[Patch,AVR]: Fix PR13697: Wrong symbols with --gc-sections

Georg-Johann Lay avr@gjlay.de
Tue May 29 06:31:00 GMT 2012


Hans-Peter Nilsson schrieb:
> On Mon, 28 May 2012, Georg-Johann Lay wrote:
>> This bug is because an empty .data will be removed by the ld machinery.
> 
> I think that bug has been fixed, at least, if you add a
> symbol-setting (one without PROVIDE) which seems slightly better
> than decorating input .data sections with KEEP, which just stops
> --gc-sections for *any* input .data.
> 
>> ld/
>>     PR 13697
>>     * scripttempl/avr.sc (.data): Keep it.
>>
> 
> (bah, I have to cutnpaste; pine doesn't quote attachments):
>>     *(.data)
>> +    /* --gc-sections will delete empty .data. This leads to wrong start
>> +       addresses for subsequent sections because -Tdata= from the command
>> +       line will have no effect, see PR13697.  Thus, keep .data  */
>> +    KEEP (*(.data))
> 
> "KEEP" is a decoration, not a separate statement.  At a minimum,
> drop the first "*(.data).  It's at least confusing, but might
> add a second copy, I'm not sure.

I am definitely no linker hero... it's just like

     *(.vectors)
     KEEP(*(.vectors))

or

    *(.init0)  /* Start here after reset.  */
     KEEP (*(.init0))
     *(.init1)
     KEEP (*(.init1))

and many others. So all this in the avr ls script is plain wrong?

> Try without --gc-sections and  gc:able .data.

er... sorry, would you outline this? I searched for "gc:able"
documentation but with no avail.

> Though, adding a symbol should do it, at least for a corrected
> linker.

You mean adding a symbol to .data by letting the compiler add a symbol
in .data for each module?



More information about the Binutils mailing list