Message-ID: <481B8321.1000305@goop.org>
Date:	Fri, 02 May 2008 14:09:53 -0700
From:	Jeremy Fitzhardinge <jeremy@...p.org>
To:	Adrian Bunk <bunk@...nel.org>
CC:	"Pallipadi, Venkatesh" <venkatesh.pallipadi@...el.com>,
	David Miller <davem@...emloft.net>, trini@...nel.crashing.org,
	mingo@...e.hu, tglx@...utronix.de, hpa@...or.com,
	akpm@...ux-foundation.org, linux-kernel@...r.kernel.org,
	"Siddha, Suresh B" <suresh.b.siddha@...el.com>,
	Linus Torvalds <torvalds@...ux-foundation.org>
Subject: Re: huge gcc 4.1.{0,1} __weak problem

Adrian Bunk wrote:
> On Wed, Apr 30, 2008 at 05:49:46AM -0700, Pallipadi, Venkatesh wrote:
>   
>>  
>>     
>>> -----Original Message----- From: David Miller 
>>> From: Venki Pallipadi <venkatesh.pallipadi@...el.com> Date: Tue, 29 
>>> Apr 2008 18:31:09 -0700
>>>
>>>       
>>>> Some flavors of gcc 4.1.0 and 4.1.1 seems to have trouble 
>>>>         
>>> understanding
>>>       
>>>> weak function definitions. Calls to function from the same 
>>>>         
>>> file where it is
>>>       
>>>> defined as weak _may_ get inlined, even when there is a 
>>>>         
>>> non-weak definition of
>>>       
>>>> the function elsewhere. I tried using attribute 'noinline' 
>>>>         
>>> which does not
>>>       
>>>> seem to help either.
>>>>
>>>> One workaround for this is to have weak functions defined in 
>>>>         
>>> different
>>>       
>>>> file as below. Other possible way is to not use weak 
>>>>         
>>> functions and go back
>>>       
>>>> to ifdef logic.
>>>>
>>>> There are few other usages in kernel that seem to depend on 
>>>>         
>>> weak (and noinline)
>>>       
>>>> working correctly, which can also potentially break and 
>>>>         
>>> needs such workarounds.
>>>       
>>>> Example -
>>>> mach_reboot_fixups() in arch/x86/kernel/reboot.c is one such 
>>>>         
>>> call which
>>>       
>>>> is getting inlined with a flavor of gcc 4.1.1.
>>>>
>>>> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@...el.com>
>>>> Signed-off-by: Suresh Siddha <suresh.b.siddha@...el.com>
>>>>         
>>> This sounds like a bug.  And if gcc does multi-file compilation it
>>> can in theory do the same mistake even if you move it to another
>>> file.
>>>
>>> We need something more bulletproof here.
>>>
>>>       
>> The references here
>> http://gcc.gnu.org/ml/gcc-bugs/2006-05/msg02801.html
>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27781
>>
>> seem to suggest that the bug is only with weak definition in the same
>> file.
>> So, having them in a different file should be good enough workaround
>> here.
>> ...
>>     
>
> A workaround here is the wrong solution since this isn't the only place 
> that suffers from this issue.
>
> We currently give a #warning for 4.1.0.
> But not for 4.1.1.
> (Accordingto the bug >= 4.1.2 is fixed.)
>
> And a #warning is not enough.
>
> The huge problem is that "empty __weak function in the same file and 
> non-weak arch function" has recently become a common pattern with 
> several new usages added during this merge window alone.
>
> And the breakages can be very subtle runtime breakages.
>
> I see only the following choices:
> - remove __weak and replace all current usages
> - move all __weak functions into own files, and ensure that also happens
>   for future usages
> - #error for gcc 4.1.{0,1}
>   

- make __weak also include noinline.  I think that's sufficient (at 
least it was when I encountered a gcc bug with these symptoms last year 
or so).

    J
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists