svn commit: r219138 - head/usr.bin/kdump

[Robert N. M. Watson]

On 3 Mar 2011, at 13:14, Alexander Leidinger wrote:

> Quoting Robert Watson <rwatson at FreeBSD.org> (from Thu, 3 Mar 2011 11:14:59 +0000 (GMT)):
> 
>> On Tue, 1 Mar 2011, Dmitry Chagin wrote:
>> 
>>> Teach kdump to decode linux syscalls names too.
>>> 
>>> Fix bug introduced in my previous commit: the kernel always dump native
>>> signal numbers, so no need to check the ABI in ktrpsig().
>> 
>> Does this mean that we're eliminating the need for the long-broken linux_kdump?
> 
> For the linux parts there is (L?)GPLed code in linux_kdump (the main reason that it is a port, I think), it is unlikely that someone sits there and will do a clean-room implementation of those linux-kernel derived things (I like to get proven wrong, anyone out there accepts the challenge? ;-) ). Dmitry provided an updated linux-kdump which contains preprocessed things (no need to have a non-default linuxulator setup). I have the corresponding PR assigned to me, but from his comments it looks like his new version depends upon his changes and does not DTRT without them (I am awaiting confirmation, and I asked about changes to let it DTRT if this is true).

On a related note, the solution for handling different OS ABIs and calls for OpenBSM was to have a file format that (a) expressed a superset of the capabilities of each platform and (b) attempted to provide common terms to describe their intersections. This has its limitations, but works fairly well in practice: you can generally have a single piece of software that makes pretty good use of BSM streams generated on Solaris, Mac OS X, or FreeBSD across those platforms and also Linux (on which OpenBSM compiles and runs, but whose kernel audit stack differs).

Robert