Re: [patch, avr] PR31124: Support rodata in flash for more AVR devices

[Georg-Johann Lay <avr@gjlay.de>]

Am 11.12.23 um 19:33 schrieb Georg-Johann Lay:
> Am 11.12.23 um 16:41 schrieb Nick Clifton:
>> Hi Johann,
>>
>>> This patch adds two new emulations in order to support rodata in
>>> flash for more devices.
>>
>> Patch approved - please apply.  (Or let me know and I will apply it 
>> for you).
> 
> Hi Nick,
> 
> I do not have write access to Binutils, at least I never asked for it.
> 
>>> What I am unsure about 
>>
>> Err - didn't you write this patch ?
> 
> I did.  But even after studying the documentation many things are
> not 100 clear.
> 
>>> is the ABSOLUTE in avrxmega2_flmap.x:
>>>
>>>    .rodata  __rodata_start   :  AT (__rodata_load_start)
>>>    {
>>>      ...
>>>       __rodata_end = ABSOLUTE(.) ;
>>>    }  > rodata
>>>
>>> No other places in the ld scripts use ABSOLUTE, e.g. for __data_end in:
>>>
>>>    .data :
>>>    {
>>>       ...
>>>       PROVIDE (__data_end = .) ;
>>>    }  > data AT> text
>>
>> The difference is, presumably, that the symbols are going to be used
>> in different ways.  The __data_end symbol, for example is more like a
>> size-of-the-data-section symbol whereas the __rodata_end symbol is more
>> of a no-more-read-only-data-after-this-address kind of symbol.
>>
>> Cheers
>>    Nick
> 
> The startup code copies __data_load_start...__data_load_end-1 to
> __data_start...__data_end-1, so the date size is the difference.
> 
> There is currently no use for the rodata symbols; I added them in
> case a user wants to compute CRC etc.
> 
> The hardware has the assertion that __rodata_load_start - __rodata_start
> must be 0 mod 0x8000.  And it should always be the case that
> __rodata_load_end - __rodata_load_start = __rodata_end - __rodata_start.
> I am just unsure if that's always the case, e.g. when the user starts
> moving stuff around with objcopy or whatnot.

When the user uses objcopy or something to patch the result from the
final link, then the ld script cannot diagnose that with ASSERT.
So presumably I am overthinking all this...

Johann

> I tested with -Ttext; and when .text reaches into .rodata, then both
> LMA and VMA are shifted in sync like it's supposed to be.
> 
> 
> Johann