[2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Mikael Pettersson]

While compiling 2.6.17-rc6 for a 486 with an NE2000 ISA ethernet card, I got:

  Building modules, stage 2.
make -rR -f /tmp/linux-2.6.17-rc6/scripts/Makefile.modpost
  scripts/mod/modpost   -o /tmp/linux-2.6.17-rc6/Module.symvers   vmlinux drivers/net/8390.o drivers/net/ne.o lib/crc32.o net/packet/af_packet.o
WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x158) and 'ne_block_input'
WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x176) and 'ne_block_input'
WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x183) and 'ne_block_input'
WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x1ea) and 'ne_block_input'
WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x251) and 'ne_block_input'
WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x266) and 'ne_block_input'
WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x29b) and 'ne_block_input'

Not sure how serious this is; the driver seems to work fine later on.

/Mikael

[PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Randy.Dunlap]

On Sat, 10 Jun 2006 14:11:42 +0200 (MEST) Mikael Pettersson wrote:

> While compiling 2.6.17-rc6 for a 486 with an NE2000 ISA ethernet card, I got:
> 
> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x158) and 'ne_block_input'
> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x176) and 'ne_block_input'
> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x183) and 'ne_block_input'
> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x1ea) and 'ne_block_input'
> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x251) and 'ne_block_input'
> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x266) and 'ne_block_input'
> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x29b) and 'ne_block_input'
> 
> Not sure how serious this is; the driver seems to work fine later on.

Doesn't look serious.  init_module() is not __init, but it calls
some __init functions and touches some __initdata.

BTW, I would be happy to see some consistent results from modpost
section checking.  I don't see all of these warnings (I see only 1)
when using gcc 3.3.6.  What gcc version are you using?
Does that matter?  (not directed at anyone in particular)

Patch below fixes it for me.  Please test/report.
---

From: Randy Dunlap <rdunlap@xenotime.net>

Fix section mismatch warnings:
WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x396) and 'cleanup_card'
WARNING: drivers/net/ne2.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x483) and 'cleanup_card'

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
---
 drivers/net/ne.c  |    2 +-
 drivers/net/ne2.c |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

--- linux-2617-rc6.orig/drivers/net/ne.c
+++ linux-2617-rc6/drivers/net/ne.c
@@ -829,7 +829,7 @@ that the ne2k probe is the last 8390 bas
 is at boot) and so the probe will get confused by any other 8390 cards.
 ISA device autoprobes on a running machine are not recommended anyway. */
 
-int init_module(void)
+int __init init_module(void)
 {
 	int this_dev, found = 0;
 
--- linux-2617-rc6.orig/drivers/net/ne2.c
+++ linux-2617-rc6/drivers/net/ne2.c
@@ -780,7 +780,7 @@ MODULE_PARM_DESC(bad, "(ignored)");
 
 /* Module code fixed by David Weinehall */
 
-int init_module(void)
+int __init init_module(void)
 {
 	struct net_device *dev;
 	int this_dev, found = 0;

Re: [PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Sam Ravnborg]

> > 
> > Not sure how serious this is; the driver seems to work fine later on.
> 
> Doesn't look serious.  init_module() is not __init, but it calls
> some __init functions and touches some __initdata.

This is the typical case with inconsistent tagging.

> BTW, I would be happy to see some consistent results from modpost
> section checking.  I don't see all of these warnings (I see only 1)
> when using gcc 3.3.6.  What gcc version are you using?
> Does that matter?  (not directed at anyone in particular)
I did not see anyone of them - strange.
I did not dig into it, but objdump -rR ne.o should tell the number of
mismatches with soem carefull checking.


> --- linux-2617-rc6.orig/drivers/net/ne.c
> +++ linux-2617-rc6/drivers/net/ne.c
> @@ -829,7 +829,7 @@ that the ne2k probe is the last 8390 bas
>  is at boot) and so the probe will get confused by any other 8390 cards.
>  ISA device autoprobes on a running machine are not recommended anyway. */
>  
> -int init_module(void)
> +int __init init_module(void)
>  {
>  	int this_dev, found = 0;

When you anyway touches the driver I suggest to name the function
<module>_init, <module>_cleanup and use module_init(), module_cleanup().

	Sam

Re: [PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Mikael Pettersson]

On Sat, 10 Jun 2006 12:13:35 -0700, Randy.Dunlap wrote:
>On Sat, 10 Jun 2006 14:11:42 +0200 (MEST) Mikael Pettersson wrote:
>
>> While compiling 2.6.17-rc6 for a 486 with an NE2000 ISA ethernet card, I got:
>> 
>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x158) and 'ne_block_input'
>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x176) and 'ne_block_input'
>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x183) and 'ne_block_input'
>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x1ea) and 'ne_block_input'
>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x251) and 'ne_block_input'
>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x266) and 'ne_block_input'
>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x29b) and 'ne_block_input'
>> 
>> Not sure how serious this is; the driver seems to work fine later on.
>
>Doesn't look serious.  init_module() is not __init, but it calls
>some __init functions and touches some __initdata.
>
>BTW, I would be happy to see some consistent results from modpost
>section checking.  I don't see all of these warnings (I see only 1)
>when using gcc 3.3.6.  What gcc version are you using?
>Does that matter?  (not directed at anyone in particular)

The messages above are from when I used gcc-4.1.1.
With gcc-3.2.3 I only see a single warning.

>Patch below fixes it for me.  Please test/report.

Worked for me too. Thanks.

/Mikael

>---
>
>From: Randy Dunlap <rdunlap@xenotime.net>
>
>Fix section mismatch warnings:
>WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x396) and 'cleanup_card'
>WARNING: drivers/net/ne2.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x483) and 'cleanup_card'
>
>Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
>---
> drivers/net/ne.c  |    2 +-
> drivers/net/ne2.c |    2 +-
> 2 files changed, 2 insertions(+), 2 deletions(-)
>
>--- linux-2617-rc6.orig/drivers/net/ne.c
>+++ linux-2617-rc6/drivers/net/ne.c
>@@ -829,7 +829,7 @@ that the ne2k probe is the last 8390 bas
> is at boot) and so the probe will get confused by any other 8390 cards.
> ISA device autoprobes on a running machine are not recommended anyway. */
> 
>-int init_module(void)
>+int __init init_module(void)
> {
> 	int this_dev, found = 0;
> 
>--- linux-2617-rc6.orig/drivers/net/ne2.c
>+++ linux-2617-rc6/drivers/net/ne2.c
>@@ -780,7 +780,7 @@ MODULE_PARM_DESC(bad, "(ignored)");
> 
> /* Module code fixed by David Weinehall */
> 
>-int init_module(void)
>+int __init init_module(void)
> {
> 	struct net_device *dev;
> 	int this_dev, found = 0;
>

Re: [PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Jan Engelhardt]

>> Doesn't look serious.  init_module() is not __init, but it calls
>> some __init functions and touches some __initdata.
>
>This is the typical case with inconsistent tagging.
>
Worse yet, I once experienced a double-definition error, that is, I had

  __init int init_module(void)  {
      /* module1 */
  }

and, in another .c file,

  __init int init_module(void) {
      /* module2 */
  }

and making them both CONFIG_...=y gave a typical double-definition link 
time error in vmlinux. The proper way (IMO) is

  static __init int blah_init(void) { ... }
  module_init(blah_init);

Then it does not even matter if blah_init is defined in two different 
modules.

>>  
>> -int init_module(void)
>> +int __init init_module(void)
>>  {
>>  	int this_dev, found = 0;
>
>When you anyway touches the driver I suggest to name the function
><module>_init, <module>_cleanup and use module_init(), module_cleanup().


Jan Engelhardt
-- 

Re: [PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Jan Engelhardt]

>
>The messages above are from when I used gcc-4.1.1.
>With gcc-3.2.3 I only see a single warning.
>
FTR, gcc 4.0.x is also 'affected'.


Jan Engelhardt
-- 

Re: [PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Mikael Pettersson]

On Sat, 10 Jun 2006 22:38:00 +0200, Sam Ravnborg wrote:
>> --- linux-2617-rc6.orig/drivers/net/ne.c
>> +++ linux-2617-rc6/drivers/net/ne.c
>> @@ -829,7 +829,7 @@ that the ne2k probe is the last 8390 bas
>>  is at boot) and so the probe will get confused by any other 8390 cards.
>>  ISA device autoprobes on a running machine are not recommended anyway. */
>>  
>> -int init_module(void)
>> +int __init init_module(void)
>>  {
>>  	int this_dev, found = 0;
>
>When you anyway touches the driver I suggest to name the function
><module>_init, <module>_cleanup and use module_init(), module_cleanup().

Maybe not: in the ne.c driver init_module() is inside #ifdef MODULE,
so conversion to ne_init() + module_init(ne_init) would be a no-op
except for making the code larger. In the non-MODULE case Space.c
calls ne_probe() directly.

/Mikael

Re: [PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Sam Ravnborg]

On Sun, Jun 11, 2006 at 04:37:31PM +0200, Mikael Pettersson wrote:
> On Sat, 10 Jun 2006 22:38:00 +0200, Sam Ravnborg wrote:
> >> --- linux-2617-rc6.orig/drivers/net/ne.c
> >> +++ linux-2617-rc6/drivers/net/ne.c
> >> @@ -829,7 +829,7 @@ that the ne2k probe is the last 8390 bas
> >>  is at boot) and so the probe will get confused by any other 8390 cards.
> >>  ISA device autoprobes on a running machine are not recommended anyway. */
> >>  
> >> -int init_module(void)
> >> +int __init init_module(void)
> >>  {
> >>  	int this_dev, found = 0;
> >
> >When you anyway touches the driver I suggest to name the function
> ><module>_init, <module>_cleanup and use module_init(), module_cleanup().
> 
> Maybe not: in the ne.c driver init_module() is inside #ifdef MODULE,
> so conversion to ne_init() + module_init(ne_init) would be a no-op
> except for making the code larger. In the non-MODULE case Space.c
> calls ne_probe() directly.
The whole purpose of marking a function __init is to place in in a
section that can be discarded after init. This has the added advantage
that it kills off some ugly #ifdef MODULE / #endif as is the case for
ne.c

Even if not discarded then the code cleaniness is preferable to #ifdef /
#endif if purpose is only to save a few bytes.

Shifting to module_init(), module_cleanup() is the only right thing to
do - and the old behaviour is not even documented in LDD3 anymore.
[At least I did not find it last time I searched].

	Sam

Re: [PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Keith Owens]

Mikael Pettersson (on Sun, 11 Jun 2006 02:51:21 +0200 (MEST)) wrote:
>On Sat, 10 Jun 2006 12:13:35 -0700, Randy.Dunlap wrote:
>>On Sat, 10 Jun 2006 14:11:42 +0200 (MEST) Mikael Pettersson wrote:
>>
>>> While compiling 2.6.17-rc6 for a 486 with an NE2000 ISA ethernet card, I got:
>>> 
>>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x158) and 'ne_block_input'
>>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x176) and 'ne_block_input'
>>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x183) and 'ne_block_input'
>>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x1ea) and 'ne_block_input'
>>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x251) and 'ne_block_input'
>>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x266) and 'ne_block_input'
>>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x29b) and 'ne_block_input'
>>> 
>>> Not sure how serious this is; the driver seems to work fine later on.
>>
>>Doesn't look serious.  init_module() is not __init, but it calls
>>some __init functions and touches some __initdata.
>>
>>BTW, I would be happy to see some consistent results from modpost
>>section checking.  I don't see all of these warnings (I see only 1)
>>when using gcc 3.3.6.  What gcc version are you using?
>>Does that matter?  (not directed at anyone in particular)
>
>The messages above are from when I used gcc-4.1.1.
>With gcc-3.2.3 I only see a single warning.

Probably over enthusiastic gcc inlining.  gcc 4 will inline functions
that are not declared as inline.  That is not so bad, except that some
versions of gcc will ignore a mismatch in function attributes and
inline a __init function into normal text, generating additional
section mismatches.  For a specific example, see

http://marc.theaimsgroup.com/?l=linux-kernel&m=113824309203482&w=2

Re: [PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Randy.Dunlap]

On Tue, 13 Jun 2006 15:35:29 +1000 Keith Owens wrote:

> Mikael Pettersson (on Sun, 11 Jun 2006 02:51:21 +0200 (MEST)) wrote:
> >On Sat, 10 Jun 2006 12:13:35 -0700, Randy.Dunlap wrote:
> >>On Sat, 10 Jun 2006 14:11:42 +0200 (MEST) Mikael Pettersson wrote:
> >>
> >>> While compiling 2.6.17-rc6 for a 486 with an NE2000 ISA ethernet card, I got:
> >>> 
> >>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x158) and 'ne_block_input'
> >>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x176) and 'ne_block_input'
> >>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x183) and 'ne_block_input'
> >>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x1ea) and 'ne_block_input'
> >>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.data:isapnp_clone_list from .text between 'init_module' (at offset 0x251) and 'ne_block_input'
> >>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x266) and 'ne_block_input'
> >>> WARNING: drivers/net/ne.o - Section mismatch: reference to .init.text: from .text between 'init_module' (at offset 0x29b) and 'ne_block_input'
> >>> 
> >>> Not sure how serious this is; the driver seems to work fine later on.
> >>
> >>Doesn't look serious.  init_module() is not __init, but it calls
> >>some __init functions and touches some __initdata.
> >>
> >>BTW, I would be happy to see some consistent results from modpost
> >>section checking.  I don't see all of these warnings (I see only 1)
> >>when using gcc 3.3.6.  What gcc version are you using?
> >>Does that matter?  (not directed at anyone in particular)
> >
> >The messages above are from when I used gcc-4.1.1.
> >With gcc-3.2.3 I only see a single warning.
> 
> Probably over enthusiastic gcc inlining.  gcc 4 will inline functions
> that are not declared as inline.  That is not so bad, except that some
> versions of gcc will ignore a mismatch in function attributes and
> inline a __init function into normal text, generating additional
> section mismatches.  For a specific example, see
> 
> http://marc.theaimsgroup.com/?l=linux-kernel&m=113824309203482&w=2

Yes, I am seeing lots of that.  :(

---
~Randy

Re: [PATCH] [2.6.17-rc6] Section mismatch in drivers/net/ne.o during modpost

[Sam Ravnborg]

 
> Probably over enthusiastic gcc inlining.  gcc 4 will inline functions
> that are not declared as inline.  That is not so bad, except that some
> versions of gcc will ignore a mismatch in function attributes and
> inline a __init function into normal text, generating additional
> section mismatches.
When using -ffunction-sections (or similar) then we ask gcc to put each
function in separate sections. In this case it is OK to mix sections.
But I agree, gcc should not mix user specified sections.

> For a specific example, see
> 
> http://marc.theaimsgroup.com/?l=linux-kernel&m=113824309203482&w=2
That's a good example - thanks!

	Sam