[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Help needed to debug an armel build/unittest failure



On 3/7/10, Iustin Pop <iusty@k1024.org> wrote:
>  >So it looks like what we're actually seeing here is one bit of corruption.
>  >
>  >I still strongly suspect a compiler bug.  Probably the compiler's emulation
>  >of 64-bit arithmetic is broken in some way.
>
>  First, are you aware of 64-bit arithmetic issues on arm(el)?

No, I know of no 64-bit arithmetic issues in any armel GCC. OpenSSL
uses 64-bit int math and its testsuite passes fine. The main ARM issue
is requiring 4-byte alignment of 4-byte word accesses (and 2- of 2-)
with silent data corruption if you access an unaligned word.
You can test for that by going
  # echo 5 > /proc/cpu/alignment
and running your test again. On unaligned access the program will then
die with SIGBUS instead of merrily going on its way with corrupted
values.
I thin

The main ARM and armel idiosyncracies are listed at
wiki.debian.org/ArmEabiFixes

However, there are four versions of GCC - 4.[1234] - installed here if
you'd like to probe that theory.

    M


Reply to: