29, 2026 at 11:58:50PM +0800, Sune Brian wrote:
> > > > > > > > > > > On Thu, Jan 29, 2026 at 11:46 PM Tom Rini <trini at konsulko.com> wrote:
> > > > > > > > > > > >
> > > > > > > > > > > > On Thu, Jan 29, 2026 at 11:43:02PM +0800, Sune Brian wrote:
> > > > > > > > > > > > > On Thu, Jan 29, 2026 at 11:34 PM Tom Rini <trini at konsulko.com> wrote:
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > On Thu, Jan 29, 2026 at 11:31:51PM +0800, Sune Brian wrote:
> > > > > > > > > > > > > > > On Thu, Jan 29, 2026 at 11:26 PM Tom Rini <trini at konsulko.com> wrote:
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > On Thu, Jan 29, 2026 at 11:25:03PM +0800, Sune Brian wrote:
> > > > > > > > > > > > > > > > > On Thu, Jan 29, 2026 at 11:19 PM Tom Rini <trini at konsulko.com> wrote:
> > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > On Thu, Jan 29, 2026 at 11:06:59PM +0800, Sune Brian wrote:
> > > > > > > > > > > > > > > > > > > On Thu, Jan 29, 2026 at 10:54 PM Tom Rini <trini at konsulko.com> wrote:
> > > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > > On Thu, Jan 29, 2026 at 10:58:21AM +0800, Sune Brian wrote:
> > > > > > > > > > > > > > > > > > > > > On Wed, Jan 28, 2026 at 10:51 PM Tom Rini <trini at konsulko.com> wrote:
> > > > > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > > > > On Wed, Jan 28, 2026 at 10:49:37PM +0800, Sune Brian wrote:
> > > > > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > > > > > Hi Tom,
> > > > > > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > > > > > The commits after c08da5d03c2a0b72e81a11ff7ca507e3a6414db3
> > > > > > > > > > > > > > > > > > > > > > > All Cyclone V boards are unable to boot, not sure this patch is the fix?
> > > > > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > > > Hi Tom,
> > > > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > > > > Yes, those commits were unintentional and have been reverted.
> > > > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > > > Had tested the last commit 6a1bdb7e952d5841f42742fefa907cae5dc8d50a
> > > > > > > > > > > > > > > > > > > > > Same situation that the boards are unable to enter the SPL phase.
> > > > > > > > > > > > > > > > > > > > > The commit 8de6e8f8a076d2c9b6d38d8563db135c167077ec works properly.
> > > > > > > > > > > > > > > > > > > > > After that things are broken.
> > > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > > Please give
> > > > > > > > > > > > > > > > > > > > https://patchwork.ozlabs.org/project/uboot/patch/20260129035512.639362-1-marek.vasut+renesas@mailbox.org/
> > > > > > > > > > > > > > > > > > > > a try as I think it might have been working before by alignment luck
> > > > > > > > > > > > > > > > > > > > rather than design.
> > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > Hi Tom,
> > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > It is the revert itself that is broken.
> > > > > > > > > > > > > > > > > > > I tested f4dfa5d3c2680322fd17fcc7c6221876e00a03c2
> > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > U-Boot SPL 2026.01-00754-gf4dfa5d3c268-dirty (Jan 29 2026 - 22:56:17 +0800)
> > > > > > > > > > > > > > > > > > > Trying to boot from MMC1
> > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > Which revert is correct.
> > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > But once dc2d8423b19d30472b02e02b41504226908a4291
> > > > > > > > > > > > > > > > > > > And revert to 431f1ce46bbffff0db8f03437a34400b11b30175
> > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > Then all these broken I don't understand why even file diff cannot
> > > > > > > > > > > > > > > > > > > show any changes.
> > > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > > Do you mean the revert is broken and fixed by alignment?
> > > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > > I mean that arbitrary changes (as part of testing these things, the
> > > > > > > > > > > > > > > > > > "-dirty" in the version string would lead to failing trees now passing)
> > > > > > > > > > > > > > > > > > can make the difference between pass/fail to boot. Did you try Marek's
> > > > > > > > > > > > > > > > > > patch I linked to?
> > > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > > Yes, I had tried. Cannot boot same as I had mentioned failing commit #.
> > > > > > > > > > > > > > > > > I apply that patch on the latest master commit.
> > > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > > The other, maybe, thing is you need a bigger SPL_SYS_MALLOC_F_LEN ?
> > > > > > > > > > > > > > > > Otherwise, you'll need to dig more and see why the device tree isn't
> > > > > > > > > > > > > > > > being found where expected on your platform, and possibly where it is
> > > > > > > > > > > > > > > > instead.
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > No the failing point is done after
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > f4dfa5d3c2680322fd17fcc7c6221876e00a03c2
> > > > > > > > > > > > > > > Revert "arm: spl: Correct alignment of .rel.dyn section"
> > > > > > > > > > > > > > > This reverts commit 380ddb4. Signed-off-by: Tom Rini <trini at konsulko.com>
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > This commit I tried to boot and works..
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Then this commit
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > dc2d8423b19d30472b02e02b41504226908a4291
> > > > > > > > > > > > > > > count rel_dyn
> > > > > > > > > > > > > > > Signed-off-by: Tom Rini <trini at konsulko.com>
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > I manually replaced the old Makefile.xpl and also worked it boot.
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Then 431f1ce46bbffff0db8f03437a34400b11b30175
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > Revert a number of incorrect commits
> > > > > > > > > > > > > > > As part of debugging the appended device tree failure, I inadvertently
> > > > > > > > > > > > > > > committed some changes as I was debugging to master, and not a private
> > > > > > > > > > > > > > > branch, and pushed them as part of the release. This reverts commit
> > > > > > > > > > > > > > > dc2d842 through 380ddb4. Signed-off-by: Tom Rini <trini at konsulko.com>
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > This is the point that everything no longer boots.
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > But between those commits there are no extra codes.
> > > > > > > > > > > > > > >
> > > > > > > > > > > > > > > After that all things no longer work.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Yes, and I need you to investigate your failure case here. There is no
> > > > > > > > > > > > > > code change between a working commit and a failing commit, which leads
> > > > > > > > > > > > > > me to the suggestions I gave for debugging.
> > > > > > > > > > > > >
> > > > > > > > > > > > > I even tried a fresh clone from https://github.com/u-boot/u-boot.
> > > > > > > > > > > > > Same situation.
> > > > > > > > > > > > > I don't understand.
> > > > > > > > > > > > > The commit missed or messed up?
> > > > > > > > > > > >
> > > > > > > > > > > > You need to look at the resulting binaries and see why the device tree
> > > > > > > > > > > > is not where it's expected to be.
> > > > > > > > > > > >
> > > > > > > > > > > > You might find it helpful looking through
> > > > > > > > > > > > https://lore.kernel.org/u-boot/CAOMZO5BiQcp2HYJgrYnPQLn8Q52Z05FKjbqVKPXxKa8D3Cny5Q@mail.gmail.com/
> > > > > > > > > > > > for some ideas.
> > > > > > > > > > >
> > > > > > > > > > > So you are suggesting even the commit between revert could introduce this issue?
> > > > > > > > > > > Aren't the revert only applied to two changes?
> > > > > > > > > >
> > > > > > > > > > What I'm saying is that the underlying issue is that the device tree is
> > > > > > > > > > not where it's expected to be. The likely root cause is that we now
> > > > > > > > > > require the device tree to be (correctly!) 8 byte aligned in memory and
> > > > > > > > > > not 4 byte aligned. In the case of appending a device tree to our binary
> > > > > > > > > > any sort of change in the code can change the size of the binary and in
> > > > > > > > > > turn mean that the address where the device tree is ends up being at a
> > > > > > > > > > different alignment.
> > > > > > > > >
> > > > > > > > > You are correct on this, but I don't understand why even reverting to
> > > > > > > > > two code changes
> > > > > > > > > could introduce this issue.
> > > > > > > > >
> > > > > > > > > The revert also included other commits?
> > > > > > > > >
> > > > > > > > > "
> > > > > > > > > Revert a number of incorrect commits
> > > > > > > > > As part of debugging the appended device tree failure, I inadvertently
> > > > > > > > > committed some changes as I was debugging to master, and not a private
> > > > > > > > > branch, and pushed them as part of the release. This reverts commit
> > > > > > > > > dc2d842 through 380ddb4.
> > > > > > > > > "
> > > > > > > > >
> > > > > > > > > This is my debug case:
> > > > > > > > >
> > > > > > > > > Correct boot:
> > > > > > > > > hexdump -Cv spl/u-boot-spl.sfp | grep 'd0 0d fe ed'
> > > > > > > > > 00001ba0 08 00 70 47 6f f0 12 00 70 47 00 bf d0 0d fe ed |..pGo...pG......|
> > > > > > > > > 0000beb0 00 00 00 00 00 00 00 00 d0 0d fe ed 00 00 08 c8 |................|
> > > > > > > > >
> > > > > > > > > Fail boot:
> > > > > > > > > hexdump -Cv u-boot-spl_a.sfp | grep 'd0 0d fe ed'
> > > > > > > > > 00001ba0 08 00 70 47 6f f0 12 00 70 47 00 bf d0 0d fe ed |..pGo...pG......|
> > > > > > > > > 0000beb0 00 00 00 00 00 00 00 00 00 00 00 00 d0 0d fe ed |................|
> > > > > > > >
> > > > > > > > Yup, so it's mis-aligned. I guess can you tell me what the defconfig
> > > > > > > > you're using is?
> > > > > > >
> > > > > > > socfpga_cyclone5_defconfig
> > > > > >
> > > > > > You might need to do some debugging yourself on this. Is that also where
> > > > > > __bss_end is? This looks like the exact case that
> > > > > > 5ffc1dcc26d3df9e2b192151936cb98014fb6e49 should have fixed. And locally,
> > > > > > I see it being aligned, and then also just making a trival change (git
> > > > > > tree is dirty) that used to lead to mis-alignment, retaining alignment.
> > > > >
> > > > > I have a question I still don't understand.
> > > > > The example you gave me is even earlier than
> > > > > f4dfa5d3c2680322fd17fcc7c6221876e00a03c2.
> > > > > But build on f4dfa5d3c2680322fd17fcc7c6221876e00a03c2 can boot normally.
> > > > > So what 431f1ce46bbffff0db8f03437a34400b11b30175 actually reverted and
> > > > > could introduce this shift from first place.
> > > > >
> > > > > I really need this to clear out for me to debug the issue.
> > > >
> > > > You might need to build with V=1 and see what's being generated to go in
> > > > to the binary file, and look at spl/u-boot-spl.map to see if things like
> > > > __bss_end match. You need to ignore the 3 commits I reverted because
>
> For the __bss_end:
>
> Boot case:
>
> hexdump -Cv u-boot-spl.sfp | grep 'd0 0d fe ed'
> 00001ba0 08 00 70 47 6f f0 12 00 70 47 00 bf d0 0d fe ed |..pGo...pG......|
> 0000beb0 00 00 00 00 00 00 00 00 d0 0d fe ed 00 00 08 c8 |................|
>
> arm-none-eabi-readelf -s spl/u-boot-spl | grep __bss_end
> 1196: ffffbeb8 0 NOTYPE GLOBAL DEFAULT 6 __bss_end
>
> This makes sense to me because it matched.
>
> Fail boot cas:
>
> hexdump -Cv u-boot-spl_a.sfp | grep 'd0 0d fe ed'
> 00001ba0 08 00 70 47 6f f0 12 00 70 47 00 bf d0 0d fe ed |..pGo...pG......|
> 0000beb0 00 00 00 00 00 00 00 00 00 00 00 00 d0 0d fe ed |................|
>
> arm-none-eabi-readelf -s spl/u-boot-spl | grep __bss_end
> 1196: ffffbeb8 0 NOTYPE GLOBAL DEFAULT 6 __bss_end
FWIW, I use: grep -E '__bss_(start|end)' spl/u-boot-spl.map
> So this is indeed broken as it now should be beb'C'?
It shouldn't be 'bebc' as that is not 8 byte aligned.
> Where to change this may I ask?
That's what I'm asking you to figure out. The __bss_end symbol is at
beb8, which is aligned, but there's 4 extra bytes between where the dtb
should be, and where it's being placed. I thought I had fixed that.
Maybe the rule for -pad.bin in scripts/Makefile.xpl is getting something
wrong (see the reverted commits for when I was investigating that) but
maybe not.
--
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20260129/419915ef/attachment.sig>
More information about the U-Boot
mailing list