[U-Boot] GCC 5.2 issue on imx28

[U-Boot] GCC 5.2 issue on imx28

[Otavio Salvador]

Hello folks,

OE-Core is preparing for upgrade to GCC 5.2 as default compiler and
mx28 is failing[1] to build with it.

1. http://errors.yoctoproject.org/Errors/Details/13878/

I am not a linker guy so could someone shed any light on this?

-- 
Otavio Salvador                             O.S. Systems
http://www.ossystems.com.br        http://code.ossystems.com.br
Mobile: +55 (53) 9981-7854            Mobile: +1 (347) 903-9750

[U-Boot] GCC 5.2 issue on imx28

[Wolfgang Denk]

Dear Otavio,

In message <CAP9ODKqqA6v6Ug4+Cx4fBdCRLchaOxL3SCZN5+iwbbFoKyq-mA@mail.gmail.com> you wrote:
> 
> OE-Core is preparing for upgrade to GCC 5.2 as default compiler and
> mx28 is failing[1] to build with it.
> 
> 1. http://errors.yoctoproject.org/Errors/Details/13878/
> 
> I am not a linker guy so could someone shed any light on this?

Does the same problem happen with the pristine mainline U-Boot tree?

The error message looks pretty cleaar to me:

../arch/arm/cpu/arm926ejs/start.S:108: undefined reference to `lowlevel_init'

In your case (mx28evk_config) this would normally come from
arch/arm/cpu/arm926ejs/mxs/mxs.c:

	inline void lowlevel_init(void) {}

Did you check the compiler release notes for differences in the
handling of such inline funtions?

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
I'd rather be led to hell than managed to heaven.

[U-Boot] GCC 5.2 issue on imx28

[Måns Rullgård]

Otavio Salvador <otavio.salvador@ossystems.com.br> writes:

> Hello folks,
>
> OE-Core is preparing for upgrade to GCC 5.2 as default compiler and
> mx28 is failing[1] to build with it.
>
> 1. http://errors.yoctoproject.org/Errors/Details/13878/
>
> I am not a linker guy so could someone shed any light on this?

There are two errors reports:

1. An undefined reference to the symbol "lowlevel_init"
2. A complaint about the ".rel.plt" section not being handled by the
   linker script.

The second error is probably caused by the first.  A quick grep turns up
this snippet in arch/arm/cpu/arm926ejs/mxs/mxs.c:

/* Lowlevel init isn't used on i.MX28, so just have a dummy here */
inline void lowlevel_init(void) {}

The semantics for non-static functions declared inline have changed in
gcc5, causing the above (empty) function not to be emitted as an
external symbol.

Since that function is only referenced from start.S, it should not be
declared inline at all.  This patch should thus fix your problem:

diff --git a/arch/arm/cpu/arm926ejs/mxs/mxs.c b/arch/arm/cpu/arm926ejs/mxs/mxs.c
index ef130ae..b1d8721 100644
--- a/arch/arm/cpu/arm926ejs/mxs/mxs.c
+++ b/arch/arm/cpu/arm926ejs/mxs/mxs.c
@@ -24,7 +24,7 @@
 DECLARE_GLOBAL_DATA_PTR;
 
 /* Lowlevel init isn't used on i.MX28, so just have a dummy here */
-inline void lowlevel_init(void) {}
+void lowlevel_init(void) {}
 
 void reset_cpu(ulong ignored) __attribute__((noreturn));
 
-- 
M?ns Rullg?rd
mans at mansr.com

[U-Boot] GCC 5.2 issue on imx28

[Otavio Salvador]

On Tue, Jul 28, 2015 at 10:39 AM, M?ns Rullg?rd <mans@mansr.com> wrote:
...
> Since that function is only referenced from start.S, it should not be
> declared inline at all.  This patch should thus fix your problem:
>
> diff --git a/arch/arm/cpu/arm926ejs/mxs/mxs.c b/arch/arm/cpu/arm926ejs/mxs/mxs.c
> index ef130ae..b1d8721 100644
> --- a/arch/arm/cpu/arm926ejs/mxs/mxs.c
> +++ b/arch/arm/cpu/arm926ejs/mxs/mxs.c
> @@ -24,7 +24,7 @@
>  DECLARE_GLOBAL_DATA_PTR;
>
>  /* Lowlevel init isn't used on i.MX28, so just have a dummy here */
> -inline void lowlevel_init(void) {}
> +void lowlevel_init(void) {}
>
>  void reset_cpu(ulong ignored) __attribute__((noreturn));

Works like a charm.

Tested-by: Otavio Salvador <otavio@ossystems.com.br>

-- 
Otavio Salvador                             O.S. Systems
http://www.ossystems.com.br        http://code.ossystems.com.br
Mobile: +55 (53) 9981-7854            Mobile: +1 (347) 903-9750

[U-Boot] GCC 5.2 issue on imx28

[Jörg Krause]

Dear M?ns Rullg?rd, Otavio Salvador,

On Di, 2015-07-28 at 14:39 +0100, M?ns Rullg?rd wrote:
> Otavio Salvador <otavio.salvador@ossystems.com.br> writes:

[snip]

> There are two errors reports:
> 
> 1. An undefined reference to the symbol "lowlevel_init"
> 2. A complaint about the ".rel.plt" section not being handled by the
>    linker script.
> 
> The second error is probably caused by the first.  A quick grep turns 
> up
> this snippet in arch/arm/cpu/arm926ejs/mxs/mxs.c:
> 
> /* Lowlevel init isn't used on i.MX28, so just have a dummy here */
> inline void lowlevel_init(void) {}
> 
> The semantics for non-static functions declared inline have changed 
> in
> gcc5, causing the above (empty) function not to be emitted as an
> external symbol.
> 
> Since that function is only referenced from start.S, it should not be
> declared inline at all.  This patch should thus fix your problem:
> 
> diff --git a/arch/arm/cpu/arm926ejs/mxs/mxs.c 
> b/arch/arm/cpu/arm926ejs/mxs/mxs.c
> index ef130ae..b1d8721 100644
> --- a/arch/arm/cpu/arm926ejs/mxs/mxs.c
> +++ b/arch/arm/cpu/arm926ejs/mxs/mxs.c
> @@ -24,7 +24,7 @@
>  DECLARE_GLOBAL_DATA_PTR;
>  
>  /* Lowlevel init isn't used on i.MX28, so just have a dummy here */
> -inline void lowlevel_init(void) {}
> +void lowlevel_init(void) {}
>  
>  void reset_cpu(ulong ignored) __attribute__((noreturn));
>  

I stumbled over the same problem. Unfortunatly, I did not find this
patch before (only the error report from Otavia) and submitted a
similar patch [1] which keeps the inline keyword.

Best regards
J?rg Krause

[1] "arm: mxs: make inline function compatible for GCC 5"
https://patchwork.ozlabs.org/patch/504043/

[U-Boot] GCC 5.2 issue on imx28

[Måns Rullgård]

J?rg Krause <joerg.krause@embedded.rocks> writes:

> Dear M?ns Rullg?rd, Otavio Salvador,
>
> On Di, 2015-07-28 at 14:39 +0100, M?ns Rullg?rd wrote:
>> Otavio Salvador <otavio.salvador@ossystems.com.br> writes:
>
> [snip]
>
>> There are two errors reports:
>> 
>> 1. An undefined reference to the symbol "lowlevel_init"
>> 2. A complaint about the ".rel.plt" section not being handled by the
>>    linker script.
>> 
>> The second error is probably caused by the first.  A quick grep turns 
>> up
>> this snippet in arch/arm/cpu/arm926ejs/mxs/mxs.c:
>> 
>> /* Lowlevel init isn't used on i.MX28, so just have a dummy here */
>> inline void lowlevel_init(void) {}
>> 
>> The semantics for non-static functions declared inline have changed 
>> in
>> gcc5, causing the above (empty) function not to be emitted as an
>> external symbol.
>> 
>> Since that function is only referenced from start.S, it should not be
>> declared inline at all.  This patch should thus fix your problem:
>> 
>> diff --git a/arch/arm/cpu/arm926ejs/mxs/mxs.c 
>> b/arch/arm/cpu/arm926ejs/mxs/mxs.c
>> index ef130ae..b1d8721 100644
>> --- a/arch/arm/cpu/arm926ejs/mxs/mxs.c
>> +++ b/arch/arm/cpu/arm926ejs/mxs/mxs.c
>> @@ -24,7 +24,7 @@
>>  DECLARE_GLOBAL_DATA_PTR;
>>  
>>  /* Lowlevel init isn't used on i.MX28, so just have a dummy here */
>> -inline void lowlevel_init(void) {}
>> +void lowlevel_init(void) {}
>>  
>>  void reset_cpu(ulong ignored) __attribute__((noreturn));
>>  
>
> I stumbled over the same problem. Unfortunatly, I did not find this
> patch before (only the error report from Otavia) and submitted a
> similar patch [1] which keeps the inline keyword.
>
> Best regards
> J?rg Krause
>
> [1] "arm: mxs: make inline function compatible for GCC 5"
> https://patchwork.ozlabs.org/patch/504043/

Since the function is only referenced from outside the C file, any use
of inline makes little sense to me.  While your patch achieves the
result of creating a linkable instance of the function, it is more
complicated than it needs to be.

-- 
M?ns Rullg?rd
mans at mansr.com

[U-Boot] GCC 5.2 issue on imx28

[Jörg Krause]

On Mi, 2015-08-05 at 20:23 +0100, M?ns Rullg?rd wrote:
> J?rg Krause <joerg.krause@embedded.rocks> writes:
> 
> > Dear M?ns Rullg?rd, Otavio Salvador,
> > 
> > On Di, 2015-07-28 at 14:39 +0100, M?ns Rullg?rd wrote:
> > > Otavio Salvador <otavio.salvador@ossystems.com.br> writes:
> > 
> > [snip]
> > 
> > > There are two errors reports:
> > > 
> > > 1. An undefined reference to the symbol "lowlevel_init"
> > > 2. A complaint about the ".rel.plt" section not being handled by 
> > > the
> > >    linker script.
> > > 
> > > The second error is probably caused by the first.  A quick grep 
> > > turns 
> > > up
> > > this snippet in arch/arm/cpu/arm926ejs/mxs/mxs.c:
> > > 
> > > /* Lowlevel init isn't used on i.MX28, so just have a dummy here 
> > > */
> > > inline void lowlevel_init(void) {}
> > > 
> > > The semantics for non-static functions declared inline have 
> > > changed 
> > > in
> > > gcc5, causing the above (empty) function not to be emitted as an
> > > external symbol.
> > > 
> > > Since that function is only referenced from start.S, it should 
> > > not be
> > > declared inline at all.  This patch should thus fix your problem:
> > > 
> > > diff --git a/arch/arm/cpu/arm926ejs/mxs/mxs.c 
> > > b/arch/arm/cpu/arm926ejs/mxs/mxs.c
> > > index ef130ae..b1d8721 100644
> > > --- a/arch/arm/cpu/arm926ejs/mxs/mxs.c
> > > +++ b/arch/arm/cpu/arm926ejs/mxs/mxs.c
> > > @@ -24,7 +24,7 @@
> > >  DECLARE_GLOBAL_DATA_PTR;
> > >  
> > >  /* Lowlevel init isn't used on i.MX28, so just have a dummy here 
> > > */
> > > -inline void lowlevel_init(void) {}
> > > +void lowlevel_init(void) {}
> > >  
> > >  void reset_cpu(ulong ignored) __attribute__((noreturn));
> > >  
> > 
> > I stumbled over the same problem. Unfortunatly, I did not find this
> > patch before (only the error report from Otavia) and submitted a
> > similar patch [1] which keeps the inline keyword.
> > 
> > Best regards
> > J?rg Krause
> > 
> > [1] "arm: mxs: make inline function compatible for GCC 5"
> > https://patchwork.ozlabs.org/patch/504043/
> 
> Since the function is only referenced from outside the C file, any 
> use
> of inline makes little sense to me.  While your patch achieves the
> result of creating a linkable instance of the function, it is more
> complicated than it needs to be.
> 

In my opinion it quite make sense to use inline for functions
referenced only from other files. The keyword is just an hint to the
compiler and why should it not make sense for other C files? However,
in this case it makes really no difference if lowlevel_init is marked
as inline, gcc will generate an "BL lowlevel_init" instruction in both
cases.

J?rg

[U-Boot] GCC 5.2 issue on imx28

[Måns Rullgård]

J?rg Krause <joerg.krause@embedded.rocks> writes:

> On Mi, 2015-08-05 at 20:23 +0100, M?ns Rullg?rd wrote:
>> J?rg Krause <joerg.krause@embedded.rocks> writes:
>> 
>> > Dear M?ns Rullg?rd, Otavio Salvador,
>> > 
>> > On Di, 2015-07-28 at 14:39 +0100, M?ns Rullg?rd wrote:
>> > > Otavio Salvador <otavio.salvador@ossystems.com.br> writes:
>> > 
>> > [snip]
>> > 
>> > > There are two errors reports:
>> > > 
>> > > 1. An undefined reference to the symbol "lowlevel_init"
>> > > 2. A complaint about the ".rel.plt" section not being handled by 
>> > > the
>> > >    linker script.
>> > > 
>> > > The second error is probably caused by the first.  A quick grep 
>> > > turns 
>> > > up
>> > > this snippet in arch/arm/cpu/arm926ejs/mxs/mxs.c:
>> > > 
>> > > /* Lowlevel init isn't used on i.MX28, so just have a dummy here 
>> > > */
>> > > inline void lowlevel_init(void) {}
>> > > 
>> > > The semantics for non-static functions declared inline have 
>> > > changed 
>> > > in
>> > > gcc5, causing the above (empty) function not to be emitted as an
>> > > external symbol.
>> > > 
>> > > Since that function is only referenced from start.S, it should 
>> > > not be
>> > > declared inline at all.  This patch should thus fix your problem:
>> > > 
>> > > diff --git a/arch/arm/cpu/arm926ejs/mxs/mxs.c 
>> > > b/arch/arm/cpu/arm926ejs/mxs/mxs.c
>> > > index ef130ae..b1d8721 100644
>> > > --- a/arch/arm/cpu/arm926ejs/mxs/mxs.c
>> > > +++ b/arch/arm/cpu/arm926ejs/mxs/mxs.c
>> > > @@ -24,7 +24,7 @@
>> > >  DECLARE_GLOBAL_DATA_PTR;
>> > >  
>> > >  /* Lowlevel init isn't used on i.MX28, so just have a dummy here 
>> > > */
>> > > -inline void lowlevel_init(void) {}
>> > > +void lowlevel_init(void) {}
>> > >  
>> > >  void reset_cpu(ulong ignored) __attribute__((noreturn));
>> > >  
>> > 
>> > I stumbled over the same problem. Unfortunatly, I did not find this
>> > patch before (only the error report from Otavia) and submitted a
>> > similar patch [1] which keeps the inline keyword.
>> > 
>> > Best regards
>> > J?rg Krause
>> > 
>> > [1] "arm: mxs: make inline function compatible for GCC 5"
>> > https://patchwork.ozlabs.org/patch/504043/
>> 
>> Since the function is only referenced from outside the C file, any 
>> use
>> of inline makes little sense to me.  While your patch achieves the
>> result of creating a linkable instance of the function, it is more
>> complicated than it needs to be.
>> 
>
> In my opinion it quite make sense to use inline for functions
> referenced only from other files. The keyword is just an hint to the
> compiler and why should it not make sense for other C files?

If using link-time optimisation, it could make sense, yes.  If not,
there is no way a function can be inlined across translation units.

> However, in this case it makes really no difference if lowlevel_init
> is marked as inline, gcc will generate an "BL lowlevel_init"
> instruction in both cases.

Correct.  Here the only caller is in an assembly file, which is not
subject to link-time optimisations.

-- 
M?ns Rullg?rd
mans at mansr.com

[U-Boot] GCC 5.2 issue on imx28

[Tom Rini]

On Tue, Jul 28, 2015 at 02:39:49PM +0100, M?ns Rullg?rd wrote:

> Otavio Salvador <otavio.salvador@ossystems.com.br> writes:
> 
> > Hello folks,
> >
> > OE-Core is preparing for upgrade to GCC 5.2 as default compiler and
> > mx28 is failing[1] to build with it.
> >
> > 1. http://errors.yoctoproject.org/Errors/Details/13878/
> >
> > I am not a linker guy so could someone shed any light on this?
> 
> There are two errors reports:
> 
> 1. An undefined reference to the symbol "lowlevel_init"
> 2. A complaint about the ".rel.plt" section not being handled by the
>    linker script.
> 
> The second error is probably caused by the first.  A quick grep turns up
> this snippet in arch/arm/cpu/arm926ejs/mxs/mxs.c:
> 
> /* Lowlevel init isn't used on i.MX28, so just have a dummy here */
> inline void lowlevel_init(void) {}
> 
> The semantics for non-static functions declared inline have changed in
> gcc5, causing the above (empty) function not to be emitted as an
> external symbol.
> 
> Since that function is only referenced from start.S, it should not be
> declared inline at all.  This patch should thus fix your problem:
> Tested-by: Otavio Salvador <otavio@ossystems.com.br>
> 
> diff --git a/arch/arm/cpu/arm926ejs/mxs/mxs.c b/arch/arm/cpu/arm926ejs/mxs/mxs.c
> index ef130ae..b1d8721 100644

After a bit re-wording of the commit message, applied to u-boot/master,
thanks!

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20150813/b1be4463/attachment.sig>