About 6 weeks ago there were a couple messages on the list
about glibc 2.2.x and embedded PowerPC. Has anyone made more
progress?
I'm trying to use glibc2.2.3 on a Walnut board, and cannot
start any processes. The MVista glibc-2.1.3 and ld.so.1 work
fine on this platform. My problem with glibc 2.2.3 looks similar
to the problems described a few weeks ago (fault when linking
libraries).
Just for reference, heres what I did to glibc 2.2.3:
I had to move sysdeps/powerpc/fclrexcept.c to the fpu subdirectory,
as it deals with the FPU (sysdeps/generic/fclrexcept.c will be used
instead). I also had to remove the fenv_const and fe_nomask references
from the Makefile.
So far as I can see __elf_machine_runtime_setup() in
sysdeps/powerpc/dl-machine.c has been updated to know that
not all PPCs have 32 byte cache lines, so I left it alone.
memset.S does still have an assumption about cache line
size ("subfic rALIGN, rALIGN, 0x20"), so I moved it aside
to let the generic version be used.
** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/
The problem we ran into with 2.2.2 was the optimized memset wasn't
aligned properly. We reverted back to the generic memset and haven't
had any trouble on 8xx. I am not aware of any similar issues on the 405,
but...
Denton Gentry wrote:
About 6 weeks ago there were a couple messages on the list
about glibc 2.2.x and embedded PowerPC. Has anyone made more
progress?
I'm trying to use glibc2.2.3 on a Walnut board, and cannot
start any processes. The MVista glibc-2.1.3 and ld.so.1 work
fine on this platform. My problem with glibc 2.2.3 looks similar
to the problems described a few weeks ago (fault when linking
libraries).
Just for reference, heres what I did to glibc 2.2.3:
I had to move sysdeps/powerpc/fclrexcept.c to the fpu subdirectory,
as it deals with the FPU (sysdeps/generic/fclrexcept.c will be used
instead). I also had to remove the fenv_const and fe_nomask references
from the Makefile.
So far as I can see __elf_machine_runtime_setup() in
sysdeps/powerpc/dl-machine.c has been updated to know that
not all PPCs have 32 byte cache lines, so I left it alone.
memset.S does still have an assumption about cache line
size ("subfic rALIGN, rALIGN, 0x20"), so I moved it aside
to let the generic version be used.
** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/