In message <041113212214.M0119131@mirage.ceres.dti.ne.jp>, you write: 

-> In article <20041111184647.25ac8f58.flynn@energyhq.es.eu.org>
-> flynn@energyhq.es.eu.org wrote:
-> 
-> > I have the same hardware as you, and this problem has been there for a
-> > long time. AFAIK this only happens with dynamic executables and, as you
-> > say, seems to be random (from a user's perspective at least).
->  :
-> > > Inspecting such a core dump with gdb always leads to the same
-> > > result: the program pointer is in illegal address space
-> > > (at least I believe so). The address is always 0x5ffe????.
-> 
-> I think address 0x5ffe0000 is used by ld.elf_so.
-> BTW, -current src/libexec/ld.elf_so/Makefile implicitly
-> specifies "DBG= -O3 -fomit-frame-pointer" on all ports
-> but I wonder if it might cause problems.

I wonder if we don't support some of the cache 'quirks' of some of
the early-rev R4000 chips -- I know some early R4600 chips had bugs
in this area and maybe the R4000s did as well.

As a first cut it might be good to look at the Linux R4000 cache code
and see if it does anything 'special' for the R4000 revs where the
problems are being reported.

--rafal

----
Rafal Boni                                                     rafal@pobox.com
  We are all worms.  But I do believe I am a glowworm.  -- Winston Churchill