On 21/07/17 16:48, Philippe Mathieu-Daudé wrote:
I submitted this patch because I spent some time debugging while QEMU was
failing silently using a MIPS kernel image which used to work, after
realizing I was in an incorrect build_dir using qemu-system-mipsel to load
a big endian image and felt stupid [1]. This dumb error can happen to other
people so I added this warning here.
Been there too. This is why we try loading images twice in pseries.
I was not aware of the ELF_LOAD_WRONG_ENDIAN related code, and at least the
MIPS arch is not using it.
As I can see in MAINTAINERS, sPAPR is "Supported" meaning "Someone is
actually paid to look after this", while there is no such paid person for
the MIPS part.
It seems each arch had a different way to load images and hw/core/loader.c
was an effort to merge common code but mostly "Supported" arch are using
it.
afaict MIPS calls load_elf() in 4 places, each checks for the return value
and already prints the message, how come that that message is not enough?
What would probably make sense here is if MIPS also printed the return code
from load_elf() but in any case it is easily visible from gdb.
While your revert does fixes your sPAPR warning issue, looking at the
problem roots I think the correct fix is to improve the MIPS port and
eventually the less loved archs to unify the loader.c calls and avoid such
problems.
I don't object reverting this patch for 2.10 and improve the loader.c usage
during 2.11 cycle, I only wonder if this is another corporate/hobbyist>
conflict of interest with corporate crushing on hobbyist instead of
Come on mate...