[PATCH] system certificates: specify byte alignment

[PATCH] system certificates: specify byte alignment

[Giuliano Procida]

The .align macro is architecture dependent. On arm64 it behaves as
.p2align. The various alignments in this file are all bytes.

So use the .balign macro to avoid unnecessary padding due to
over-alignment.

Signed-off-by: Giuliano Procida <gprocida@google.com>
---
 certs/system_certificates.S | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/certs/system_certificates.S b/certs/system_certificates.S
index 003e25d4a17e..ea6984b427c9 100644
--- a/certs/system_certificates.S
+++ b/certs/system_certificates.S
@@ -4,7 +4,7 @@
 
 	__INITRODATA
 
-	.align 8
+	.balign 8
 	.globl system_certificate_list
 system_certificate_list:
 __cert_list_start:
@@ -20,14 +20,14 @@ __cert_list_end:
 system_extra_cert:
 	.fill CONFIG_SYSTEM_EXTRA_CERTIFICATE_SIZE, 1, 0
 
-	.align 4
+	.balign 4
 	.globl system_extra_cert_used
 system_extra_cert_used:
 	.int 0
 
 #endif /* CONFIG_SYSTEM_EXTRA_CERTIFICATE */
 
-	.align 8
+	.balign 8
 	.globl system_certificate_list_size
 system_certificate_list_size:
 #ifdef CONFIG_64BIT
@@ -36,7 +36,7 @@ system_certificate_list_size:
 	.long __cert_list_end - __cert_list_start
 #endif
 
-	.align 8
+	.balign 8
 	.globl module_cert_size
 module_cert_size:
 #ifdef CONFIG_64BIT
-- 
2.51.0.384.g4c02a37b29-goog

Re: [PATCH] system certificates: specify byte alignment

[Will Deacon]

On Fri, Sep 12, 2025 at 11:00:15AM +0100, Giuliano Procida wrote:
> The .align macro is architecture dependent. On arm64 it behaves as
> .p2align. The various alignments in this file are all bytes.
> 
> So use the .balign macro to avoid unnecessary padding due to
> over-alignment.
> 
> Signed-off-by: Giuliano Procida <gprocida@google.com>
> ---
>  certs/system_certificates.S | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)

Does certs/revocation_certificates.S need fixing as well?

Will

Re: [PATCH] system certificates: specify byte alignment

[Giuliano Procida]

On Mon, 22 Sept 2025 at 17:38, Will Deacon <will@kernel.org> wrote:
>
> On Fri, Sep 12, 2025 at 11:00:15AM +0100, Giuliano Procida wrote:
> > The .align macro is architecture dependent. On arm64 it behaves as
> > .p2align. The various alignments in this file are all bytes.
> >
> > So use the .balign macro to avoid unnecessary padding due to
> > over-alignment.
> >
> > Signed-off-by: Giuliano Procida <gprocida@google.com>
> > ---
> >  certs/system_certificates.S | 8 ++++----
> >  1 file changed, 4 insertions(+), 4 deletions(-)
>
> Does certs/revocation_certificates.S need fixing as well?

Yes. I will post a v2.

> Will

[PATCH v2] certs: specify byte alignment

[Giuliano Procida]

The alignments specified in system_certificates.S and
revocation_certificates.S are intended to be byte quantities.

However, the .align macro is architecture dependent and on arm64 it
behaves as .p2align. So use the .balign macro to avoid unnecessary
padding due to over-alignment.

Signed-off-by: Giuliano Procida <gprocida@google.com>
---
 certs/revocation_certificates.S | 4 ++--
 certs/system_certificates.S     | 8 ++++----
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/certs/revocation_certificates.S b/certs/revocation_certificates.S
index f21aae8a8f0e..a53b49d49931 100644
--- a/certs/revocation_certificates.S
+++ b/certs/revocation_certificates.S
@@ -4,14 +4,14 @@
 
 	__INITRODATA
 
-	.align 8
+	.balign 8
 	.globl revocation_certificate_list
 revocation_certificate_list:
 __revocation_list_start:
 	.incbin "certs/x509_revocation_list"
 __revocation_list_end:
 
-	.align 8
+	.balign 8
 	.globl revocation_certificate_list_size
 revocation_certificate_list_size:
 #ifdef CONFIG_64BIT
diff --git a/certs/system_certificates.S b/certs/system_certificates.S
index 003e25d4a17e..ea6984b427c9 100644
--- a/certs/system_certificates.S
+++ b/certs/system_certificates.S
@@ -4,7 +4,7 @@
 
 	__INITRODATA
 
-	.align 8
+	.balign 8
 	.globl system_certificate_list
 system_certificate_list:
 __cert_list_start:
@@ -20,14 +20,14 @@ __cert_list_end:
 system_extra_cert:
 	.fill CONFIG_SYSTEM_EXTRA_CERTIFICATE_SIZE, 1, 0
 
-	.align 4
+	.balign 4
 	.globl system_extra_cert_used
 system_extra_cert_used:
 	.int 0
 
 #endif /* CONFIG_SYSTEM_EXTRA_CERTIFICATE */
 
-	.align 8
+	.balign 8
 	.globl system_certificate_list_size
 system_certificate_list_size:
 #ifdef CONFIG_64BIT
@@ -36,7 +36,7 @@ system_certificate_list_size:
 	.long __cert_list_end - __cert_list_start
 #endif
 
-	.align 8
+	.balign 8
 	.globl module_cert_size
 module_cert_size:
 #ifdef CONFIG_64BIT
-- 
2.51.0.534.gc79095c0ca-goog

Re: [PATCH v2] certs: specify byte alignment

[Will Deacon]

On Tue, Sep 23, 2025 at 09:13:42AM +0100, Giuliano Procida wrote:
> The alignments specified in system_certificates.S and
> revocation_certificates.S are intended to be byte quantities.
> 
> However, the .align macro is architecture dependent and on arm64 it
> behaves as .p2align. So use the .balign macro to avoid unnecessary
> padding due to over-alignment.
> 
> Signed-off-by: Giuliano Procida <gprocida@google.com>
> ---
>  certs/revocation_certificates.S | 4 ++--
>  certs/system_certificates.S     | 8 ++++----
>  2 files changed, 6 insertions(+), 6 deletions(-)

Acked-by: Will Deacon <will@kernel.org>

Will

Re: [PATCH v2] certs: specify byte alignment

[Xi Ruoyao]

On Tue, 2025-09-23 at 10:08 +0100, Will Deacon wrote:
> On Tue, Sep 23, 2025 at 09:13:42AM +0100, Giuliano Procida wrote:
> > The alignments specified in system_certificates.S and
> > revocation_certificates.S are intended to be byte quantities.
> > 
> > However, the .align macro is architecture dependent and on arm64 it
> > behaves as .p2align. So use the .balign macro to avoid unnecessary
> > padding due to over-alignment.
> > 
> > Signed-off-by: Giuliano Procida <gprocida@google.com>
> > ---
> >  certs/revocation_certificates.S | 4 ++--
> >  certs/system_certificates.S     | 8 ++++----
> >  2 files changed, 6 insertions(+), 6 deletions(-)
> 
> Acked-by: Will Deacon <will@kernel.org>
> 
> Will

Hi,

It seems this patch somehow got lost?

Kexy found MIPS is specifically affected by the .align directive: on
MIPS it not only behaves as p2align and introduces a lot of extra
padding, but also moved before the __revocation_list_end label in the
case of

__revocation_list_end:

    .align 8

in revocation_certificates.S.  It seems the GNU assembler moves the
.align earlier to mimic the IRIX assembler behavior described at page
110 of https://irix7.com/techpubs/007-2418-006.pdf:


    Labels immediately preceding an automatic or explicit
    alignment are also realigned. For example, foo:
    .align 3; .word 0 is the same as .align 3;
    foo: .word 0.

Kexy's AI proposed the same fix as this patch.  So maybe we should apply
it now?

-- 
Xi Ruoyao <xry111@xry111.site>

Re: [PATCH v2] certs: specify byte alignment

[Giuliano Procida]

Hi.

On Thu, 6 Aug 2026 at 14:51, Xi Ruoyao <xry111@xry111.site> wrote:
>
> On Tue, 2025-09-23 at 10:08 +0100, Will Deacon wrote:
> > On Tue, Sep 23, 2025 at 09:13:42AM +0100, Giuliano Procida wrote:
> > > The alignments specified in system_certificates.S and
> > > revocation_certificates.S are intended to be byte quantities.
> > >
> > > However, the .align macro is architecture dependent and on arm64 it
> > > behaves as .p2align. So use the .balign macro to avoid unnecessary
> > > padding due to over-alignment.
> > >
> > > Signed-off-by: Giuliano Procida <gprocida@google.com>
> > > ---
> > >  certs/revocation_certificates.S | 4 ++--
> > >  certs/system_certificates.S     | 8 ++++----
> > >  2 files changed, 6 insertions(+), 6 deletions(-)
> >
> > Acked-by: Will Deacon <will@kernel.org>
> >
> > Will
>
> Hi,
>
> It seems this patch somehow got lost?
>
> Kexy found MIPS is specifically affected by the .align directive: on
> MIPS it not only behaves as p2align and introduces a lot of extra
> padding, but also moved before the __revocation_list_end label in the
> case of
>
> __revocation_list_end:
>
>     .align 8
>
> in revocation_certificates.S.  It seems the GNU assembler moves the
> .align earlier to mimic the IRIX assembler behavior described at page
> 110 of https://irix7.com/techpubs/007-2418-006.pdf:
>
>
>     Labels immediately preceding an automatic or explicit
>     alignment are also realigned. For example, foo:
>     .align 3; .word 0 is the same as .align 3;
>     foo: .word 0.
>
> Kexy's AI proposed the same fix as this patch.  So maybe we should apply
> it now?

That seems like a good idea. A colleague raised it again this week.
We were considering just applying it to ACK and forgetting about upstream.
I'm glad there's other interest though.

Could you raise this with the relevant maintainer?

> --
> Xi Ruoyao <xry111@xry111.site>

Thanks,
Giuliano.

Re: [PATCH v2] certs: specify byte alignment

[Xi Ruoyao]

On Thu, 2026-08-06 at 15:32 +0100, Giuliano Procida wrote:
> Hi.
> 
> On Thu, 6 Aug 2026 at 14:51, Xi Ruoyao <xry111@xry111.site> wrote:
> > 
> > On Tue, 2025-09-23 at 10:08 +0100, Will Deacon wrote:
> > > On Tue, Sep 23, 2025 at 09:13:42AM +0100, Giuliano Procida wrote:
> > > > The alignments specified in system_certificates.S and
> > > > revocation_certificates.S are intended to be byte quantities.
> > > > 
> > > > However, the .align macro is architecture dependent and on arm64 it
> > > > behaves as .p2align. So use the .balign macro to avoid unnecessary
> > > > padding due to over-alignment.
> > > > 
> > > > Signed-off-by: Giuliano Procida <gprocida@google.com>
> > > > ---
> > > >  certs/revocation_certificates.S | 4 ++--
> > > >  certs/system_certificates.S     | 8 ++++----
> > > >  2 files changed, 6 insertions(+), 6 deletions(-)
> > > 
> > > Acked-by: Will Deacon <will@kernel.org>
> > > 
> > > Will
> > 
> > Hi,
> > 
> > It seems this patch somehow got lost?
> > 
> > Kexy found MIPS is specifically affected by the .align directive: on
> > MIPS it not only behaves as p2align and introduces a lot of extra
> > padding, but also moved before the __revocation_list_end label in the
> > case of
> > 
> > __revocation_list_end:
> > 
> >     .align 8
> > 
> > in revocation_certificates.S.  It seems the GNU assembler moves the
> > .align earlier to mimic the IRIX assembler behavior described at page
> > 110 of https://irix7.com/techpubs/007-2418-006.pdf:
> > 
> > 
> >     Labels immediately preceding an automatic or explicit
> >     alignment are also realigned. For example, foo:
> >     .align 3; .word 0 is the same as .align 3;
> >     foo: .word 0.
> > 
> > Kexy's AI proposed the same fix as this patch.  So maybe we should apply
> > it now?
> 
> That seems like a good idea. A colleague raised it again this week.
> We were considering just applying it to ACK and forgetting about upstream.
> I'm glad there's other interest though.
> 
> Could you raise this with the relevant maintainer?

Hmm, get_maintainers.pl shows David Howells and David Woodhouse are the
relevant maintainers and they are already CC'ed.  Or is the
get_maintainers.pl output inaccurate?


-- 
Xi Ruoyao <xry111@xry111.site>