Re: [bisected] 2.6.29-rc1 Firefox crashing on page load

[Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>]

Justin Madru wrote:
> Hiroshi Shimamoto wrote:
>> Justin Madru wrote:
>>   
>>> Hiroshi Shimamoto wrote:
>>>     
>>>> Justin Madru wrote:
>>>>   
>>>>       
>>>>> Hiroshi Shimamoto wrote:
>>>>>     
>>>>>         
>>>>>> Justin Madru wrote:
>>>>>>       
>>>>>>           
>>>>>>> Ok, finally bisected the bug, but the commit isn't related to networking!
>>>>>>> I did: git revert 4217458dafaa57d8e26a46f5d05ab8c53cf64191
>>>>>>> on current git and that fixed the bug.
>>>>>>>
>>>>>>> By the way, how do I undo my git revert, so I'm back to a pristine tree?
>>>>>>> I want to drop my changes - the revert.
>>>>>>> Also how do I find the commit that merged/pulled in this commit?
>>>>>>>
>>>>>>> commit 4217458dafaa57d8e26a46f5d05ab8c53cf64191
>>>>>>> Author: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
>>>>>>> Date: Fri Dec 5 17:17:09 2008 -0800
>>>>>>>
>>>>>>> x86: signal: change type of paramter for sys_rt_sigreturn()
>>>>>>>
>>>>>>> Impact: cleanup on 32-bit
>>>>>>>
>>>>>>> Peter pointed this parameter can be changed.
>>>>>>>
>>>>>>> Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
>>>>>>> Signed-off-by: Ingo Molnar <mingo@elte.hu>
>>>>>>>         
>>>>>>>             
>>>>>> Hi,
>>>>>> thanks for reporting.
>>>>>>
>>>>>> I'm not sure why this commit affects.
>>>>>> Can you check vmlinux? size, objdump, etc.
>>>>>> On my environment, the generated code looks same.
>>>>>>
>>>>>> Thanks,
>>>>>> Hiroshi
>>>>>>
>>>>>>       
>>>>>>           
>>>>> I tried to do an objdump but it gave an error. How am I suppose to do an 
>>>>> objdump?
>>>>>
>>>>> $ objdump -x /boot/vmlinuz-2.6.29-rc2-git
>>>>> objdump: /boot/vmlinuz-2.6.29-rc2-git: File format not recognized
>>>>> $ readelf -a /boot/vmlinuz-2.6.29-rc2-git
>>>>> readelf: Error: Unable to seek to 0xc031f2eb for section headers
>>>>> readelf: Error: Not an ELF file - it has the wrong magic bytes at the start
>>>>> $ ls -la /boot/vmlinuz-2.6.29-rc2-git
>>>>> -rw-r--r-- 1 root root 2286480 2009-01-19 18:44 /boot/vmlinuz-2.6.29-rc2-git
>>>>>     
>>>>>         
>>>> not vmlinuz, vmlinux.
>>>> vmlinuz is a compressed kernel, it is not ELF file.
>>>> Usually vmlinux is generated in top of source directory.
>>>> I can see disassembled image with objdump -d vmlinux.
>>>>
>>>> Thanks,
>>>> Hiroshi
>>>>
>>>>   
>>>>       
>>> Sorry, didn't know. I tried objdump -d vmlinux, but the resulting file 
>>> is ~50MB!
>>> Is there a smaller section you're interested in, because that would be 
>>> hard to send.
>>> Or am I doing it wrong again?
>>>     
>> that's OK. No need to send the disassembled result.
>> You can diff disassembled file.
>> I guess your GCC generates different code when the patch is reverted.
>>
>> Thanks,
>> Hiroshi
>>
>>   
> Ok, I tried, I really did. But no mater what I did the diff was even bigger
> if not the same size as the original dump! I even tried to cut out the 
> first two fields with
>     cut -f3- dump-revert > dump-revert2
> but that wouldn't shrink the diff size (100,000+ lines).

I'm sorry for lack of explanations.
The patch changes sys_rt_sigreturn(), it means GCC might generate
different code for that function. And might be different size, it causes
all codes after sys_rt_sigreturn() will looks different.

So I think you can see the difference begins from sys_rt_sigreturn().

> 
> Does gcc output fluctuate this wildly with such little change?
> Is there a config option that allows gcc to randomize the output?
> I recompiled just to make sure, and even two recompiles of the same 
> source (no revert)
> resulted in a different sha1 hash (didn't test the objdump diff).
> The only thing I can think of is that I compile with -O3 instead of -O2.
> 
> I'm sorry, but unless you have any suggestions I've decided to just upload
> the vmlinux files to my server (<40KB up speed) 
> http://jdserver.homelinux.org/bugreports/002/
> and rapidshare
> http://rapidshare.com/files/186890846/vmlinux.tar.bz2.html
> http://rapidshare.com/files/186890848/data-norevert.tar.bz2.html
> http://rapidshare.com/files/186890849/data-revert.tar.bz2.html

thanks for uploading vmlinux, will see the difference.
And sorry again, actually, it's just for my curiosity.
I'd like to know what difference this is.
On my environment GCC(4.2.4-1ubuntu3 and 4.3.2) generates same binary...

> 
> By the way Ingo has already added the revert to tip/master
> http://git.kernel.org/?p=linux/kernel/git/tip/linux-2.6-tip.git;a=commit;h=1d03950243904c89b8854c10516932b029c2a8fa

Yeah, Ingo knows what GCC does. I don't know so much.

Thanks,
Hiroshi