Re: [PATCH] move BUG_TABLE into RODATA

["Jan Beulich" <jbeulich-Et1tbQHTxzrQT0dZR+AlfA@public.gmane.org>]

>>> Sam Ravnborg <sam-uyr5N9Q2VtJg9hUCZPvPmw@public.gmane.org> 22.04.08 20:07 >>>
>Hi Jan.
>
>> --- linux-2.6.25/arch/x86/kernel/vmlinux_64.lds.S	2008-04-17 04:49:44.000000000 +0200
>> +++ 2.6.25-bug-table-rodata/arch/x86/kernel/vmlinux_64.lds.S	2008-03-04 11:14:13.000000000 +0100
>> @@ -19,7 +19,7 @@ PHDRS {
>>  	data PT_LOAD FLAGS(7);	/* RWE */
>>  	user PT_LOAD FLAGS(7);	/* RWE */
>>  	data.init PT_LOAD FLAGS(7);	/* RWE */
>> -	note PT_NOTE FLAGS(4);	/* R__ */
>> +	note PT_NOTE FLAGS(0);	/* ___ */
>>  }
>
>Is this intentional in this patch?

Yes. While it may be possible to do this separately, it seemed safer to
me to keep it together since I only tested it this way.

>>  SECTIONS
>>  {
>> @@ -40,16 +40,14 @@ SECTIONS
>>  	_etext = .;		/* End of text section */
>>    } :text = 0x9090
>>  
>> +  NOTES :text :note
>> +
>>    . = ALIGN(16);		/* Exception table */
>>    __ex_table : AT(ADDR(__ex_table) - LOAD_OFFSET) {
>>    	__start___ex_table = .;
>>  	 *(__ex_table)
>>    	__stop___ex_table = .;
>> -  }
>> -
>> -  NOTES :text :note
>> -
>> -  BUG_TABLE :text
>> +  } :text = 0x9090
>
>And I do not see the 0x9090 justified.
>Is this something to do with the fact that 0x90 equals NOP on x86?

Yes, for consistency I think this ought to be repeated here - while
not strictly necessary (__ex_table is data, hence padding with
zeroes rather than NOPs would be fine), it seems more safe to have
it here so that we don't catch occasional linker bugs. Ultimately I
would think __ex_table should go into RODATA, too, which would
eliminate the question.

Jan