Subject: Re: another kdump problem
To: None <Charles-Hannum@deshaw.com>
From: Niklas Hallqvist <niklas@appli.se>
List: tech-kern
Date: 09/05/1995 20:43:02
>>>>> "Charles" == Charles Hannum <Charles-Hannum@deshaw.com> writes:

Niklas>    3 When not in append mode and when running on a kernel
Niklas> that supports the sysctls, ktrace will output a header
Niklas> containing a magic number and the emulation/syscall table in
Niklas> the ktrace.out dump.

Charles> I'm not sure this is right.  When appending or using `-i' to
Charles> trace children, more than one process may be represented in
Charles> the ktrace file.  It seems to me that you need a special
Charles> record for the syscall list, tagged with a pid and recorded
Charles> internally to kdump(1) by pid, and allow for the syscall list
Charles> to be replaced (i.e. after execve(2)).

But I do, in a way.  I dump *all* syscall lists into the file
initially.  Kdump maintains a pid->emulation mapping, with a default
of the native "netbsd" emulation used when a new pid is seen.
Whenever an EMUL record is seen in the stream the mapping is changed.
This works fine as all syscall lists are available.

Further comments?

Niklas

Niklas Hallqvist	Phone: +46-(0)31-40 75 00
Applitron Datasystem	Fax:   +46-(0)31-83 39 50
Molndalsvagen 95	Email: niklas@appli.se
S-412 63  GOTEBORG	WWW:   Here
Sweden