I'm a little skeptical that this is a bug outside of mx or ib. Scott do you get the same behavior from valgrind if you use tcp? Or even get the same hang on sighup if you use tcp?

-sam

On Jan 18, 2007, at 3:33 PM, Murali Vilayannur wrote:

Hi Scott,
Just a shot into the wild...
Can you do
% mv /lib/tls /lib/tls.old

and then rerun the server with valgrind..
Quite possibly this is a valid bug in pvfs2, just wanted to find out
if the error message is any different after you do the above..
Don't forget to revert the rename
% mv /lib/tls.old /lib/tls
after you are done.
thanks,
Murali

On 1/18/07, Scott Atchley <[EMAIL PROTECTED]> wrote:
On Jan 18, 2007, at 3:07 PM, Sam Lang wrote:

>> To answer my own question, send a HUP, then continue the process
>> in GDB. :-)
>
> You can also do 'signal SIGHUP' at the gdb prompt.

That is good to know. Thanks.

> So this looks like its just related to the invalid free in the mx
> code (and the ib code I guess).  You could run it in valgrind to
> verify that probably.  Can you set the map pointer to null and
> check on the next attempt to free it?
>
> -sam

When I try to use valgrind, the server fails to start with:

...
[D 01/18 15:35] bmi_mx: memory used at end of initialization 17200540.
[D 01/18 15:35] bmi_mx: entering BMI_mx_set_info.
[D 01/18 15:35] bmi_mx: leaving BMI_mx_set_info.
[D 01/18 15:35] bmi_mx: entering BMI_mx_set_info.
[D 01/18 15:35] bmi_mx: leaving BMI_mx_set_info.
[E 01/18 15:35] TROVE:DBPF:Berkeley DB: read: 0x1f9fc7f0, 4096: Bad
address <-----------
[E 01/18 15:35] TROVE:DBPF:Berkeley DB: //scratch/atchley/4e901077/
collection_attributes.db: Bad address
[E 01/18 15:35] src/server/pvfs2-server.c line 1077: Error
initializing filesystem pvfs2-fs
[E 01/18 15:35]         [bt] ./sbin/pvfs2-server [0x8054db6]
[E 01/18 15:35]         [bt] ./sbin/pvfs2-server [0x8054661]
[E 01/18 15:35] [bt] ./sbin/pvfs2-server(main+0x4b0) [0x8054048]
[E 01/18 15:35]         [bt] /lib/tls/libc.so.6(__libc_start_main
+0xf4) [0x1ba42974]
[E 01/18 15:35]         [bt] ./sbin/pvfs2-server(aio_cancel64+0x69)
[0x8053971]
[E 01/18 15:35] Error: Could not initialize server interfaces; aborting.
[E 01/18 15:35] Error: Could not initialize server; aborting.
[D 01/18 15:35] bmi_mx: entering BMI_mx_finalize.
...

Setting the inout_parameter pointer to NULL has no effect.

Scott

_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers



_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers

Reply via email to