| Commit message (Collapse) | Author | Age |
|---|
| |\ |
|
| | | |
|
| |/ |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
Table-driven reservations should always have the maximum fixed
priority. Thus default to this behavior, both for convenience and to
avoid accidental misconfiguration.
|
| |
|
|
|
|
|
|
| |
resctrl is a new tool intended to configure reservations in
reservation-aware plugins. For now, it only supports the partitioned
reservations (P-RES) plugin.
(While at it, also update .gitignore)
|
| |
|
|
|
|
|
|
|
|
| |
The -r flag allows specifying a "virtual CPU" (in addition to a
physcial CPU to migrate to) that may be interpreted by a scheduler
plugin to refer to a reservation to which the task should be added.
The -v flag causes rtspin to output timing information on stdout and
can be useful for early-stage plugin debugging (to see that something
is happening at all).
|
| |
|
|
|
|
|
| |
Switch rt_launch to use a more modern way to launch real-time tasks.
This makes rt_launch more flexible, makes it more similar to rtspin,
and also allows to let us get rid of a chunk of old, inflexible code
in liblitmus that wasn't really serving any useful purpose anymore.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch replaces the algorithm used to compute CPU affinity masks.
The old algorithm was fragile and would break on systems with multiple
levels of shared caches. (As found on some older Intel chips, and as
one would probably find on a system with hyper-threading enabled.) The
bug stems from fact that Linux enumerates consecutive CPUs in a way
that distributes them across the physical system, thereby reducing
contention resources (e.g., cache).
Cluster size no longer needs to be provided now that cluster/CPU
mappings are explicit in /proc/litmus/domains/.
This keeps the following migration functions, but marks them as
deprecated:
- be_migrate_to_cluster()
- cluster_to_first_cpu()
- partition_to_cpu()
Although the deprecated interfaces are supported, the implementations
for these functions call the new be_migrate_to_domain() and
domain_to_first_cpu() functions.
[bbb: resolved several merge conflicts]
|
| |
|
|
|
|
| |
Fixed build on ARMv7
Signed-off-by: Roy Spliet <rspliet@mpi-sws.org>
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
The sproadic_*() macros have become unwieldy. This patch
replaces those convenience macros for global, clustered,
and partitioned scheduling. A part of this API change
is the explicit use of nanosecond time-values. Prior APIs
have used lt_t (litmus time), which had an implied time
scale of nanoseconds.
/bin apps and test suite also updated to use revised API.
Modifications to the test suite are mostly centered around
using nanoseconds instead of milliseconds.
|
| |
|
|
|
|
|
|
|
| |
This patch adds cluster scheduling options to
rtspin and rt_launch. The convenience routines
in litmus.h were also updated to facilitate
clustered scheduling. For partitioned scheduling,
just set cluster_size = 1 (default size for rtspin
and rt_launch).
|
| |
|
|
|
|
|
|
|
|
|
| |
This patch replaces be_migrate_to() with several
be_migrate_*() APIs to automatically assign CPU affinity
masks from a task's partition (or cluster) assignment.
Routines are release-master-aware such that the release
master (if one exists) will NOT be included in the
task's affinity mask. (Note that release-master
avoidance may be overridden by calling
__be_migrate_thread_to_cluster().)
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This adds the uncache test tool (bin/uncache.c). The tool
can be used to test Litmus's char device driver for allocating
uncacheable CPU memory. The tool runs various checks and
gathers basic cache vs. main memory statistics.
In the future, uncache could be extended to quantify the
benefits of the L1, L2, and L3 caches, instead of treating
them as a black box.
Note: Uncache works best when compiled with '-O2'. While '-O2'
has not been added to the Makefile, other code was updated (code
in tests/ and rtspin), to compile with -O2.
DEPENDS UPON LITMUS-RT PATCH 888d097deb6d1fdc0c89a4f9667fd81cf416cfc7.
|
| |
|
|
|
|
| |
This is useful for testing locking protocol implementations (not
intended for benchmarking!), and also serves to document how to use
the LITMUS^RT locking API.
|
| |
|
|
| |
This mirrors the option in rtspin.
|
| |
|
|
|
|
|
| |
The -f option required manually *also* setting
the -w option. This is no longer required.
While at it, simplify the code by using the /proc interface.
|
| |
|
|
|
| |
Add a wrapper for /proc/litmus/stats to make it easy to query the
number of tasks already waiting for a task set release.
|
| |
|
|
| |
Also add the priority parameter, which was missing.
|
| |
|
|
|
|
|
|
| |
Updated APIs to support arbitrary deadlines. Added macros for implicit
deadlines.
Note: Had to tweak Makefile to support gcc version >= 4.6 (moved -lrt to the
end of the link command).
|
| |
|
|
|
|
| |
Use the kernel's notion of fixed priorities.
Also, add some clarifying comments to the
task setup wrappers.
|
| |
|
|
|
|
| |
PCP was only used for DPCP before
tests: add some basic tests for PCP under P-FP
|
| |
|
|
|
|
|
|
|
| |
dissertation (branch bbb-diss)
I took the unchanged code but removed references
to OMLP which was and is not implemented
tests: changed so that they work for P-FP
|
| |
|
|
|
| |
The timing functions are quite handy when building benchmark tasks.
Avoid copy&paste reuse by making them available via the library.
|
| |
|
|
|
| |
We want the positive loop length, not the negative, to prevent jobs
from overruning their budget.
|
| |
|
|
|
| |
Using cputime() is much more accurate than the old delay loop
auto-configuration. There is no good reason to keep it around.
|
| |
|
|
|
| |
Still need to figure out what to do with the period is less than the
exec_cost (or budget).
|
| |
|
|
|
| |
rtspin produces prettier schedules if it transitions to non-real-time
mode before terminating.
|
| |
|
|
| |
This hopefully helps a bit to clarify the output.
|
| |
|
|
|
|
| |
This should only trigger if the execution time tracking is broken (as
it was on District10). With the emergency exit, we at least prevent
the system from becoming unresponsive.
|
| | |
|
| |
|
|
| |
Useful to test how the kernel handles PRECISE_ENFORCEMENT.
|
| |
|
|
|
|
|
|
| |
how their execution budgets should be enforced: NO_ENFORCEMENT,
QUANTUM_ENFORCEMENT, and PRECISE_ENFORCEMENT (unsupported).
NOTE: Users of NO_ENFORCEMENT must call sleep_next_period() at the
end of every job to signal to the kernel that its job is complete.
|
| |
|
|
|
| |
Can be used to dump the delay loop that rtspin
is using.
|
| | |
|
| | |
|
| |
|
|
|
|
| |
People got confused how to use them.
They were never meant to be useful to outside developers.
So, just get rid of them now.
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
this finally appears to work reasonably on Flare
|
| |
|
|
|
| |
Don't just blindly overrun the budget.
Instead, try to measure a spin loop and hope that we get good results.
|
| |
|
|
|
| |
Run this tool (and a background workload) to measure system call
overheads.
|
| |\
| |
| |
| |
| | |
Conflicts:
Makefile
|
| | | |
|
| | | |
|
