architecture paper draft

[Frank Ch. Eigler]

Hi -


On Fri, Feb 11, 2005 at 09:38:11AM +0000, Richard J Moore wrote:
> [...]
> Have we come to a parting of the ways? Is kprobes the right
> mechanism on which to build a DTRACE-like capability?

Well, it's the one we need to start with, since that's the one we have,
and to a large extent it works well.  When systemtap users later need
faster probing than the int3 method probides, systemtap could be
retargeted to generate modules that use this other API; or else
kprobes can be extended.

The big doubt in my mind is what to do about user-space.  We seem
to lack a reliable way of accessing memory there from the full
variety of contexts into which probes could be placed.  For
portability, we may have to impose severe translation- or
run-time restrictions on choices of probe points that talk
to user-space.  Investigating drastically different mechanisms,
such as auxiliary user-space processes talking ptrace, may be
necessary.

With luck, the systemtap level script writer may be kept unaware
of these issues, to the extent that the tool may automatically
choose amongst the various possible back-ends based on an
analysis of the script.


- FChE
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/systemtap/attachments/20050211/fdf060be/attachment.sig>