Thanks for the detailed explanation.
I've been using the word 'segfault' in tinyos for everything
that would look like a segfault in UNIX systems such as
null pointers, pointers out-of-bound, etc...
although I am not claiming that I remembered the exact definition
that I think I've read some time ago....
Thanks
- jpaek
Philip Levis wrote:
On Nov 10, 2007, at 9:07 PM, Jeongyeup Paek wrote:
Maybe there is a misunderstanding...
If you write a code that has segmentation fault,
then the mote will restart.
And I don't think this 'behavior' is a bug
since it is the only thing it can do.
To be more precise, there is no such thing as a "segmentation fault" in
TinyOS; a segfault is when a program accesses a memory address for which
there is no valid virtual memory segment, and so no valid
virtual->physical address mapping. In UNIX/Windows/etc., typically the
OS does not allow you to define a segment that covers the first page of
the address space, so that null pointers always segfault.
When you have no virtual memory (read: microcontrollers), this becomes a
bit trickier. It turns out that the low physical addresses are memory
mapped registers. So if you dereference a null pointer, you end up
accessing things like the register that controls whether interrupts are
enabled or not. This can sometimes appear as a reset.
Another thing that can cause a reset is if you have a stack overflow. It
turns out that the reset vector is sometimes at code address 0. So if
you have an array on the stack of size X and write X+Y 0x00 bytes, if Y
is big enough, you will overwrite the return address to be 0.
This second case appeared on micaz nodes pre 2.0.2; the way the SPI bus
was implemented, if you send a packet of length 0 it would send a packet
of length 65535. This is a pretty bad bug: writing 65535 bytes of value
0x00 would blow away the heap and stack, causing the return address on
the stack to be 0x0000 and... a reset. It turned out that this was
caused by corrupted 802.15.4 packets whose PHY length byte was 0; this
is clearly a nad length, but the code never considered it might happen,
so it triggered the SPI bug.
It's stuff like this that has made core start leaning towards default-on
but disableable Deputy support.
Phil
--
Jeongyeup Paek
Ph.D. student
Embedded Networks Laboratory
Department of Computer Science
University of Southern California
http://enl.usc.edu/~jpaek
_______________________________________________
Tinyos-help mailing list
[email protected]
https://www.millennium.berkeley.edu/cgi-bin/mailman/listinfo/tinyos-help
