Re: [linuxdc-dev]Problem with porting SoX

[Adam J. R. <ad...@yg...>]

	I am not a SuperH developer, but here is some advice on how
you might want to proceed.

>From: Adrian McMenamin <ad...@mc...>

>./sox: Killing process "sox" due to unaligned access
>To be honest I still don't know enough about Linux development to know what 
>"unaligned access" is other than it sounds like a linker fix up problem.

	I imagine the error message came from do_address_error at
line 473 of linux-2.5.1-pre1/arch/sh/kernel/traps.c (may be different
for other kernels).

	By the way, here is a handy shell command for searching
for error messages or other strings in the kernel source tree.
When you have a mysterious error message, it is a good idea to try
tis on the kernel and other potentially related source trees if you
have them handy, like the sources for glibc and the applications
that you were running.

% find /usr/src/linux -type f | xargs egrep 'due to unaligned access'
/usr/src/linux/arch/sh/kernel/traps.c:          printk(KERN_NOTICE "Killing process \"%s\" due to unaligned access\n", current->comm);

	"unaligned access" sounds like an attempt to do a 32-bit
access to an address that is not a multiple of 4 or 16-bit memory
access to an address that is not a multiple of 2.  Most architectures
allow unaligned access from hardware, but some, apparently including
SuperH, do not.  From the source code near that error message in
traps.c, it looks like the kernel emulates around ordinary unaligned
reads and writes, but that there are a number number of cases where the
kernel has to give up and generate this error, such as when jumping to
an address that is an odd number.  So, probably your next step is to
add a few printk's to the different points of do_address_error that
jump or fall through to the "uspace_segv:" label so you can see
which case is actually generating the error.  You might also want
to print out the program counter ("regs->pc") to try to figure out
what address is causing the problem (check with gdb or something).
At the very least, you should get some idea of whether or not you
were accessing an otherwise legitimate unaligned address or had
jumped to a garbage address.

Adam J. Richter     __     ______________   4880 Stevens Creek Blvd, Suite 104
ad...@yg...     \ /                  San Jose, California 95129-1034
+1 408 261-6630         | g g d r a s i l   United States of America
fax +1 408 261-6631      "Free Software For The Rest Of Us."