[Crash-utility] [PATCH v2 0/3] Kdump core analysis support for PPC32

[Suzuki K. Poulose]

On 01/04/12 01:40, Dave Anderson wrote:
>
>
> ----- Original Message -----
>> The following series implements the kdump core analysis support
>> for PPC32. I have posted the KDUMP kernel support patches for PPC440x
>> here :
>>   http://lists.ozlabs.org/pipermail/linuxppc-dev/2011-December/094994.html
>>
>> You need upstream git snapshot of kexec-tools for kdump support on
>> PPC440x.
>>
>> These patches are based on crash-6.0.2
>>
>> ---
>>
>> Suzuki K. Poulose (3):
>>        [ppc] Enable stack trace display for KDUMP cores
>>        [ppc][netdump] Read register set from ELF Note
>>        [ppc] Support PPC32 Core analysis on PPC64 host
>>
>>
>>   configure.c |   14 ++++++++
>>   netdump.c   |   77 +++++++++++++++++++++++++++++++++++++++++++++
>>   ppc.c       |  101
>>   +++++++++++++++++++++++++++++++++++++++++++++++++++--------
>>   3 files changed, 178 insertions(+), 14 deletions(-)
>>
>> --
>> Suzuki Poulose
>
>
> Like the target=X86 and target=ARM builds on x86_64, I installed
> several extra required libraries (glibc-devel.ppc, libgcc.ppc,
> ncurses-libs.ppc, ncurses-devel.ppc and zlib-static.ppc), and
> I finally got it to build OK with "setarch ppc make target=PPC".
>

> However, I forgot the "setarch ppc" the first time I built it, and
> it went ahead and built everything with both "-m64" and "-m32" set,
> ending up with an (unwanted) 64-bit binary.  So it might be worth
> preventing that in configure.c?  I'll leave that up to you.

We have to fix this in gdb-7.3.1.patch and the Makefile, which hard codes
"-m64" based on the value of 'arch'.  Have done this in the next version.

>
> But there are several compiler complaints (some are actual errors)
> with the updated ppc.c file that need to be fixed:
>
> ...
> cc -c -g -DPPC -m32 -D_FILE_OFFSET_BITS=64 -fPIC -DGDB_7_3_1  ppc.c -Wall -O2 -Wstrict-prototypes -Wmissing-prototypes -fstack-protector
> ppc.c: In function ‘ppc_kdump_stack_frame’:
> ppc.c:1142: warning: too many arguments for format
> ppc.c:1167: warning: ‘return’ with a value, in function returning void
> ppc.c: In function ‘get_ppc_frame’:
> ppc.c:1259: warning: unused variable ‘offset’
> ppc.c: In function ‘ppc_dump_irq’:
> ppc.c:1354: warning: the address of ‘typename’ will always evaluate as ‘true’
> ppc.c:1468: warning: the address of ‘typename’ will always evaluate as ‘true’
> ppc.c: At top level:
> ppc.c:1239: warning: ‘ppc_get_pc’ defined but not used
> ppc.c:1223: warning: ‘ppc_get_sp’ defined but not used
> ...
>
> Can you fix these so that "make warn" will compile it cleanly?

Yep, done. I will send the new version with the changes.

Thanks
Suzuki