Subject: Re: sparse core dumps?
To: Jason Thorpe <>
From: Nathan J. Williams <>
List: tech-kern
Date: 06/02/2004 18:02:21
Jason Thorpe <> writes:

> On Jun 2, 2004, at 2:25 PM, Hubert Feyrer wrote:
> > Would it be possible to allow core dumps to be "sparse"?
> ELF format core dumps can already be "sparse".  I'm pretty sure, for
> example, that we don't dump the entire stack.

The ELF format allows for them to be sparse, and sure, we special-case
the conventional process stack, but it would be good to avoid dumping
tons of data we know is empty, in the form of anon ranges without
allocated pages.  This is one of the reasons core dumps of threaded
programs are so large, as well; the thread stacks don't get this
special-case treatment.

The current uvm_coredump_walkmap() interface doesn't pass enough
information to deal with this, though; the abstraction would need to
be made a bit richer to permit finding and iterating over the
non-empty pages. The project calls for a bit of UVM guru-hood.

        - Nathan