Previous work

Hi there,

I've been doing record/replay mechanism for a while on the Linux kernel. Judging by https://github.com/mozilla/rr/wiki/Kernel-API-wishlist it might be useful to take some of the code from the Scribe's kernel implementation.
While Scribe does not rely on performance counters, the ideas are similar. There is a major difference with shared memory as Scribe modifies the page table mechanism to track memory access ordering.

Here's how it works (screencast included):
http://velvetpulse.com/2012/11/27/scribe-the-deterministic-transparent-record-replay-engine/

Papers:
http://viennot.biz/sigmetrics2010_scribe.pdf
http://viennot.biz/sosp2011_racepro.pdf
http://viennot.biz/asplos2013_dora.pdf

Sources:
https://github.com/nviennot/linux-2.6-scribe
https://github.com/columbia/mreplay (mutable replay)

VM to play with Scribe:
http://velvetpulse.s3.amazonaws.com/vm/scribe-ubuntu-1204.tar.bz2

Let me know if there is some interest to salvage some of Scribe's code.

Start with the rr Kernel API wishlist and the linked Scribe implementation at github.com/nviennot/linux-2.6-scribe. Read the referenced Scribe papers and compare its shared-memory and page-table approach with rr. The issue does not define a specific change or completion criteria, so the intended outcome would need maintainer agreement.