[PATCH 11/11] libdl: Add powerpc large memory and small data support.

[Chris Johns]

On 5/2/19 5:32 pm, Sebastian Huber wrote:
> On 05/02/2019 07:19, Sebastian Huber wrote:
>> On 05/02/2019 02:43, chrisj at rtems.org wrote:
>>> +    /*
>>> +     * Set bsp_section_sdata_bss_maximum_size to 1 to use the maximum, eg:
>>> +     *
>>> +     *  bsp_section_sdata_bss_maximum_size = 1;
>>> +     */
>>> +    bsp_section_sdata_bss_default_size =
>>> +        DEFINED(bsp_section_sdata_bss_maximum_size) ? 65536 : 0;
>>> +    /*
>>> +     * Set bsp_section_sdata_bss_size to the system's sdata and bss data
>>> size, eg:
>>> +     *
>>> +     *  bsp_section_sdata_bss_maximum_size = 4096;
>>> +     */
>>> +    bsp_section_sdata_sbss_size = DEFINED(bsp_section_sdata_bss_size) ?
>>> +        bsp_section_sdata_bss_size : bsp_section_sdata_bss_default_size;
>>
>> I would not make it more complicated than necessary. The systems using libdl
>> probably all have network support and file systems. For them wasting 64KiB
>> shouldn't be an issue. I would rename the bsp_section_sdata_bss_maximum_size
>> to something like bsp_section_use_maximum_sdata_size since it is actually not
>> a size.

The "small data" region is formally sdata and sbss sections which is why I used
the extended labeling. I wanted to make it clear this allocates space for sdata
_and_ sbss.

I agree there will be a hard split of BSPs, those that define the full size and
those who do not use libdl. The only issue is a small amount of data is needed
for some of the tests in the testsuite to pass. I expect a run of excluding
libdl tests. This is an unfortunate situation and I am not sure how else to
handle it.

I have not updated which BSPs get the full sdata size. I am not sure which need it.

> Maybe we can simply use:
> 
> /*
>  * Define bsp_section_sdata_fixed_size to set the size of the small-data area to
> a fixed size as specified, otherwise the size of the small-data area is defined
> by the section content.
>  *
>  * bsp_section_sdata_fixed_size = 65536;
>  */

I provided the max value to avoid errors related to getting this value correct
in a number of places.

Chris