Linker Set Based Initialization

Tue Dec 8 17:13:20 UTC 2015

Hello Joel, 

before you start with wild guessing, please look at the patch: 

https://git.rtems.org/rtems/commit/?id=b618d8cfc54f84d4ed03dc7b7fa510c872e6128a 

----- Am 8. Dez 2015 um 16:45 schrieb Joel Sherrill <joel at rtems.org>: 

> On Tue, Dec 8, 2015 at 9:09 AM, Sebastian Huber <
> sebastian.huber at embedded-brains.de > wrote:

>> On 08/12/15 16:03, Joel Sherrill wrote:

>>> What BSPs/architectures have you tested?

>> I temporarily moved the splinkersets01 test to the samples/ticker and tested
>> that all BSPs build and link this test.

>> I executed the splinkersets01 test on sis, psim and arm_realview_pbx_a9_qemu.

>>> Is this something that breaks on a per BSP basis? or per architecture basis?
>>> I am assuming that since it is linker based, each BSP could have broken
>>> linkcmds.
>>> Is that right?

>> It breaks on a per linker command file basis. Since all the maintained BSPs use
>> a linkercmds.base, which shouldn't be a big issue.

> That means a LOT of the BSPs are broken.

Why do you think BSPs are broken? If you look at my commit, then you will see that I edited every linker command file by hand! Please note that there are no changes for the ARM (except the GBA BSP, which is a removal candidate if you ask me), SPARC and i386. They already use the linker sets for the libbsd. 

> You have defined maintained in your own way.

You can use other definitions and end up likely with the same set of BSPs. 

> There are only a handful of architectures with linkcmds.base in them:

> ./or1k/shared/startup/linkcmds.base
> ./arm/shared/startup/linkcmds.base
> ./m68k/shared/startup/linkcmds.base
> ./powerpc/tqm8xx/startup/linkcmds.base
> ./powerpc/gen5200/startup/linkcmds.base
> ./powerpc/shared/startup/linkcmds.base
> ./sparc/shared/startup/linkcmds.base

> I am not sure how many BSPs within arm, m68k, or powerpc actually use the
> linkcmds.base.

> By my count, 13 of 94 BSP families have linkcmds.base.

>> For requirements on the linker command file, see new chapter in user manual.
>> However BSPs should not deal with this in copy and paste linker command files
>> and instead use a linkercmds.base file.

> So 85% of the BSP families don't use linkcmds.base and by the above statement,
> they must immediately be migrated to linkcmds.base.

No, nothing must be migrated. In fact such a migration would be very risky. You just need two section descriptions in the linker command file (see user manual chapter). 

> Unless you have a plan to address this problem, I am on the side of rejecting
> the
> part of this patch that changes the initialization. And the issue must be
> addressed
> before this can be merged.

I am not aware of any issues, except the dependency on the GNU linker. 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20151208/05d681ac/attachment.html>