With help from Sergey Bozhko <sbozhko@mpi-sws.org>
and Meenal Gupta <meenalg727@gmail.com>.

Pierre Roux authored 1 year ago and Björn Brandenburg committed 1 year ago

This refactoring is the result of a joint effort by Pierre Roux, Sergey Bozhko, and Björn Brandenburg. 

Fixes: #111

Pierre Roux authored 1 year ago and Björn Brandenburg committed 1 year ago

Revise some definitions of interference with too much repetition.

Sergey Bozhko authored 1 year ago and Björn Brandenburg committed 1 year ago

In the lemma I'm working on now, I need a bool version of this
definition. As we discussed, we should avoid two identical versions
(Prop + bool) whenever possible

Proofs should avoid destruct'ing the busy-window definitions all
over the place. This patch removes all instances of busy-window
`destruct`-ing from the `results` module.

Some old code sites in `analysis` remain. See also #97.

Future code additions should avoid adding new instances of
busy-window `destruct`-ing.

Remove dependency on ideal uniprocessors from
`prosa.analysis.facts.busy_interval.carry_in` as much as currently
possible, and move the remaining lemma to the `ideal.carry_in`
submodule.

See also: #112

To make room for a generic notion of `is_idle`.

Generalize:

- `preemption_time`
- priority compliance
- priority facts
- task_schedule.v

... to work with any processor model (and not just ideal
uniprocessors).

Closes: #78

This only moves definitions in a new file,
not changing anything else.

Meenal Gupta authored 2 years ago and Björn Brandenburg committed 2 years ago

Following up on a discussion [1] on the transitivity / reflexivity
/ totality of fixed-priority policies, this patch introduces the
corresponding definitions for FP and JLFP policies.

Additionally, auxiliary lemmas are provided that show that the
FP➔JLFP and JLFP➔JLDP coercions preserve transitivity / reflexivity /
totality. 

The auxiliary lemmas are made available to `rt_{e}auto`, so usually
should not show up in proof scripts. 


[1]: !265 (comment 88687)
 

Pierre Roux authored 2 years ago and Björn Brandenburg committed 2 years ago

This is replaying a part of
!244

Pierre Roux authored 2 years ago and Björn Brandenburg committed 2 years ago

There is already another_hep_job_interference_P
that get automatically combined with negPP.

Pierre Roux authored 2 years ago and Björn Brandenburg committed 2 years ago

Reading proofs about ELF, some hypotheses were neither easy to write
nor to read.

It appeared the use of implicit arguments in
model/priorities/classes.v was probably a bit too aggressive. Thus,
this patch makes some arguments explicits, for the sake of
improved clarity.

Changes:

* This commit generalises aRTA to "multi-stage" aRTA. The general idea
is explained in file [analysis/abstract/abstract_rta.v]. Short idea is
that the fixpoint equation can be extended to a sequence of fixpoint
equations. Solution to an equation can be used in later fixpoints. This
way one can support move expressive models of execution. The prior
version of aRTA (applicable to ideal uni-processors) is now an
instantiation of the new RTA theorem.

* Interference and Interfering workload are now type-classes. Note that
this changes files in directory [results], since [Variables] were
replaced with [Instance] declarations.

* IBF supports arbitrary parameters (not just relative arrival time).
IBF parametrised by the relative arrival time is not expressive enough
to support restricted-supply analysis. It was generalized to support a
larger class of parameters

* Rename [run to completion] file into [lower bound on service]. The
file was slightly generalized, now  it derives a lower bound on any
pre-defined amount of service.

* Generalize instantiations of interference and interfering workload.
Now, the definitions do not directly destruct the processor state.
Also, definitions for these functions have been made opaque.