Backport of ARM cache, RTL and TMS570 to RTEMS 4.11 Was: Simple question about RTL

Pavel Pisa pisa at cmp.felk.cvut.cz
Tue Oct 4 21:47:49 UTC 2016


Hello Chris,

On Tuesday 04 of October 2016 23:10:20 Chris Johns wrote:
> On 04/10/2016 09:51, Pavel Pisa wrote:
> > Hello Chris and Joel,
> >
> > I would like to correct my mistake which breaks RTEMS 4.11 build for
> > non-ARM architectures, as fast as possible. Proposed solution on devel
> > list.
> >
> > [PATCH] libdl/rtl-obj.c: synchronize cache should not depend on
> > CPU_CACHE_LINE_BYTES.
> >
> > [PATCH] bsps/arm: do not introduce CPU_CACHE_LINE_BYTES in 4.11 and
> > correct CPU_STRUCTURE_ALIGNMENT.
> >
> > I have checked 4.11 build with patches for Raspberry Pi 2 and TMS570.
> > including RTL functionality on targets. I have tested 4.11 that it
> > works on TMS570 with lwIP in build without initialization when
> > loaded to SDRAM with external init and standalone RTEMS Flash version
> > which includes complete RTEMS driven initialization.
> >
> > I have tested that i386 builds and checked to run ticker,
> > my dlopen and PCI i82557b/fxp1 betworking tests in QEMU.
> > i386 build has been done mainly to check that I have not
> > introduced problem to non-ARM architectures this time.
>
> Great.
>
> > Do you agree with fixes?
>
> Yes.
>
> > What is your opinion for CPU_CACHE_LINE_BYTES in 4.11.
> > Do you agree/prefer that it should not be introduced back
> > (which I have done by mistake) and kept as 4.12 specific?
>
> I agree it should not be introduced back into 4.11.

OK, I have pushed changes to 4.11 and I have changed master
to use the same rtems_cache_get_maximal_line_size() in RTL
as is used 4.11 now to have support consistent between branches
and have chance to catch eventual problem even on master.

Best wishes,

              Pavel




More information about the devel mailing list