[PATCH v2 00/15] Fair scheduling deadline server fixes

From: Joel Fernandes (Google)
Date: Tue Mar 12 2024 - 21:25:13 EST


Hello,
The deadline server [1] allows RT tasks to run on a system safely, while not
wasting CPU that RT tasks may not get on an idle system due to RT throttling.

Here are patches that are mostly fixes that we found while testing out the
deadline server [1] for ChromeOS. It snowballed from 10 to 15 patches as I
found that my unit test was breaking, and then we also saw some crashes in the
field related to the dl_timer! All of that is fixed.

There is also a fix to core scheduling among several other fixes.
Appreciate a thorough review. I kept all the patches on top of Daniel's and
Peter's patches because I will let them squash it with appropriate attribution
to the contributors.

These patches are based on Daniel's preview branch for v6:
https://git.kernel.org/pub/scm/linux/kernel/git/bristot/linux.git/?h=dl_server_v6

Daniel mentioned he is working on fixing the fair server interface issues [2].
These patches apply cleanly on his preview version.

[1] https://lore.kernel.org/all/cover.1699095159.git.bristot@xxxxxxxxxx/
[2] https://lore.kernel.org/all/091ca2ea-202d-4685-92ea-529186a94f0a@xxxxxxxxxx/


Joel Fernandes (Google) (12):
sched/core: Add clearing of ->dl_server in put_prev_task_balance()
sched/core: Fix priority checking for DL server picks
sched/core: Fix picking of tasks for core scheduling with DL server
sched/debug: Use unsigned long for cpu variable to prevent cast errors
selftests/sched: Add a test to verify that DL server works with core
scheduling
selftests/sched: Migrate cs_prctl_test to kselfttest
admin-guide/hw-vuln: Correct prctl() argument description
sched: Fix build error in "sched/rt: Remove default bandwidth control"
sched/deadline: Mark DL server as unthrottled before enqueue
sched/deadline: Make start_dl_timer callers more robust
sched/deadline: Do not restart the DL server on replenish from timer
sched/deadline: Always start a new period if CFS exceeded DL runtime

Suleiman Souhlal (2):
sched: server: Don't start hrtick for DL server tasks
sched/deadline: Reverse args to dl_time_before in replenish

Youssef Esmat (1):
sched/core: Clear prev->dl_server in CFS pick fast path

.../admin-guide/hw-vuln/core-scheduling.rst | 4 +-
include/linux/sched.h | 3 +-
kernel/sched/core.c | 46 +++-
kernel/sched/deadline.c | 87 ++++--
kernel/sched/debug.c | 4 +-
kernel/sched/fair.c | 22 +-
kernel/sched/rt.c | 2 +
kernel/sched/sched.h | 3 +-
tools/testing/selftests/sched/Makefile | 17 +-
tools/testing/selftests/sched/common.c | 24 ++
tools/testing/selftests/sched/common.h | 8 +
.../selftests/sched/cs_dlserver_test.c | 254 ++++++++++++++++++
tools/testing/selftests/sched/cs_prctl_test.c | 74 ++---
13 files changed, 456 insertions(+), 92 deletions(-)
create mode 100644 tools/testing/selftests/sched/common.c
create mode 100644 tools/testing/selftests/sched/common.h
create mode 100644 tools/testing/selftests/sched/cs_dlserver_test.c

--
2.34.1